Algorithmic Writing Assistance on Jobseekers' Resumes Increases

Algorithmic Writing Assistance on Jobseekers’

Resumes Increases Hires

Emma van Inwegen

MIT

Zanele Munyikwa

MIT

John J. Horton

MIT & NBER

March 7, 2023

Abstract

There is a strong association between the quality of the writing in a resume for new labor

market entrants and whether those entrants are ultimately hired. We show that this

relationship is, at least partially, causal: a ﬁeld experiment in an online labor market

was conducted with nearly half a million jobseekers in which a treated group received

algorithmic writing assistance. Treated jobseekers experienced an 8% increase in the

probability of getting hired. Contrary to concerns that the assistance is taking away a

valuable signal, we ﬁnd no evidence that employers were less satisﬁed. We present a

model in which better writing is not a signal of ability but helps employers ascertain

ability, which rationalizes our ﬁndings.

1 Introduction

For most employers, the ﬁrst exposure to a job candidate is typically a written resume. The

resume contains information about the applicant—education, skills, past employment, and

so on—that the employer uses to draw inferences about the applicant’s suitability for the

job. Conveying this information is the most important function of the resume. A better-

written resume—without any change in the underlying facts—might make it easier for the

employer to draw the correct inferences, which could lead to a greater chance of an interview

or job offer. We call this the “clarity view” of the role of resume writing quality. However, a

resume might not merely be a conduit for match-relevant information; the resume’s writing

itself could signal ability. In particular, the quality of the writing might be informative about

the jobseeker’s communication skills, attention to detail, or overall quality. This is another

reason better writing could lead to a greater chance of an interview or a job offer. We call

this the “signaling view” of the role of resume writing quality.

In this paper, we explore how resume writing quality affects the hiring process using

both observational data and a ﬁeld experiment. First, using observational data from a large

online labor market, we document a strong positive relationship between writing quality

and hiring (and not simply callbacks). This relationship persists even after controlling for

other factors that might otherwise explain the relationship. In terms of magnitude, one

additional error on a job seeker’s resume is associated with 1.4% fewer hires. However, this

is only an association and there are other potential reasons writing quality could be corre-

lated with hiring even with our controls. Second, we report the results of a ﬁeld experiment

in which we vary writing quality in the same market. We are primarily interested in un-

derstanding if resume writing quality has a causal effect on job market outcomes, and in

distinguishing between the “clarity view” and “signaling view.”

In order to address the question of causality, we intercept new jobseekers at the resume-

writing stage of registering for the platform and randomly offer some of them—the treat-

ment group—algorithmic writing assistance. Others—the control group—had the status

quo experience of no assistance. This writing assistance creates random variation in writ-

ing quality. The algorithmic writing assistance was a service provided by a company to

the platform and embedded in the text box where jobseekers input their resumes. We will

discuss in depth what this assistance, which we refer to as Algorithmic Writing Service pro-

vides, but generally, it makes writing better by identifying common errors and offering the

writer suggestions on how to address those errors.

In the experimental data, there is a very strong “ﬁrst stage,” in that those treated had

better-written resumes on several quantiﬁable dimensions. For example, we ﬁnd fewer

spelling and grammar errors in the resumes of the treated group of jobseekers. Positive

effects on resume quality were concentrated among the low-end of the distribution in writ-

ing quality, as jobseekers with already excellent resumes can beneﬁt little from writing

assistance.

After creating a resume, jobseekers engage in search, which may or may not lead to a

job. We observe job search behavior and outcomes for both treated and control jobseekers.

Treated workers did not send out more applications than workers in the control group,

nor did they propose higher wages. This is a convenient result because our interest is in

employers’ decision-making, even though randomization was at the level of the jobseeker.

If jobseekers had altered their application behavior—perhaps sending more applications

because they know they have a stronger case to make—we might wrongly attribute greater

job-market success to the resume rather than this endogenous effort.

Our primary outcome of interest is the effect of writing assistance on hiring. We ﬁnd that

treated jobseekers had a 8% increase in their probability of being hired at all relative to the

control group. The 95% conﬁdence interval on the percentage increase in hiring is (3%,13%.)

They also had 7.8% more job offers over the experimental period than those in the control

group. In terms of the matches themselves, treated workers’ hourly wages were 10% higher

than the hourly wages of workers in the control group. However, it is important to remember

this is a conditional result and could simply be due to composition changes in which workers

are hired.

In the “signaling view” the treatment would remove or at least weaken a credible signal

of jobseeker ability. If this is the case, this should leave employers disappointed. Unique

to our setting, we have a measure of employer disappointment, as both sides privately rate

each other at the conclusion of the contract. Although these ratings have been shown to

become inﬂated over time (Filippas et al., Forthcoming) and can be distorted when they are

public and reciprocal (Bolton, Greiner and Ockenfels, 2013), they are still a useful signal of

worker performance (Fradkin et al., 2021; Cai et al., 2014). If employers are disappointed

with the performance of the worker, this would likely manifest in lower employer ratings at

the conclusion of the contract. We ﬁnd no evidence that this is the case.

We look at the impact of the treatment to the public and private numerical ratings the

employers give to the workers, as well as the “sentiment” of the written text of reviews

(which are less prone to inﬂation(Filippas et al., Forthcoming)). We ﬁnd no signiﬁcant treat-

ment effects for any of these ratings, with both positive and negative point estimates. For

example, the average rating of the “quality of work” completed by workers in the control

group is 4.768, and 4.772 in the treatment group. The average private rating given to work-

ers in the control group was 8.63 on a ten-point scale, with average ratings in the treatment

group of 8.56. The sentiment of the review text in the treatment was slightly more positive,

but the effect was close to zero and not signiﬁcant. Given the 10% higher average wages

in the treatment group, if employers were simply tricked into hiring worse workers gener-

ally, these higher wages should have made it even more likely to ﬁnd a negative effect on

ratings (Luca and Reshef, 2021).

One possible explanation for our results is that employers are simply wrong in regarding

resume writing quality as informative about ability. However, the “clarity view” can also ra-

tionalize our results without making this assumption. It is helpful to formalize this notion

to contrast it with the more typical signaling framing of costly effort and hiring. To that end,

we present a simple model where jobseekers have heterogeneous private information about

their productivity but can reveal their type via writing a “good” resume. This is not a sig-

naling model where more productive workers face lower resume-writing costs—any worker,

by writing a good resume, will reveal their information, and this cost is assumed to be inde-

pendent of actual productivity. Our model has heterogeneous “good” resume writing costs.

We show that writing assistance shifting the cost distribution can generate our ﬁndings of

more hires, higher wages, and equally satisﬁed employers.

Our main contribution is to compare the “clarity view” and “signaling view” for the pos-

itive relationship between writing and hiring. Our main substantive ﬁnding is evidence for

the “clarity view.” We can do this because we can trace the whole matching process from re-

sume creation all the way to a measure of post-employment satisfaction. Helping jobseekers

have better-looking resumes helped them get hired (consistent with both explanations), but

we ﬁnd no evidence that employers were later disappointed (which is what the “signaling

view” explanation would predict). We also contribute more broadly by showing the impor-

tance of text in understanding matching (Marinescu and Wolthoff, 2020). The notion that

better writing can help a reader make a better purchase decision is well-supported in the

product reviews literature (Ghose and Ipeirotis, 2010) but is a relatively novel ﬁnding in la-

bor markets. In one related example, (Sajjadiani et al., 2019) analyze resumes of applicants

to public school teaching jobs and ﬁnd that spelling accuracy is associated with a higher

probability of being hired. And Hong, Peng, Burtch and Huang (2021) show that workers

who directly message prospective employers (politely) are more likely to get hired, but the

politeness effect is muted when the workers’ messages contain typographic errors.

In addition to the general theoretical interest in understanding hiring decisions, there

are practical implications to differentiating between these two views of the resume. If the

“clarity view” is more important, then any intervention that encourages better writing is

likely to be beneﬁcial. There will likely be little loss in efﬁciency if parties are better in-

formed. Even better, as we show, the kind of assistance that improves clarity can be deliv-

ered algorithmically. These interventions are of particular interest because they have zero

marginal cost (Belot et al., 2018; Briscese et al., 2022; Horton, 2017), making a positive re-

turn on investment more likely, a consideration often ignored in the literature (Card et al.,

2010). On the other hand, if the “signaling view” is more important, then providing such

writing assistance will mask important information and lead to poor hiring decisions.

Unlike general advice, algorithmic interventions are adaptive. In our study, the algo-

rithm took what the jobseeker was trying to write as input and gave targeted, speciﬁc advice

on improvement. This is likely more immediately useful than more vague recommendations,

such as telling jobseekers to “omit needless words.” This advice comes in the form of rec-

ommendations that are predicted to improve the resume’s effectiveness. A limitation of our

study is that we cannot speak to crowd-out effects (Crépon et al., 2013), which are relevant

to discuss the welfare implications of any labor market intervention. However, this concern

is somewhat secondary to our narrower purpose of understanding how employers make deci-

sions with respect to resumes. Additionally, given that in our setting, new entrants compete

with established jobseekers on the platform, we anticipate the crowd-out effect will be small,

and perhaps even welcome if at the expense of more established workers, given the obstacles

new entrants face (Pallais, 2013).

In addition to exploring an AI technology in a real labor market, we contribute to a large

literature on how experimentally varying applicant attributes affects callback rates (Moss-

Racusin et al., 2012; Bertrand and Mullainathan, 2003; Kang et al., 2016; Farber et al.,

2016). While we are not the ﬁrst to show that writing matters in receiving callbacks from

employers (Sterkens et al., 2021; Martin-Lacroux and Lacroux, 2017), we are the ﬁrst to do

so on such a massive scale and with natural variation in writing quality

. Our experiment

involves 480,950 jobseekers which is an order of magnitude larger than the next largest

experiments. Another beneﬁt is that we do not need to guess how workers might make

mistakes on their resumes, as it is workers and not researchers writing their resumes. Ad-

ditionally, unique in this literature, we can follow the induced changes all the way through

hiring and even post-employment assessment which allows us to answer our “clarity view”

vs. “signaling view” questions.

The rest of the paper proceeds as follows. Section 2 describes the online labor market

which serves as the focal market for this experiment. Section 3 reports the experimental re-

sults of the treatment effects on writing quality and subsequent labor market outcomes. In

Section 4 we present a simple model that can rationalize our ﬁndings. Section 5 concludes.

2 Empirical context and experimental design

The setting for this experiment is a large online labor market. Although these markets are

online, with a global audience, and with lower search costs (Goldfarb and Tucker, 2019), they

are broadly similar to more conventional markets (Agrawal et al., 2015). Employers post job

descriptions, jobseekers apply, and there are interviews followed by hiring and managing.

One distinctive feature of online labor markets is that both the employer and the worker

provide ratings for each other at the end of a contract.

Because of the many similarities between on and ofﬂine labor markets, a growing body

of research uses online labor markets as a setting, often through randomized experiments.

These studies contribute to the theory in longstanding questions about labor markets, such

as deepening our understanding of the mechanisms and processes by which employers and

workers ﬁnd jobs. Online labor markets also allow researchers to broaden the range of

questions in which it is possible to make causal estimates (Horton, 2010; Barach and Horton,

2021) because platforms store detailed data on things like applications, text, length of time

spent working on an application, speed of hire, and much more.

While the reason this preference exists is not known, recruiters report, anecdotally, caring about a re-

sume’s writing quality (Oreopoulos, 2011).

Many studies on online labor markets identify and measure phenomena that are rele-

vant to labor markets both online and ofﬂine. Like the ofﬂine labor market, online labor

markets have been shown to have hiring biases (Chan and Wang, 2018). But, Agrawal et al.

(2016) shows that these biases tend to be ameliorated with experience and that general, em-

ployers are able to learn as they hire (Kokkodis and Ransbotham, 2022). And Stanton and

Thomas (2016) shows that in an online labor market, agencies (which act as quasi-ﬁrms)

help workers ﬁnd jobs and break into the marketplace.

2.1 Search and matching on the platform

A would-be employer writes job descriptions, labels the job opening with a category (e.g.,

“Graphic Design”), lists required skills, and then posts the job opening to the platform web-

site. Jobseekers generally learn about job openings via electronic searches. They submit

applications, including a wage bid and a cover letter. In addition to jobseeker-initiated ap-

plications, employers can also use the interface to search worker proﬁles and invite workers

to apply to particular jobs. The platform uses the jobseeker’s history and ratings on the

platform to recommend jobseekers to would-be employers (Horton, 2017). Despite platforms

making algorithmic recommendations, none are based on the writing quality of their re-

sume. In terms of selection, Pallais (2013) shows that employers in an online labor market

care about workers’ reputation and platform experience when hiring. After jobseekers sub-

mit applications, employers screen the applicants, decide whether to give interviews, and

then whether to make an offer(s).

2.2 Experimental intervention at the resume-writing stage of pro-

ﬁle creation

When new jobseekers sign up to work on the platform, their ﬁrst step is to register and

create their proﬁle. This proﬁle serves as the resume with which they apply for jobs. This

proﬁle includes a list of skills, education, and work experience outside of the platform, as

well as a classiﬁcation of their primary job category (e.g., “Graphic Design”), mirroring what

employers select when posting a job. The interface consists of a text box for a proﬁle title

and a longer one for a proﬁle description. Jobseekers either enter their proﬁle information

on the spot or they can copy and paste it from somewhere else.

During the experimental period, jobseekers registering for the platform were randomly

assigned to an experimental cell. The experimental sample comprises jobseekers who joined

the platform between June 8th and July 14th, 2021. For treated jobseekers, the text boxes

for the proﬁle description are checked by the Algorithmic Writing Service. Control jobseek-

Figure 1: Example of the Algorithmic Writing Service ’s interface showing suggestions on

how to improve writing

Notes: Example of the Algorithmic Writing Service applied to a paragraph of text. To receive the suggestions,

users hover their mouse over the underlined word or phrase. For example, if you hover over the ﬁrst clause

“Rooms that are tiny" underlined in blue, “Tiny rooms" will pop up as a suggestion.

ers received the status quo experience. The experiment included 480,950 jobseekers, with

50% allocated to the treated cell. Table 1 shows that it was well-balanced and the balance

of pre-treatment covariates was consistent with a random process.

2.3 The algorithmic writing assistance

Words and phrases which are spelled wrong or used incorrectly are underlined by the Algo-

rithmic Writing Service. See Figure 1 for an example of the interface with an example of

the text “marked up” by the Algorithmic Writing Service. By hovering a mouse cursor over

the underlined word or phrase, the user sees suggestions for ﬁxing spelling and grammar

errors. The Algorithmic Writing Service also gives advice about punctuation, word usage,

phrase over-use, and other attributes related to clarity, engagement, tone, and style.

2.4 Platform proﬁle approval

When jobseekers ﬁnish setting up their proﬁles, they have to wait to be approved by the

platform. The platform approves jobseekers who have ﬁlled out all the necessary informa-

tion and uploaded an ID and bank details. The platform can also reject jobseekers at their

discretion. However, platform rejection is somewhat rare. About 10 percent of proﬁles are

rejected, usually as a part of fraud detection or because the jobseekers leave a completely

empty proﬁle. 46% of workers who were allocated into the experiment upon registration

complete and submit their proﬁles. About 41% of workers who begin registering get all the

way through the approval process.

As approval is made following proﬁle creation, this platform step creates a potential

problem for interpreting any intervention that changes proﬁle creation. For example, it

could be that better writing just led to a greater probability of platform approval. Or, it

could have caused jobseekers to be more likely to complete the registration process and

submit their proﬁle, both of which could effect hiring. While unlikely, this is possible, and

we do several things to deal with this potential issue.

First, see whether there is any evidence of selection. We ﬁnd no evidence that treated

jobseekers were more likely to be approved. We show that treated jobseekers are no more

likely to submit their proﬁles and that approval too is unaffected by the treatment

Second, in our main analysis, we condition on proﬁle approval in our regressions. We also

do robustness checks where we report the same analysis not conditioned on proﬁle approval

and where we control for proﬁle approval as a covariate. All our results are robust to these

strategies and are described in Section 3.11.

Once a jobseeker is approved, they can begin applying for jobs posted on the platform.

Their proﬁle will include their resume and a “proﬁle hourly wage" which is the wage offer to

employers searching for workers. After they complete their ﬁrst job on the platform, their

proﬁle also shows the worker’s actual wages and hours worked on jobs found through the

platform.

2.5 Description of data used in the analysis

The dataset we use in the analysis consists of the text of jobseekers’ resumes as well as all of

their behavior on the platform between the time they registered and August 14th, 2021, one

month after allocations ended. We construct jobseeker level data including the title and text

of their proﬁle, the number of applications they send in their ﬁrst month on the platform, the

number of invitations to apply for jobs they receive, the number of interviews they give, and

the number of contracts they form with employers. The most common categories worker’s

list as their primary job categories are Design & Creative, Writing, Administrative Support,

and Software Development, in order of frequency.

In Table 1 we present summary statistics about the jobseekers in the full experimental

See Appendix Table 12.

Table 1: Comparison of jobseeker covariates, by treatment assignment

Treatment

mean:

TRT

Control

mean:

CTL

Difference in means:

TRT

−

CTL

p-value

Full sample description: N = 480,948

Resume submitted 0.456 (0.001) 0.455 (0.001) 0.001 (0.001) 0.452

Platform approved 0.407 (0.001) 0.406 (0.001) 0.002 (0.001) 0.186

Resume length 32.911 (0.116) 32.860 (0.117) 0.051 (0.165) 0.755

Proﬁle hourly rate 18.843 (0.126) 18.917 (0.126) -0.074 (0.178) 0.676

Flow from initial allocation into analysis sample

Treatment (N) Control (N) Total (N)

Total jobseekers allocated 240,232 240,718 480,950

,→ who submitted their proﬁles 109,639 109,604 219,243

,→ and were approved by the platform 97,860 97,610 195,470

,→ with non-empty resumes 97,480 97,221 194,701

Pre-allocation attributes of the analysis sample: N = 194,700

From English-speaking country 0.182 (0.001) 0.183 (0.001) -0.002 (0.002) 0.362

US-based 0.141 (0.001) 0.143 (0.001) -0.002 (0.002) 0.222

Specializing in writing 0.166 (0.001) 0.168 (0.001) -0.002 (0.002) 0.151

Specializing in software 0.115 (0.001) 0.115 (0.001) 0.000 (0.001) 0.770

Resume length 70.394 (0.222) 70.260 (0.222) 0.135 (0.314) 0.668

Notes: This table reports means and standard errors of various pre-treatment covariates for the treatment

group and the control group. The ﬁrst panel shows the post-allocation outcomes of the full experimental

sample i) proﬁle submission, ii) platform approval, iii) length of resume in the number of words, iv) proﬁle

hourly wage rate in USD. The means of proﬁle hourly rate in treatment and control groups are only for those

proﬁles which report one. The reported p-values are for two-sided t-tests of the null hypothesis of no difference

in means across groups. The second panel describes the ﬂow of the sample from the allocation to the sample

we use for our experimental analysis. The complete allocated sample is described in the ﬁrst line, with each

following line deﬁned cumulatively. The third panel looks at pre-allocation characteristics of the jobseekers in

the sample we use for our analysis, allocated jobseekers with non-empty resumes approved by the platform. We

report the fraction of jobseekers i) from the US, UK, Canada, or Australia, ii) from the US only, iii) specializing

in writing jobs, iv) specializing in software jobs, and v) the mean length of their resumes in the number of

words.

sample as well as the sample conditioned on platform approval. 17% of the jobseekers spec-

ify that writing jobs are their primary area of work. Only 14% of jobseekers are based in the

US, and over 80% are based in a country where English is not the native language.

2.6 Constructing measures of writing quality

We do not observe the changes that the Algorithmic Writing Service suggested—we simply

observe the resumes that result. As such, we need to construct our own measures of writing

quality to determine if the treatment was delivered.

Algorithmic Writing Service gives suggestions to writers about how to improve text

along several dimensions. Perhaps the most straightforward measure of writing quality

is spelling. To see if the treatment impacted spelling errors, we take each worker’s proﬁle

and check if each word appears in an English language dictionary. We use the dictionary

hunspell, which is based on MySpell dictionaries and is the basis for the spell checker for

Google Chrome, Firefox, and Thunderbird.

As many of the resumes are for technical jobs, they often contain industry-speciﬁc terms

such as “UX” or brand names like “Photoshop.” To prevent these from being labeled as er-

rors, we augmented the list of words in the dictionary by checking the 1,000 most commonly

“misspelled” words in our sample and adding non-errors manually.

Spelling is not the only measure of writing quality. To broaden our measures, we use

LanguageTool, an open-source software that ﬁnds many errors that a simple spell checker

cannot detect, to understand employers care about measures of writing quality other than

simply the number of spelling mistakes. LanguageTool is a rule-based dependency parser

that identiﬁes errors (rule violations) and categorizes them. Some example categories in-

clude “Nonstandard Phrases,” “Commonly Confused Words,” “Capitalization,” and “Typog-

raphy.” For example, the nonstandard phrase “I never have been" would be ﬂagged with a

suggestion to replace it with “I have never been.”

2.7 Spelling errors are associated with lower hiring probabilities

in the observational data

Before presenting results of the experiment, we explore the relationship between resume

writing quality and hiring in observational data from this market. We begin by studying the

most unambiguous measure of writing quality: spelling. In Figure 2 we plot the relationship

between hiring outcomes and the percentage of words spelled correctly on the resumes of

all jobseekers who registered for the platform over the month of May 7th through June 7th,

2021, prior to the experiment. Because the distribution of percent correctly spelled is so left

skewed, we truncate the sample to only those who spell at least 75% of the words in their

resumes correctly. This window includes 97% of jobseekers. The x-axis is deciles between

75% and 100% of words spelled correctly.

For a more detailed explanation of all of the rule categories, see Table 7 in Appendix A.

Figure 2: Association between spelling errors and hiring outcomes in the observational data

Number of contracts in total

Probability of being hired at least once

(0.751,0.9]

(0.9,0.932]

(0.932,0.947]

(0.947,0.957]

(0.957,0.966]

(0.966,0.972]

(0.972,0.978]

(0.978,0.983]

(0.983,0.989]

(0.989,0.996]

(0.996,1]

(0.751,0.9]

(0.9,0.932]

(0.932,0.947]

(0.947,0.957]

(0.957,0.966]

(0.966,0.972]

(0.972,0.978]

(0.978,0.983]

(0.983,0.989]

(0.989,0.996]

(0.996,1]

0.02

0.04

0.06

0.08

Percentage of Words Spelled Correctly in Profile

Notes: These data show the relationship between the percentage of correctly spelled words on a jobseekers’

resume with various hiring outcomes. A 95% conﬁdence interval is plotted around each estimate. The sample

is of all new jobseekers who registered and were approved for the platform between June 1st and June 7th,

2021, and had resumes with more than 10 words. Plots are truncated at those who spelled at least 75% of the

words in their resume correctly.

Job seekers with resumes with fewer spelling errors are more likely to be hired. In the

left facet, the y-axis is the number of contracts a jobseeker forms in their ﬁrst month on the

platform. Jobseekers with over 99% of the words in their resume spelled correctly are hired

nearly three times more in their ﬁrst month on the platform than jobseekers with less than

90% spelled correctly. In the right facet, the y-axis is the probability that a jobseeker is ever

hired in their ﬁrst month on the platform. However, as is visible in both facets, resumes

with 100% of words spelled correctly are much less likely to receive interest from employers.

This is likely because those resumes tend to be much shorter than the others—the average

length of a resume that has zero spelling errors is only 47 words long.

2.8 The association between various kinds of writing errors and

hiring probabilities

Moving beyond spelling, in Table 2, we show the correlation between hiring outcomes on

each type of language error in the resumes in the observational data.

In the ﬁrst speciﬁca-

tion we show the correlation between the error rate for the various types of language errors

In Table 8 of Appendix A we summarize the occurrence of other types of errors within the observational

data.

and the number of contracts formed over the jobseeker’s ﬁrst month on the platform. In the

second speciﬁcation the outcome is simply whether or not the jobseeker was ever hired in

their ﬁrst month. In Columns (3) and (4), we control for the jobseekers’ proﬁle hourly rate

and their job category. Resumes with more per word grammar errors, typos, typography

errors, and miscellaneous errors are all hired less. This linear model places some unreason-

able assumptions like constant marginal effects on the relationship between various writing

errors and hiring. There may be interactions between these error types. However, it is still

useful to summarize the relationships. We can see generally negative relationships between

a higher writing error rate and hiring.

Interestingly, more style errors positively predict hiring. While initially surprising, style

errors are often caused by language being unnecessarily ﬂowery. Some examples of style

errors are “Moreover, the street is almost entirely residential” and “Doing it this way is

more easy than the previous method.” This implies that despite employers’ dislike of most

writing errors, they forgive or even prefer this kind of ﬂowery language.

For robustness we repeat these analysis in levels in Appendix Table 17. And in Appendix

Table 18 we collapse all error types into one measure of Total Errors and report these results

in both levels and normalized by resume length. The negative relationship between writing

errors and hiring persists in all speciﬁcations.

3 Effects of the treatment

We look at two main kinds of experimental results. First, we examine how the treatment

affected the text of resumes. We are looking to see whether there is a “ﬁrst stage.” Next,

we look at market outcomes for those treated workers. For convenience, we present these

treatment effects as percent changes, in Figures 3 and 5.

3.1 Algorithmic writing assistance improved writing quality

The ﬁrst step is to measure the effect the Algorithmic Writing Service has on writing in the

treatment group. We start with the fraction of words in the resume spelled incorrectly. In

the control group, resumes are 70 words long on average. Even the worst spellers spell most

of the words correctly, and an average resume has 96% of the words spelled correctly.

To understand the effects of the treatment on other types of writing errors we return to

the more ﬁne-grained LanguageTool deﬁnitions of writing errors. In Figure 3, we look at

the effect of treatment on the number of each type of writing error, normalized by resume

Table 2: Hiring outcomes predicted based on language errors (normalized by word count) in

observational data

Dependent variable:

Number of Contracts Hired Number of Contracts Hired

(1) (2) (3) (4)

Capitalization Error −0.075 −0.038 −0.055 −0.026

(0.048) (0.025) (0.045) (0.023)

Possible Typo −0.030

∗∗

−0.022

∗∗∗

−0.016 −0.013

∗∗

(0.013) (0.007) (0.012) (0.006)

Grammar Error −0.534

∗∗∗

−0.314

∗∗∗

−0.360

∗∗∗

−0.210

∗∗∗

(0.097) (0.051) (0.092) (0.047)

Punctuation Error −0.0001 0.0002 −0.0002 0.0001

(0.006) (0.003) (0.006) (0.003)

Typography Error −0.098

∗∗∗

−0.069

∗∗∗

−0.066

∗∗∗

−0.050

∗∗∗

(0.026) (0.014) (0.025) (0.013)

Style Error 0.261

∗∗

0.130

∗∗

0.234

∗∗

0.115

∗∗

(0.119) (0.062) (0.112) (0.058)

Miscellaneous Error −0.414

∗∗∗

−0.220

∗∗∗

−0.252

∗

−0.121

(0.151) (0.079) (0.143) (0.074)

Redundant Phrases −0.433 −0.264 −0.240 −0.149

(0.437) (0.229) (0.414) (0.213)

Nonstandard Phrases 0.804 −0.124 0.699 −0.193

(1.681) (0.882) (1.591) (0.819)

Commonly Confused Words −0.761 −0.331 −0.531 −0.190

(0.618) (0.324) (0.584) (0.301)

Collocations −0.637 −0.380

∗

−0.438 −0.262

(0.434) (0.228) (0.411) (0.211)

Semantic Error −0.340 −0.532 −0.191 −0.445

(1.112) (0.583) (1.052) (0.541)

Constant 0.053

∗∗∗

0.036

∗∗∗

0.036

∗∗∗

0.026

∗∗∗

(0.002) (0.001) (0.002) (0.001)

Controls X X

Observations 65,114 65,114 65,114 65,114

0.001 0.002 0.106 0.140

Notes: This table analyzes correlation between various writing errors on jobseekers’ resumes and their hiring

outcomes. The independent variables, writing errors, are divded by the number of words in the jobseekers’

resume. Number of Contracts is deﬁned as the number of unique jobs they work over the month after they

Columns (3) and (4) include controls for proﬁle hourly rate and job category. Writing errors are deﬁned by

LanguageToolR. The sample is made up of all jobseekers who registered for the platform in the week before

the experiment who submitted non-empty resumes.

Signiﬁcance indicators: p ≤ 0.10 : ∗, p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.

Figure 3: Effect of the algorithmic writing assistance on writing quality measures

Commonly Confused Words

Miscellaneous Errors

Collocations

Capitalization Errors

Nonstandard Phrases

Grammar Errors

Typographic Errors

Redundant Phrases

Punctuation Errors

Possible Typo

Semantic Errors

Style Errors

All Error Types

−30%

−20%

−10%

10%

20%

Percentage (%) Difference between Treatment and Control Group

Notes: This plot shows the effect of the treatment on various writing errors in jobseekers’ resumes. Point

estimates are the percentage change in the dependent variable versus the control group for the treatment

groups. A 95% conﬁdence interval based on standard errors calculated using the delta method is plotted

around each estimate. The experimental sample is of all new jobseekers who registered and were approved

for the platform between June 8th and July 14th, 2021, and had non-empty resumes, with N = 194,701.

Regression details can be found in Tables 10 and 11 of the Appendix.

Figure 4: Effect of treatment on percentage of words spelled correctly, by deciles

OLS Estimate

−0.2

0.0

0.2

0.4

0.6

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.0

Treatment Effect in (%)

Notes: This plot shows the effect of the treatment on the percentage of words spelled correctly in jobseekers’

resumes, by deciles. The experimental sample is of all new jobseekers who registered and were approved for

the platform between June 8th and July 14th, 2021, and had non-empty resumes, with N = 194,701.

length.

Our outcomes of interest are the error rate for each type, so we normalize each type

of error to the number of words in the resume. For treatment effects measured in percentage

terms we calculate the standard errors using the delta method.

We ﬁnd that jobseekers in the control group had signiﬁcantly higher rate of errors of the

following types: capitalization, collocations, commonly confused words, grammar, spelling,

possible typos, miscellaneous, and typography. We ﬁnd larger treatment effects for errors

associated with writing clarity than for many others. For example, two of the largest mag-

nitudes of differences in error rate were commonly confused words and collocations, where

two English words are put together that are not normally found together. Interestingly,

the treatment group had more “style” errors, paralleling our results from the observational

data, Table 2.

3.2 Algorithmic assistance helped the worst writers more

The treatment was predominantly effective for jobseekers at the bottom of the spelling dis-

tribution. In Figure 4 we report results from a quantile regression on the effect of the

treatment on the percentage of words they spelled correctly. The effect is concentrated in

jobseekers in the bottom half of the spelling distribution. The treatment effect is largest for

jobseekers below the 30% decile, with effects decreasing at each decile until the median at

which point the treatment did not affect spelling.

The treatment had no effect on the length of resumes—see Table 9 in Appendix A.

Figure 5: Effect of algorithmic writing assistance on hiring outcomes

Number of applications

Mean worker wage bid

Number of invitations to apply

Number of interviews

Hired

Number of contracts

Mean hourly rate for worked jobs

−5%

10%

15%

20%

Percentage (%) Difference between Treatment and Control Group

Notes: This analysis looks at the effect of treatment on hiring outcomes on jobseekers in the experimental

sample. The x-axis is the difference in the mean outcome between jobseekers in the treated group and the

control group. A 95% conﬁdence interval based on standard errors calculated using the delta method is plotted

around each estimate. The experimental sample is of all new jobseekers who registered and were approved

for the platform between June 8th and July 14th, 2021, and had non-empty resumes, with N = 194,701.

Regression details on the number of applications and wage bid can be found in Table 4. Regression details

on invitations, interviews, hires, and the number of contracts can be found in Appendix Table 13. Regression

details on hourly wages can be found in Table 6.

3.3 Heterogeneous treatment effects to spelling

A natural question is whether effects differed by jobseeker background. In Table 3 we inter-

act pre-randomization jobseeker attributes with the treatment. We can see that jobseekers

from the US, from English-speaking countries,

and who are writers all do better in “lev-

els.” We ﬁnd that jobseekers from countries that are not native English speaking experience

signiﬁcantly larger treatment effects to the fraction of words they spell correctly than their

anglophone counterparts.

Table 3: Effects of writing assistance on error rate

Dependent variable:

Total Error Rate x 100

(1) (2) (3) (4)

Algo Writing Treatment (Trt) −0.578

∗∗∗

−0.697

∗∗∗

−0.667

∗∗∗

−0.581

∗∗∗

(0.073) (0.081) (0.079) (0.080)

Anglophone Country −4.767

∗∗∗

(0.133)

Trt ×Anglo 0.609

∗∗∗

(0.188)

US −4.683

∗∗∗

−4.388

∗∗∗

(0.147) (0.104)

Trt × US 0.561

∗∗∗

(0.208)

Writer −0.729

∗∗∗

(0.138)

Trt × Writer −0.046

(0.195)

Constant 7.978

∗∗∗

8.854

∗∗∗

8.651

∗∗∗

8.732

∗∗∗

(0.052) (0.057) (0.056) (0.058)

Observations 187,858 187,858 187,858 187,858

0.0003 0.012 0.010 0.010

Notes: In Column (1) we show the overall effect of the treatment to the number of errors on a jobseekers’

resume divided by the number of words. In Column (2) we interact the treatment with a dummy variable for

if the jobseeker is from the US, UK, Canada, or Australia. In Column (3) we interact the treatment with a

dummy for if the jobseeker is in the US. In Column (4) we interact the treatment with a dummy for if the

jobseeker lists Writing as their primary category of desired work. The experimental sample is of all new

jobseekers who registered and were approved by the platform between June 8th and July 14th, 2021 and had

non-empty resumes. Signiﬁcance indicators: p ≤ 0.10 : ∗, p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.

We deﬁne whether a jobseeker is from a native English-speaking country, by whether they login to the

platform from USA, UK, Canada, or Australia.

3.4 Treated workers did not change their job search strategy or

behavior

One potential complication in our desire to focus on employer decision-making is that the

treatment could have impacted jobseekers search behavior or intensity. Suppose treated

jobseekers changed their behavior, knowing they had higher quality resumes. In that case,

we could not interpret our treatment effect as being driven by employers’ having improved

perceptions of treated jobseekers. However, we ﬁnd no evidence that jobseekers changed

their search behavior due to the treatment. In the ﬁrst facet of Figure 5, the outcome is the

number of applications a jobseeker sends out over their ﬁrst 28 days after registering. We

ﬁnd no effect of the treatment on the number of applications sent.

In the second facet, the outcome is the mean wage bid proposed by the jobseekers on

their applications in their ﬁrst 28 days on the platform. Average wage bids in both the

treatment and control group were $24 per hour. The lack of effects on jobseekers’ behaviors

makes sense because they were unaware of the treatment.

Table 4 show the effects of the treatment on jobseekers application behavior. In Column

(1) we see whether treated jobseekers applied for more jobs than those in the control group

over the experimental period and ﬁnd they did not. In Column (2) we ﬁnd that treated

jobseekers do not apply to more hourly jobs than those in the control group. They also could

have bid for higher wages knowing they had better-looking resumes. In Column (3) we see

no evidence of this, where we narrow the sample to only applications to hourly jobs and look

at the effect of the treatment on hourly wage bids.

3.5 The treatment did not affect employer recruiting

Employers were able to seek out workers using the platform’s search feature to invite job-

seekers to apply to their job openings. In Figure 5’s third facet from the top, the outcome is

the number of invitations to apply for a job that the jobseeker receives in their ﬁrst month.

We ﬁnd the effect of the treatment on employer invitations is a precise zero. In the fourth

facet from the top, the outcome is the number of interviews a jobseeker gives over their ﬁrst

month on the platform. We ﬁnd that this is also zero. Table 13 Column (4) provides the

details of this regression.

Although it may seem surprising given the results on hires and contracts, it makes sense

given that our experimental sample consists of only new jobseekers to the platform. New

entrants almost never appear in the search results when employers search for jobseekers,

given that their rank is determined by their platform history.

In the fourth facet of Figure 5, we show no effect of the treatment on number of inter-

Table 4: Effects of writing assistance on jobseekers’ application behavior

Dependent variable:

Num Applications Num Hourly Applications Mean Hourly Wage Bid

(1) (2) (3)

Algo Writing Treatment 0.008 0.008 −0.232

(0.027) (0.017) (0.402)

Constant 2.337

∗∗∗

1.235

∗∗∗

24.230

∗∗∗

(0.019) (0.012) (0.284)

Observations 194,701 194,701 65,411

0.00000 0.00000 0.00001

Notes: This table analyzes the effect of the treatment on jobseekers’ application behavior. The experimental

sample is made up of all new jobseekers who registered and were approved by the platform between June

8th and July 14th, 2021 and had non-empty resumes. The outcome in Column (1) is the number of total

applications a jobseeker sent out between the time the experiment began and one month after it ended. The

outcome in Column (2) is the number of speciﬁcally hourly applications sent out in that same time period.

The outcome in Column (3) is the mean hourly wage bid they proposed for those hourly jobs, and the sample

narrows to only jobseeker who submitted at least one application to an hourly job.

Signiﬁcance indicators: p ≤ 0.10 : ∗, p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.

views. Interviews, while technically feasible, are very uncommon on this platform. In the

control group the average jobseeker gives 0.2 interviews over the course of their ﬁrst month

after registering. Table 13 Column (3) provides the details of this regression.

3.6 Treated jobseekers were more likely to be hired

The treatment raised jobseekers’ hiring probability and the number of contracts they formed

on the platform. In the ﬁfth facet of Figure 5, the outcome is a binary indicator for whether

or not a jobseeker is ever hired in their ﬁrst 28 days on the platform. During the experi-

ment, 3% of jobseekers in the control group worked at least one job on the platform. Treated

jobseekers see an 8% increase in their likelihood of being hired in their ﬁrst month on the

platform. In Table 5 Column (1) we report these results in levels.

Jobseekers in the treated group formed 7.8% more contracts overall. In the sixth facet

of Figure 5, the outcome is the number of contracts a jobseeker worked on over their ﬁrst

month.

3.7 Hourly wages in formed matches were higher

Treated workers had 10% higher hourly wages than workers in the control group.

Table 5: Effects of writing assistance on hiring, by sub-groups

Dependent variable:

Hired x 100

(1) (2) (3) (4)

Algo Writing Treatment (Trt) 0.247

∗∗∗

0.223

∗∗

0.242

∗∗∗

0.237

∗∗∗

(0.080) (0.088) (0.086) (0.088)

Anglophone Country 2.508

∗∗∗

(0.146)

Trt ×Anglo 0.155

(0.207)

US 2.602

∗∗∗

(0.161)

Trt × US 0.072

(0.228)

Writer −0.293

∗

(0.151)

Trt × Writer 0.060

(0.214)

Constant 3.093

∗∗∗

2.632

∗∗∗

2.719

∗∗∗

3.142

∗∗∗

(0.057) (0.063) (0.061) (0.062)

Observations 194,703 194,703 194,703 194,703

0.00005 0.003 0.003 0.0001

Notes: This table analyzes the effect of the treatment on whether or not a jobseeker was ever hired on the

platform in the month after they joined, times 100. In Column (1) we show the overall effect of the treatment

to hiring. In Column (2) we interact the treatment with a dummy variable for if the jobseeker is from the US,

UK, Canada, or Australia. In Column (3) we interact the treatment with a dummy for if the jobseeker is in the

US. In Column (4) we interact the treatment with a dummy for if the jobseeker lists Writing as their primary

category of desired work. The experimental sample is of all new jobseekers who registered and were approved

by the platform between June 8th and July 14th, 2021 and had non-empty resumes. Signiﬁcance indicators:

p ≤ 0.10 : ∗, p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.

In the seventh facet, the outcome is the mean hourly rate workers earned in jobs they

worked over their ﬁrst month on the platform.

In the control group, workers on average made $16.80 per hour. In the treatment group,

workers made $18.48 per hour, a signiﬁcant difference at the 0.042 level. Since workers did

not bid any higher, this result suggests that employers are hiring more productive workers,

or that they thought the treated workers were more productive. If it is the latter, the “sig-

naling view” would predict that employers would then be disappointed with the workers

they hired, which we should be able to observe in worker ratings.

Table 6: Effect of algorithmic writing assistance on average wages and ratings of worked

jobs

Dependent variable:

Hourly wage rate Private rating Positive text review Recieved rating Recieved text review

(1) (2) (3) (4) (5)

Algo Writing Treatment 1.685

∗∗

−0.077 −0.004 −0.002 0.006

(0.830) (0.082) (0.009) (0.012) (0.008)

Constant 16.796

∗∗∗

8.633

∗∗∗

0.874

∗∗∗

0.624

∗∗∗

0.138

∗∗∗

(0.605) (0.059) (0.006) (0.008) (0.006)

Observations 2,816 4,250 4,529 6,263 6,263

0.001 0.0002 0.00005 0.00001 0.0001

Notes: This analysis looks at the effect of treatment on outcomes of worked jobs for jobseekers in the exper-

imental sample. Column (1) deﬁnes hourly wage rate as the mean hourly wage rate paid for all hourly jobs

worked. Column (2) deﬁnes private rating as the mean private rating on all jobs given by employers to the

workers after the job ended. In Column (3) we take the text of the reviews left by employers on each job and

use sentiment analysis (model: distilbert-base-uncased-ﬁnetuned-sst-2-english) to impute whether the review

is positive or negative, labeled one or zero. The outcome is the mean of these ratings over all worked jobs in the

sample. Column (4) is the percentage of contracts worked where the freelancer recieved any private rating.

And Column (5) is the percentage of contracts worked where the freelancer recieved any text based review.

The experimental sample is of all new jobseekers who registered and were approved for the platform between

June 8th and July 14th, 2021 and had non-empty resumes. Signiﬁcance indicators: p ≤ 0.10 : ∗, p ≤ 0.05 : ∗∗

and p ≤ .01 : ∗ ∗ ∗.

Hourly wage rates for new entrants are not representative of rates on the platform. If a new entrant gets

hired for their ﬁrst job, they tend to experience rapid wage growth.

3.8 Employers satisfaction was unaffected by the treatment

At the end of every contract, employers rate and review the workers by reporting both pub-

lic and private rating to the platform. Private ratings are not shared with the worker. In

the control group, workers had an average private rating of 8.63. In Table 6 we show that

treated workers who formed any contracts over the experimental period did not have statis-

tically different private ratings than workers in the control group. In Column (2) we show

that workers in the treated group have an average private rating of 8.56 with a standard

error of 0.08.

When the employers give these ratings they are also able to leave text reviews. While

numerical ratings have become inﬂated in recent years, Filippas et al. (Forthcoming) show

that the sentiments associated with the text of reviews has increased signiﬁcantly less over

time. This means that text reviews are likely more informative about the workers’ quality

than the numerical ratings. We use a BERT text classiﬁcation model (HF Canonical Model

Maintainers, 2022) to label each review as having positive or negative sentiment. These

classiﬁcations are signiﬁcantly correlated with the private ratings, with a Pearson correla-

tion coefﬁcient of 0.54. In Column (3) of Table 6 we show that the treated workers’ average

text reviews are not statistically different from the average sentiment of the reviews for con-

trol workers. We may also worry that if employers are less happy with the workers quality

or productivity, that they may be less likely to leave a review at all. In Column’s (4) and

(5) we show that workers in the treatment group are not more or less likely to receive any

rating or text reviews than workers in the control group.

Lastly, in Appendix Table 16 we report the results of the effect of the treatment on the

employers’ public ratings of the workers. Each outcome is a public rating the employers

give to the workers at the end of a contract. Employers rate the workers communication,

skills, quality of work, availability, cooperation, and ability to make deadlines. Each rating

is given on a ﬁve point scale. There is less variation in the public ratings than in the private

ones, and the average rating for each attribute is over 4.75 stars. Like the private ratings,

there are no signiﬁcant effects of the treatment to any of the ratings, including to workers’

communication skills. And the point estimate of the treatment effect to the quality of the

work done is even positive.

3.9 How much power do we have to detect worse contractual out-

comes?

Given the null effect of the treatment to ratings, a natural question is how much power is

available to detect effects. While we do ﬁnd a substantial increase in hiring—8%—these

marginal hires are mixed in with a much larger pool of “inframarginal” hires that would

likely be hired anyway, but for our intervention. How much worse could those marginal

applicants have been and still get our results to private ratings in the treatment?

Let I indicate “inframarginal” jobseekers who would have been hired in the treatment

or control. Let M indicate “marginal” jobseekers who are only hired in the treatment. For

workers in the control group, the average private rating will be

. But for the treat-

ment, the mean rating is a mixture of the ratings for the inframarginal and the ratings for

the induced, marginal applicants, and so

+ τ

1 + τ

(1)

where τ is the treatment effect. We assume no substitution, making our estimates conser-

vative. The sampling distribution of the mean rating for the marginal group is

(1 + τ) −

(2)

Our course,

, τ and

are all themselves random variables. Furthermore, they are not

necessarily independent. To compute the sampling distribution of

, we bootstrap sample

both the hiring regressions and the private feedback regressions on the experimental sam-

ple.

Because we do not have feedback on workers who are never hired, we use the estimates

values to calculate

. Figure 6 shows the sampling distribution of

The treatment actual rating is plotted as a dotted line and the and control actual rating

is plotted as a solid vertical line. The distribution is centered at these mean values.

The dashed line indicates the control mean rating minus one standard deviation in the

private ratings (where the standard deviation is 2.4). Comparing this value to the distribu-

tion of

, this value (at the dashed line) lies at only about 0.025 of the density. In short,

it would be quite surprising for us to get the results we have—an 8% increase in hires and

slightly higher (but not signiﬁcant ratings) if the actual marginal hires were a standard

deviation worse.

3.10 Heterogeneous treatment effects to hiring

We might have expected the treatment to have differential hiring effects on the subgroups of

interest, particularly since the treatment disproportionately impacted the fraction of words

We deﬁne this sample as the workers allocated into the experiment who were approved by the platform

and had non-empty resumes. From this we bootstrap sample with replacement. We run the hiring regressions

on this sample and the ratings regressions on the same samples, narrowed to only those workers who were

ever hired.

Figure 6: Sampling distribution of the private ratings of marginal hired jobseekers

Trt

Mean

Rating

Ctl

Mean

Rating

Ctl Mean Rating − 1 Std Dev

0.00

0.25

0.50

0.75

1.00

4 6 8

Bootstrap estimate of mean private ratings for induced (marginal) hires

CDF

spelled correctly in non-native English speakers’ resumes. In hiring outcomes, we might

expect, for example, that native English or US-based jobseekers would beneﬁt less, while

writers might beneﬁt more—though as we saw earlier, writers already make few errors.

However, for these same jobseekers, the treatment might do less.

We have already shown above in Table 3 that the treatment disproportionately impacted

the fraction of words spelled correctly in non-native English speakers’ resumes. If we look

downstream to hiring outcomes, in Table 5, we interact the same groups with the treatment

and look at their effect on the probability they were hired. The point estimates are generally

quite imprecise and we lack the power to conclude much. While non-native English speak-

ers’ writing might beneﬁt more from the treatment, it does not translate into more hires

relative to native English speakers.

3.11 Robustness checks

In our main analysis we narrow the sample to only those jobseekers whose proﬁles were

approved by the platform. In Appendix Table 14 we run a similar regression on the full

experimental sample, but we include proﬁle approval as a control to see if it affects the

estimates. In this analysis, we ﬁnd that the treatment effect on the number of hires is

slightly smaller than in the analysis conditional on platform approval—conditioning the

sample on only jobseekers whose proﬁles were approved has an estimate of 7.8% while it is

10% in the full sample. The effect on the probability of any hire is 8% in the sample of only

approved jobseekers and 8% in the unconditional sample. This approach and narrowing

the sample to only approved jobseekers would “block” the approval channel. In Appendix

Table 15 we report the same analysis not conditioned on proﬁle approval. None of these

robustness checks change the direction or signiﬁcance of any of the hiring estimates, and

the slightly larger estimates in the unconditional sample are unsurprising because platform

approval is a necessary condition for a jobseeker to be hired.

4 A simple model of the “clarity view” of resume writing

In this section, we formalize a rational model of how the writing intervention could (a)

increase hiring but (b) not lead to worse matches. We formalize the argument that better

writing allowed employers to better ascertain who was a potential match with a simple

model, and show how this kind of interplay between resume quality and hiring could exist

in equilibrium.

4.1 A mass of jobseekers with heterogeneous productivity

There is a unit mass of jobseekers. If hired, their productivity is θ

. Workers are either

high-type (θ = θ

) or low-type (θ = θ

), with θ

> θ

. Workers know their own type. It is

common knowledge that the fraction of high types in the market is γ. All workers, if hired,

are paid their expected productivity, from the employer’s point of view. Hires only last one

unit of time.

4.2 Jobseekers decide whether to put into resume-writing

Before being hired, jobseekers write resumes. Jobseekers must decide whether to put effort

e ∈ {0,1} into writing that resume. Effort itself is not observable. The cost of this effort is

jobseekers-speciﬁc and there is a distribution of individual resume effort costs. The support

of the cost distribution is [0,

c]. The distribution has mass everywhere and the CDF is F

and PDF is f . Jobseekers who put in no effort have resume costs of 0, while those that put

in effort have a cost of c

. Critically, this cost is independent of a jobseeker’s type i.e., there

is no Spence-like assumption that better workers ﬁnd it cheaper to create better resumes

(Spence, 1978).

Before making an offer, ﬁrms observe a signal of jobseekers’ type on their resume, R ∈

{0,1}. With effort, a high-type jobseeker generates an R = 1 signal; without effort, R = 0. A

low-type jobseeker generates R = 0 no matter what.

Clearly, low-types will never put in effort. The question is whether a high type will put in

effort. The decision hinges on whether the cost of resume effort is worth the wage premium

it creates. Let w

R=0

be the wage paid in equilibrium to jobseekers with R = 0. Note that

R=1

= θ

, as there is no uncertainty about a jobseeker’s type if R = 1.

A jobseeker i who is a high-type will choose e = 1 if θ

−w

R=0

) > c

. The marginal high-

type jobseeker is indifferent between putting in effort or not, and has a resume-writing cost

c, where

c = θ

− w

R=0

(

c). (3)

This implies that there are F(c)γ jobseekers that choose e = 1. These are the high-type

jobseekers with relatively low resume writing costs. The remaining [1 − F(c)]γ high-type

jobseekers choose e = 0. They are pooled together with the 1 − γ jobseekers that choose e = 0

because they are low-types.

From the employer’s perspective, if they believe that the resume effort cost of the marginal

high-type jobseekers is

c, the probability an R = 0 jobseekers is high-type is

R=0

(

c) =

1 − F(

1/γ − F(

. (4)

The wage received by an R = 0 worker is

R=0

(

c) = θ

+ (θ

− θ

R=0

(

c) (5)

When the cost of the marginal jobseeker is higher, more jobseekers ﬁnd it worth choosing

e = 1, as F

(

c) > 0. This leaves fewer high-types in the R = 0 pool, and so

d p

R=0

< 0. (6)

4.3 The equilibrium fraction of high-type workers putting effort

into resume-writing

In equilibrium, there is some marginal high-type jobseeker indifferent between e = 0 and

e = 1, and so

(θ

− θ

)(1 − p

R=0

(

∗

)) =

∗

Figure 7 illustrates the equilibrium i.e., the cost where the marginal jobseeker is indif-

ferent between e = 0 and e = 1. The two downward-sloping lines are the pay-offs to the

marginal jobseeker for each

c. The pay-off to R = 1 is declining, as the wage is constant (at

) but the cost is growing linearly. The pay-off to R = 0 is also declining, from Equation 6.

Both curves are continuous.

Note that when the marginal jobseeker has

c = 0, there is just a point-mass of high-types

that have a cost that low, i.e., f (

c). Because the marginal jobseeker is indifferent between

Figure 7: Equilibrium determination of the marginal high-type jobseeker indifferent be-

tween putting effort into a resume

γ + (1 − γ)θ

−

Cost to

Marginal

Worker

∗

Resume writing

costs decrease

Payoff to marginal

H-type worker

when R = 0

Payoff to marginal

H-type worker

when R = 1

putting in effort and not putting in effort, jobseekers with costs of even ε will not put in

effort. Since no one ﬁnds it worthwhile to put in effort the R = 0 pool is just the expected

value of all jobseekers. And the wage is w

R=0

(

c) = γθ

+ (1 − γ)θ

. The marginal jobseeker

pays nothing, so the pay-off is θ

At the other extreme,

c =

c, all but a point mass of jobseekers have a cost less than

this. Since the marginal jobseeker is indifferent between putting in effort at a cost of

c, any

jobseeker with cost

c − ε or below will put in effort. Then the R = 0 pool is purely low-types

and the wage is θ

. For the R = 1 market, the marginal jobseeker has a cost of

c so the

pay-off is θ

−

c. We know θ

> γθ

+ (1 − γ)θ

. And by assumption, θ

> θ

−

c, and so by

the intermediate value theorem, an equilibrium

∗

exists on (0,

c).

4.4 A shift in the resume writing cost distribution leads to more

high-type workers choosing to exert effort

Now suppose a technology comes along that lowers—or at least keeps the same—resume

writing costs for all jobseekers. This would shift F higher for all points except the endpoints

of the support, creating a new distribution of costs that ﬁrst-order stochastically dominates

the other.

Before determining the new equilibrium, note that no matter the marginal

c, when F

increases, the probability that an R = 0 worker is a high-type declines, as

d p

= −

(F − 2)

< 0. (7)

This shifts the w

R=0

curve down everywhere, without changing the endpoints.

Because w

R=1

−

c is downward sloping, it intersects w

R=0

(

c) at a higher value of

c. At

the new equilibrium, the marginal jobseeker has resumes costs of

∗0

, where

∗0

∗

. At

this new equilibrium, more jobseekers choose e = 1, causing more R = 1 signals. This lowers

wages for the R = 0 group.

4.5 The effects of lower costs are theoretically ambiguous

Note that this shift in costs is not Pareto improving—low-types are made worse off as they

ﬁnd themselves in a pool with fewer high-types. Furthermore, because workers are all paid

their expected product, the surplus maximizing outcome would be for everyone to choose

R = 0. Resume effort purely changes around the allocation of the wage bill, not the total

amount. Total surplus is

γ + (1 − γ)θ

−

c f (c)dc, (8)

which is maximized at

c = 0, i.e., when no one ﬁnds it worthwhile to choose effort. However,

with a shift in cost distribution (raising F), what matters is whether the marginal decrease

in costs for all inframarginal workers i..e, those with c <

c outweighs the costs borne by the

(newly) marginal jobseekers who choose to put in effort.

In our model, all job offers are accepted. However, if we think of jobseekers as having

idiosyncratic reservation values that determine whether they accept an offer, the shift in

costs makes it more likely that high-types will accept an offer, while making it less likely

that low-types will accept an offer. This is consistent with results where there is a greater

chance an employer hires at all in the treatment. It is also consistent with our result of

higher wages. Finally, if we think of employer ratings being a function of surplus, our ﬁnding

of no change in employer satisfaction is also consistent, as employers are, in all cases, just

paying for expected productivity.

5 Conclusion

Employers are more likely to hire new labor market entrants with better-written resumes.

We argue that better writing makes it easier for employers to decide to hire a particular

worker. We show results from a ﬁeld experiment in an online labor market where treated

workers were given algorithmic writing assistance. These jobseekers were 8% more likely

to get hired and formed 7.8% more contracts over the month-long experiment. While one

might have expected writing quality to be a valuable indicator of worker quality, the treat-

ment did not affect employers’ ratings of hired workers. We provide a model of the hiring

process where the cost of exerting effort on a resume is lowered by the algorithmic writing

assistance, which helps employers to distinguish between high and low-type workers.

One possibility is that the beneﬁts to treated workers came at the expense of other work-

ers, as both treated- and control-assigned workers compete in the same market. Crowd-out

concerns have been shown to be important with labor market assistance (Crépon et al.,

2013). However, even if additional hires came from experienced workers, this is likely still

a positive result. New labor market entrants are uniquely disadvantaged (Pallais, 2013) in

online labor markets. To the extent that the gains to new workers come partially at the

expense of experienced workers, this is likely a good trade-off.

Conceptualizing AI/ML innovation and proliferation as a fall in the cost of prediction

technology ﬁts our setting (Agrawal et al., 2018b,a). Writing a resume is, in part, an applied

prediction task—what combination of words and phrases, arranged in what order, are likely

to maximize my pay-off from a job search? The Algorithmic Writing Service reduces the

effort or cost required for making these decisions. When revising their resumes, rather than

identifying errors in their own predictions themselves, jobseekers with access to Algorithmic

Writing Service specify their target audience and writing goals are given suggestions for

error correction and cleaned up writing. Furthermore, the treatment, by lowering the costs

of error-free writing for at least some jobseekers, causes them to do better at writing their

resumes.

These kinds of algorithmic writing assistance will likely “ruin” writing as a signal of

ability. With the proliferation of writing technologies with capabilities far beyond what

is explored here (Brown et al., 2020), even if the “signaling view” was at one time true,

technological changes are likely to make it not true in the future.

References

Agrawal, Ajay, John Horton, Nicola Lacetera, and Elizabeth Lyons, “Digitization

and the contract labor market,” Economic analysis of the digital economy, 2015, 219.

, Joshua Gans, and Avi Goldfarb, “Prediction, judgment, and complexity: a theory of

decision-making and artiﬁcial intelligence,” in “The economics of artiﬁcial intelligence:

An agenda,” University of Chicago Press, 2018, pp. 89–110.

, , and , Prediction machines: the simple economics of artiﬁcial intelligence, Harvard

Business Press, 2018.

, Nicola Lacetera, and Elizabeth Lyons, “Does standardized information in online

markets disproportionately beneﬁt job applicants from less developed countries?,” Jour-

nal of international Economics, 2016, 103, 1–12.

Barach, Moshe A and John J Horton, “How do employers use compensation history?

Evidence from a ﬁeld experiment,” Journal of Labor Economics, 2021, 39 (1), 193–218.

Belot, Michèle, Philipp Kircher, and Paul Muller, “Providing Advice to Jobseekers at

Low Cost: An Experimental Study on Online Advice,” The Review of Economic Studies,

10 2018, 86 (4), 1411–1447.

Bertrand, Marianne and Sendhil Mullainathan, “Are Emily and Greg More Em-

ployable than Lakisha and Jamal? A Field Experiment on Labor Market Discrimina-

tion.(2003),” Amer. Econ. Rev., 2003, 94, 991.

Bolton, Gary, Ben Greiner, and Axel Ockenfels, “Engineering trust: reciprocity in the

production of reputation information,” Management science, 2013, 59 (2), 265–285.

Briscese, Guglielmo, Giulio Zanella, and Veronica Quinn, “Providing Government

Assistance Online: A Field Experiment with the Unemployed,” Journal of Policy Analysis

and Management, 2022, 41 (2), 579–602.

Brown, Tom, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan,

Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda

Askell et al., “Language models are few-shot learners,” Advances in neural information

processing systems, 2020, 33, 1877–1901.

Cai, Hongbin, Ginger Zhe Jin, Chong Liu, and Li an Zhou, “Seller reputation: From

word-of-mouth to centralized feedback,” International Journal of Industrial Organization,

2014, 34, 51–65.

Card, David, Jochen Kluve, and Andrea Weber, “Active labour market policy evalua-

tions: A meta-analysis,” The Economic Journal, 2010, 120 (548), F452–F477.

Chan, Jason and Jing Wang, “Hiring preferences in online labor markets: Evidence of a

female hiring bias,” Management Science, 2018, 64 (7), 2973–2994.

Crépon, Bruno, Esther Duﬂo, Marc Gurgand, Roland Rathelot, and Philippe

Zamora, “Do labor market policies have displacement effects? Evidence from a clustered

randomized experiment,” The Quarterly Journal of Economics, 2013, 128 (2), 531–580.

Farber, Henry S, Dan Silverman, and Till Von Wachter, “Determinants of callbacks to

job applications: An audit study,” American Economic Review, 2016, 106 (5), 314–18.

Filippas, Apostolos, John Joseph Horton, and Joseph Golden, “Reputation Inﬂation,”

Marketing Science, Forthcoming.

Fradkin, Andrey, Elena Grewal, and David Holtz, “Reciprocity and unveiling in two-

sided reputation systems: Evidence from an experiment on Airbnb,” Marketing Science,

2021, 40 (6), 1013–1029.

Ghose, Anindya and Panagiotis G Ipeirotis, “Estimating the helpfulness and economic

impact of product reviews: Mining text and reviewer characteristics,” IEEE transactions

on knowledge and data engineering, 2010, 23 (10), 1498–1512.

Goldfarb, Avi and Catherine Tucker, “Digital economics,” Journal of Economic Litera-

ture, 2019, 57 (1), 3–43.

HF Canonical Model Maintainers, “distilbert-base-uncased-ﬁnetuned-sst-2-english (Re-

vision bfdd146),” 2022.

Hong, Yili, Jing Peng, Gordon Burtch, and Ni Huang, “Just DM Me (Politely): Di-

rect Messaging, Politeness, and Hiring Outcomes in Online Labor Markets,” Information

Systems Research, 2021, 32 (3), 786–800.

Horton, John J., “Online labor markets,” Internet and Network Economics: 6th Interna-

tional Workshop, WINE 2010, Stanford, CA, USA, December 13-17, 2010. Proceedings,

2010.

, “The Effects of Algorithmic Labor Market Recommendations: Evidence from a Field

Experiment,” Journal of Labor Economics, 2017, 35 (2), 345–385.

Kang, Sonia K, Katherine A DeCelles, András Tilcsik, and Sora Jun, “Whitened

résumés: Race and self-presentation in the labor market,” Administrative Science Quar-

terly, 2016, 61 (3), 469–502.

Kokkodis, Marios and Sam Ransbotham, “Learning to successfully hire in online labor

markets,” Management Science, 2022.

Luca, Michael and Oren Reshef, “The effect of price on ﬁrm reputation,” Management

Science, 2021, 67 (7), 4408–4419.

Marinescu, Ioana and Ronald Wolthoff, “Opening the black box of the matching func-

tion: The power of words,” Journal of Labor Economics, 2020, 38 (2), 535–568.

Martin-Lacroux, Christelle and Alain Lacroux, “Do Employers Forgive Applicants’ Bad

Spelling in Résumés?,” Business and Professional Communication Quarterly, sep 2017, 80

(3), 321–335.

Moss-Racusin, Corinne A, John F Dovidio, Victoria L Brescoll, Mark J Graham,

and Jo Handelsman, “Science faculty’s subtle gender biases favor male students,” Pro-

ceedings of the national academy of sciences, 2012, 109 (41), 16474–16479.

Oreopoulos, Philip, “Why do skilled immigrants struggle in the labor market? A ﬁeld ex-

periment with thirteen thousand resumes,” American Economic Journal: Economic Pol-

icy, 2011, 3 (4), 148–71.

Pallais, Amanda, “Inefﬁcient Hiring in Entry-level Labor Markets,” American Economic

Review, 2013.

Sajjadiani, Sima, Aaron J Sojourner, John D Kammeyer-Mueller, and Elton Myk-

erezi, “Using machine learning to translate applicant work history into predictors of per-

formance and turnover.,” Journal of Applied Psychology, 2019, 104 (10), 1207.

Spence, Michael, “Job market signaling,” in “Uncertainty in economics,” Elsevier, 1978,

pp. 281–306.

Stanton, Christopher T and Catherine Thomas, “Landing the ﬁrst job: The value of

intermediaries in online hiring,” The Review of Economic Studies, 2016, 83 (2), 810–854.

Sterkens, Philippe, Ralf Caers, Marijke De Couck, Michael Geamanu, Victor Van

Driessche, and Stijn Baert, “Costly Mistakes: Why and When Spelling Errors in Re-

sumes Jeopardise Interview Chances,” Working paper, 2021.

A Appendix

Table 7: Description of Error Rule Categories with Examples

Category Description Examples

American En-

glish Phrases

Sentence favors the American English

spelling of words.

apologize, catalog, civilization, defense

British English,

Oxford Spelling

Sentence favors the British English

spelling of words.

apologise, catalogue, civilisation, defence

Capitalization Rules about detecting uppercase words

where lowercase is required and vice

versa.

This house is old. it was built in 1950.

I really like Harry potter.

Collocations A collocation is made up of two or more

words that are commonly used together in

English. This refers to an error in this

type of phrase.

Undoubtedly, this is the result of an extremely dy-

namic development of Lublin in the recent years.

I will take it in to account.

It’s batter to be save then sorry.

Commonly Con-

fused Words

Words that are easily confused, like ’there’

and ’their’ in English.

I have my won bed.

Their elicit behavior got the students kicked out of

school.It’s the worse possible outcome.

Grammar Violations related to system of rules that

allow us to structure sentences.

Tom make his life worse.

A study like this one rely on historical and present

data.This is best way of dealing with errors.

Miscellaneous Miscellaneous rules that don’t ﬁt else-

where.

This is best way of dealing with errors.

The train arrived a hour ago.

It’s nice, but it doesn’t work. (inconsistent apostro-

phes)

Nonstandard

Phrases

I never have been to London.

List the names in an alphabetical order.

Why would a man all of the sudden send ﬂowers?

Possible Typo Spelling issues. It’a extremely helpful when it comes to homework.

We haven’t earned anything.This is not a HIPPA

violation.

Punctuation Error in the marks, such as period,

comma, and parentheses, used in writing

to separate sentences and their elements

and to clarify meaning.

"I’m over here, she said.

Huh I thought it was done already.

The U.S.A is one of the largest countries.

Redundant

Phrases

Redundant phrases contain words that

say the same thing twice. When one of the

words is removed, the sentence still makes

sense. Sometimes the sentence has to be

slightly restructured, but the message re-

mains the same.

We have more than 100+ customers.

He did it in a terrible way.

The money is sufﬁcient enough to buy the sweater.

Semantics Logic, content, and consistency problems. It allows us to both grow, focus, and ﬂourish.

On October 7, 2025 , we visited the client.This was

my 11nd try.

Style General style issues not covered by other

categories, like overly verbose wording.

Moreover, the street is almost entirely residential.

Moreover, it was named after a poet.

Doing it this way is more easy than the previous

method.

I’m not very experienced too.

Anyways, I don’t like it.

Typography Problems like incorrectly used dash or

quote characters.

This is a sentence with two consecutive spaces.

I have 3dogs.The price rose by $12,50.

I’ll buy a new T—shirt.

Table 8: Summary statistics on error counts and rates in the control group

Total Errors Error Rate

All Error Types 4.390 (10.257) 0.080 (0.252)

Capitalization Errors 0.140 (0.543) 0.004 (0.029)

Possible Typo 2.312 (8.768) 0.041 (0.106)

Grammar Errors 0.219 (0.589) 0.004 (0.014)

Punctuation Errors 0.425 (1.629) 0.008 (0.213)

Typographic Errors 0.821 (2.985) 0.016 (0.051)

Style Errors 0.317 (0.890) 0.004 (0.011)

Miscellaneous Errors 0.103 (0.391) 0.002 (0.009)

Redundant Phrases 0.025 (0.162) 0.000 (0.003)

Nonstandard Phrases 0.002 (0.050) 0.000 (0.001)

Commonly Confused Words 0.010 (0.117) 0.000 (0.002)

Collocations 0.012 (0.121) 0.000 (0.003)

Semantic Errors 0.003 (0.061) 0.000 (0.001)

Notes: This table reports means and standard errors of the writing errors in the resumes of the control group.

The ﬁrst column displays the average total error count and the second column displays the average error rate

(total errors normalized by word count). Writing errors are deﬁned by LanguageToolR. The sample is made up

of all jobseekers in the control group of the experimental sample who submitted non-empty resumes and were

approved by the platform.

Table 9: Effects of writing assistance on length of resume

Dependent variable:

Number of words in resume

Algo Writing Treatment 0.128

(0.314)

Constant 70.541

∗∗∗

(0.223)

Observations 194,701

0.00000

Notes: This table analyzes the effect of the treatment on the number of words

in a jobseeker’s resume. The sample is made up of all jobseekers in the exper-

imental sample who submitted non-empty proﬁles and were approved by the

platform. Signiﬁcance indicators: p ≤ 0.10 : ∗, p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.

Table 10: Effect of Treatment to Writing Errors, Page 1

Dependent variable:

Spelling Capitalization Possible Typo Grammar Punctuation Typography Style

(1) (2) (3) (4) (5) (6) (7)

Algo Writing Treatment −0.001

∗

−0.0005

∗∗∗

−0.002

∗∗∗

−0.0005

∗∗∗

−0.0004 −0.002

∗∗∗

0.0003

∗∗∗

(0.0005) (0.0001) (0.0005) (0.0001) (0.0004) (0.0003) (0.0001)

Constant 0.036

∗∗∗

0.003

∗∗∗

0.041

∗∗∗

0.004

∗∗∗

0.010

∗∗∗

0.015

∗∗∗

0.004

∗∗∗

(0.0003) (0.00005) (0.0003) (0.00004) (0.0003) (0.0002) (0.00004)

Observations 187,858 187,858 187,858 187,858 187,858 187,858 187,858

0.00002 0.0003 0.0001 0.0004 0.00000 0.0004 0.0001

Notes: This table analyzes the effect of the treatment on all types of writing errors, normalized by resume

length. Writing errors are deﬁned by LanguageToolR, and divided by the number of words in a jobseekers’

resume to calculate their error rate. The sample is made up of all jobseekers in the experimental sample who

completed the platform registration page and submitted non-empty resume. Signiﬁcance indicators: p ≤ 0.10 :

∗, p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.

Table 11: Effect of Treatment to Writing Errors, Page 2

Dependent variable:

Miscellaneous Redundant Phrases Nonstandard Phrases Commonly Confused Words Collocations Semantics

(1) (2) (3) (4) (5) (6)

Algo Writing Treatment −0.0004

∗∗∗

−0.00003

∗∗

−0.00001 −0.00005

∗∗∗

−0.0001

∗∗∗

0.00001

(0.00003) (0.00001) (0.00000) (0.00001) (0.00001) (0.00001)

Constant 0.002

∗∗∗

0.0004

∗∗∗

0.00003

∗∗∗

0.0001

∗∗∗

0.0003

∗∗∗

0.0001

∗∗∗

(0.00002) (0.00001) (0.00000) (0.00001) (0.00001) (0.00000)

Observations 187,858 187,858 187,858 187,858 187,858 187,858

0.001 0.00002 0.00001 0.0002 0.0001 0.00000

Notes: This table analyzes the effect of the treatment on all types of writing errors, normalized by resume

length. Writing errors are deﬁned by LanguageToolR, and divided by the number of words in a jobseekers’

resume to calculate their error rate. The sample is made up of all jobseekers in the experimental sample

who completed the platform registration page and submitted non-empty resume. Signiﬁcance: p ≤ 0.10 : †,

p ≤ 0.05 : ∗, p ≤ 0.01 : ∗∗ and p ≤ .001 : ∗ ∗ ∗.

Table 12: Effects of writing assistance on proﬁle submission and platform approval

Dependent variable:

Proﬁle submitted x 100 Approved x 100

(1) (2) (3)

Algo Writing Treatment 0.107 0.200 0.186

(0.144) (0.133) (0.142)

Constant 45.532

∗∗∗

89.057

∗∗∗

40.550

∗∗∗

(0.102) (0.094) (0.100)

Observations 480,950 219,243 480,950

0.00000 0.00001 0.00000

Notes: This table analyzes the effect of the treatment on whether or not a jobseeker’s proﬁle was submitted

and approved. In Column (1) the outcome is 100 times a binary indicator for whether or not the jobseeker

completed platform registration and submitted their resume, on the full experimental sample. In Column

(2) the outcome is 100 times a binary indicator for whether or not the platform approved the resume, on the

sample of only those jobseekers who submitted their resumes. In Column (3) the outcome is 100 times a binary

indicator for whether or not the platform approved the resume, on the full experimental sample. Signiﬁcance

indicators: p ≤ 0.10 : ∗, p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.

Figure 8: Daily allocations of jobseekers into experimental cells

4000

5000

6000

7000

8000

Jun 15

Jul 01

Jul 15

Daily Allocations into Experiment

Control

Treatment

Notes: This plot shows the daily allocations into the treatment and control cells for the experimental sample

of 480,950 new jobseekers to the platform.

Table 13: Effect of algorithmic writing assistance on hiring outcomes

Dependent variable:

Num Contracts Hired x 100 Num Hourly Interviews Num Invitations

(1) (2) (3) (4)

Algo Writing Treatment 0.004

∗∗

0.247

∗∗∗

0.002 0.001

(0.002) (0.080) (0.004) (0.003)

Constant 0.047

∗∗∗

3.093

∗∗∗

0.210

∗∗∗

0.142

∗∗∗

(0.001) (0.057) (0.003) (0.002)

Observations 194,703 194,703 194,703 194,703

0.00003 0.00005 0.00000 0.00000

Notes: This analysis looks at the effect of treatment on hiring outcomes on jobseekers in the experimental

sample. Column (1) deﬁnes Number of Contracts as the number of unique jobs they work over the month

after they register for the platform. Column (2) deﬁnes Hired x 100 as one hundred times the probability the

jobseeker was hired over that month. Column (3) is the number of interviews they gave over that month. And

the Column (4) outcome Invitations is the number of times they were recruited to a job over their ﬁrst month.

The experimental sample is of all new jobseekers who registered and were approved for the platform between

June 8th and July 14th, 2021 and had non-empty resumes. Signiﬁcance indicators: p ≤ 0.10 : ∗, p ≤ 0.05 : ∗∗

and p ≤ .01 : ∗ ∗ ∗.

Table 14: Effect of algorithmic writing assistance on hiring outcomes, controlling for plat-

form approval

Dependent variable:

Num Contracts Hired x 100 Num Hourly Interviews Num Invitations

(1) (2) (3) (4)

Algo Writing Treatment 0.001

∗∗

0.100

∗∗∗

0.001 0.0004

(0.001) (0.032) (0.002) (0.001)

Approved by Platform 0.049

∗∗∗

3.204

∗∗∗

0.210

∗∗∗

0.142

∗∗∗

(0.001) (0.033) (0.002) (0.001)

Constant −0.001 −0.050

∗

−0.0003 0.00001

(0.001) (0.027) (0.001) (0.001)

Observations 480,952 480,952 480,952 480,952

0.011 0.019 0.030 0.023

Notes: This analysis looks at the effect of treatment on hiring outcomes on jobseekers in the experimental

sample. Column (1) deﬁnes Number of Contracts as the number of unique jobs they work over the month

after they register for the platform. Column (2) deﬁnes Hired x 100 as one hundred times the probability the

jobseeker was hired over that month. Column (3) is the number of interviews they gave over that month. And

the Column (4) outcome Invitations is the number of times they were recruited to a job over their ﬁrst month.

The sample used in this analysis is the entire experimental sample. Signiﬁcance indicators: p ≤ 0.10 : ∗,

p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.

Table 15: Effect of algorithmic writing assistance on hiring outcomes, unconditional on plat-

form approval

Dependent variable:

Num Contracts Hired x 100 Num Hourly Interviews Num Invitations

(1) (2) (3) (4)

Algo Writing Treatment 0.002

∗∗

0.106

∗∗∗

0.001 0.001

(0.001) (0.033) (0.002) (0.001)

Constant 0.019

∗∗∗

1.249

∗∗∗

0.085

∗∗∗

0.058

∗∗∗

(0.0005) (0.023) (0.001) (0.001)

Observations 480,952 480,952 480,952 480,952

0.00001 0.00002 0.00000 0.00000

Notes: This analysis looks at the effect of treatment on hiring outcomes on jobseekers in the experimental

sample. Column (1) deﬁnes Number of Contracts as the number of unique jobs they work over the month

after they register for the platform. Column (2) deﬁnes Hired x 100 as one hundred times the probability the

jobseeker was hired over that month. Column (3) is the number of interviews they gave over that month. And

the Column (4) outcome Invitations is the number of times they were recruited to a job over their ﬁrst month.

The sample used in this analysis is the entire experimental sample. Signiﬁcance indicators: p ≤ 0.10 : ∗,

p ≤ 0.05 : ∗∗ and p ≤ .01 : ∗ ∗ ∗.