Short answer: some do, some don’t, and we could find no good measurement of how many. Whether a cover letter changes anything is a different question, and that one has evidence, though less of it than the internet implies. In one experiment, a letter written for the specific job moved callbacks. A generic one made much less difference, if any. Then AI made tailoring cheap on one gig-work platform, and there the advantage faded.
The closest thing to a direct answer sits in a small survey published by a company that sells cover letter writing: 87% of 236 surveyed hiring professionals said they read them. It says nothing about whether reading one changed anything.
Most of what circulates on this topic is worse than uncertain. The percentages are largely untraceable, at least one is credited to a company we could find no such publication from, and one carries a rejection rate that appears in no source we could find.
Why you cannot just look the number up
Two reasons. The first is ordinary. A survey that asks recruiters whether they read cover letters produces an answer about what recruiters say. It cannot produce an answer about what changes a hiring decision. Those are separate claims, and they get spliced together constantly.
The second is that the numbers you will find are often not attached to the study they name.
Where the “91% of executives” figure comes from
Take one of the pages that comes up for this question. Cover Letter Copilot states that “91% of executives include them (Robert Half, 2026)” and that “84% eliminate candidates who don’t include them for leadership roles,” the first credited inline to “Robert Half, 2026” and the second, in a later section, to a “Robert Half 2026 executive hiring report.”
The 91% is real. It is also from February 14, 2012, and it belongs to a survey published by OfficeTeam, based on telephone interviews with more than 1,000 senior managers at companies with 20 or more employees. The release says: “More than nine in 10 (91 percent) executives polled said cover letters are valuable when evaluating job candidates.” Valuable when evaluating candidates. The release names OfficeTeam, which was a Robert Half division at the time, so the company credit is the one part of “Robert Half, 2026” that holds up. Its companion finding was that 79% of those managers said it’s common to receive cover letters even when resumes arrive electronically. It also publishes the breakdown behind the 91%: 21% said “very valuable,” 70% “somewhat valuable,” 9% “not valuable at all.”
The 84% is not in that release. No figure about eliminating or rejecting anyone appears in it. And Robert Half’s own cover letter hub publishes no statistics of any kind. It is a list of how-to articles.
The numbers with no origin at all
The same page credits a 53% callback lift to “CareerBuilder, 2026.” That number is real too, and the only published 53% we could trace belongs to a field experiment run by ResumeGo between 2019 and 2020. It is the only field experiment on this question that we could find, and it is still circulating in 2026 under another company’s name. Five more percentages on that page, 83 and 56 and 72 and 49 and an 87 that is not the 87% above, are credited to a “2026 survey by TopResume and ResumeLab,” and two others, 68 and 81, to a “2026 survey by Harvard Business Review.” We traced none of those seven.
| Claim as it circulates | Credited to | What it is | Year | Verdict |
|---|---|---|---|---|
| 91% of executives include a cover letter | “Robert Half, 2026” | 91% of executives called cover letters valuable when evaluating candidates, from phone interviews with 1,000+ senior managers | 2012 | Real 2012 OfficeTeam survey, but the finding changed; no Robert Half 2026 publication traced |
| 84% eliminate candidates without one for leadership roles | “Robert Half 2026 executive hiring report” | Nothing we could trace. Not in the 2012 release, and not on Robert Half’s cover letter hub, which publishes no statistics | n/a | No origin found |
| 53% higher callback rates with personalized letters | “CareerBuilder, 2026” | ResumeGo’s three-arm callback experiment, 7,287 applications | 2019-2020 | Matches ResumeGo 2019-2020; the page credits a different company and year |
Which leaves exactly one of the three claims with a study behind it, and it is not a survey.
What did the one real experiment find?
ResumeGo sent 7,287 fictitious job applications to live postings on job boards including ZipRecruiter, Glassdoor and Indeed between July 15, 2019 and January 10, 2020. The applications went out in three versions: no cover letter, a generic one summarizing the applicant’s qualifications, and one tailored to the exact job. Interview callbacks were counted for 30 days after each submission.
The headline everyone quotes is the 53%. Tailored letters drew callbacks at 16.4%, against 10.7% for applications with no letter at all. That is a relative lift on a small base. In absolute terms, subtracting ResumeGo’s own published rates, it is 5.7 percentage points.
The number underneath it gets quoted far less. The generic letter, the kind you write once and reuse, drew 12.5%. Against no letter at all, that is a gap of 1.8 percentage points, by our own arithmetic on ResumeGo’s published rates, or a 17% lift on that 10.7% base. So about two thirds of the total gap comes from tailoring rather than from attaching a document, and the remaining third is the generic letter.
What its citations leave out
Two things about this experiment that its citations leave out. ResumeGo sells resume writing services, which its own page says in its opening line, and cover letter writing services, which its site sells alongside them. A result showing that cover letters work is a result the company profits from. And the three callback rates exist only as data labels on a chart image on that page. There is no results table. The write-up carries no per-group sample size, no confidence interval and no significance test. Any per-arm figure is an assumption of an equal split, which ResumeGo never claims. Whether that 1.8-point generic gap is a real difference or noise is not something anyone outside the company can say.
The same page carries a second, separate instrument: a survey of 236 hiring professionals, 87% of whom said they read cover letters. How those 236 were recruited is not disclosed. In the experiment on that same page, adding a generic letter moved the callback rate by under two points. Saying you read something and being moved by it are different measurements.
One more thing about the experiment. It finished in January 2020, well before ChatGPT existed, so every letter in it was written by a person.
What changed when AI made tailoring cheap?
In September 2025, Jingyi Cui, Gabriel Dias and Justin Ye posted Signaling in the Age of AI: Evidence from Cover Letters. It is a working paper and it has not been peer reviewed.
What the 2025 paper measured
The setting is Freelancer.com, which in 2023 launched a native tool that auto-writes a cover letter tailored to the job description. Access depended on membership plan, which handed the authors a treated and an untreated group to compare. Their final sample is 2,511,592 bids from 17,759 workers across two skill categories, PHP and internet marketing. Freelancer.com supplied the data. The work was funded by the Cowles Foundation at Yale.
On callbacks: the authors estimate that using the tool raised the probability of getting one by 3.56 percentage points, a 51% increase against a pre-rollout callback rate of 7.02%. Across everyone who got access, whether or not they used it, the estimate is 0.43 points.
The callback result is significant at the 10% level and not at the 5% level. The effect tapered off after two months. On actual awards, the paper says its estimates are “too imprecise to draw definitive conclusions,” so nothing here shows anyone won more work, only that more replies came back. And this is a gig-work bidding platform. There is no ATS, no recruiter, no HR function, and a bid is not a cover letter in the sense you would recognize.
A second 51% appears in the same paper, and it has nothing to do with the first. After the tool arrived, the correlation between how closely a cover letter matched the job posting and whether it drew a callback fell by half. Employers shifted toward other signals, the paper reports, including workers’ prior work histories.
In 2019 a tailored letter cost real effort, and the cost is what made it informative. A tool that produces the same artifact for nothing strips the cost out, and the authors describe their result as consistent with what theory predicts if the technology reduces the signal content of the letter.
What this cannot tell you is whether the same thing happened in corporate hiring. We could find no replication of the ResumeGo experiment since AI writing tools arrived.
Does NACE’s employer survey even ask about cover letters?
Barely. The National Association of Colleges and Employers surveys employers every year about new-graduate hiring, and on cover letters its report is close to silent.
Job Outlook 2024, fielded August 2 to September 18, 2023 with 255 employer respondents, contains zero occurrences of the phrase “cover letter” in the full report.
Job Outlook 2026, fielded August 7 to September 22, 2025, contains exactly one, inside its Skills-Based Hiring section, and it is bundled with the resume: “Create a skills-based resume and/or cover letter,” selected by 50.0% of the 108 employers who answered that item, a question put only to employers already practising skills-based hiring.
So there is no standalone cover letter question in either, and nothing in the two to build a trend line from. That is the shape of the evidence here: a 2012 telephone survey whose finding is routinely restated as something it did not say, one experiment and one attitude survey both published by a company that sells cover letters, one non-peer-reviewed working paper about a gig-bidding platform that is not corporate hiring, a pile of percentages with no traceable origin, and an annual employer survey that does not ask the question on its own.
So should you write one?
Write one when the form asks for one. That is not a finding from any study. A blank required field is a rejection you learn nothing from.
Write a specific one, or skip it. If you are not going to make it specific to this posting, the generic version is worth about 1.8 points on a chart nobody outside the company can audit, and the effort is worth more on another application. This is the only claim on this page with support from two unrelated bodies of evidence: ResumeGo’s tailored arm beat both its generic arm and its no-letter arm in 2019-2020, and in the 2025 working paper, workers who used a tool that tailors letters for them got more callbacks. Neither is strong alone. One is a vendor experiment with no published precision. The other is a non-peer-reviewed paper about gig bidding, with an effect significant only at the 10% level. It faded in two months, and its estimates on hiring are too imprecise to read either way. They point the same way from completely different angles, which is worth more than either on its own.
None of it tells you your own odds. ResumeGo measured fictitious applications across unnamed industries and seniority levels in 2019 and 2020, and published no breakdown by job level, so there is nothing in it about entry-level roles specifically.
On AI, the working paper does not say that AI-written letters are bad. It says the link between a letter matching the posting and a reply coming back weakened by half once the matching became free. The same authors found that time spent editing an AI draft correlates with success. That is an association, on a gig platform, and it is the only measurement we found on the question.
If you are deciding where the effort goes, the letter is the smaller lever next to how many applications a search takes. What happens to the rest of your application after you hit submit is a separate question, and we traced that one here: what an applicant tracking system does and does not do.
How we know this: the 91% figure and its methodology come from the OfficeTeam release published on PR Newswire on February 14, 2012, read directly; the claim that no rejection figure appears in it is based on a full read of that release. The callback rates come from ResumeGo’s own research page and the chart image on it, which is the only place those three numbers are published. The AI findings come from v2 of the arXiv working paper by Cui, Dias and Ye, dated November 2025, read in full rather than from its abstract, which carries almost none of its numbers; the significance, fade and awards caveats are the paper’s own wording; the gig-platform caveat is ours. The NACE counts come from full-text searches of the Job Outlook 2024 and 2026 reports. That OfficeTeam was a Robert Half division is not in the 2012 release. It comes from Robert Half International’s Form 10-K for the fiscal year ended December 31, 2011, which names OfficeTeam among the company’s divisions. Where we did arithmetic on someone else’s published figures, which covers the 5.7-point and 1.8-point gaps and the 17% and two-thirds figures, we have said so. Two widely quoted figures were deliberately given no row of their own: a “never read” percentage whose published values are specific to different subgroups rather than to recruiters as a whole, and an 83% we did not trace to its own publisher’s page. Last checked September 8, 2026.