Direct answer
Two numbers dominate AI-recruiting content: 27% and 87%. Both are real, both are published, and they are not measuring the same thing. SHRM found that 27% of organizations use AI specifically in recruiting. The 87% figure counts something else entirely. If you quote either one as "how many companies use AI in hiring", you are reporting a number whose denominator you have not checked.
What SHRM actually measured
SHRM published its State of AI in HR 2026 report on 31 March 2026. The survey was fielded 5 to 23 December 2025 through SHRM’s Voice of Work Research Panel. It found that 39% of organizations have adopted AI somewhere in HR, with recruiting as the single largest application at 27% of organizations, ahead of HR technology at 21%, learning and development at 17%, and employee experience at 14%. Adoption splits hard by size: roughly 33% at small organizations, around 35% at midsize, and 60% at the largest. If you work at an enterprise, the 27% headline understates what you see around you, and that is not a contradiction.
Operating notes
- Establish the denominator before quoting any adoption percentage.
- Organizations and individuals are different populations; 27% and 72% can both be right.
- A sample size you cannot explain usually means the original report was never opened.
- Drop any statistic whose attribution changes between publishers.
- Pair vendor-published figures with an independent source.
The sample size that appears two ways
You will see SHRM’s sample cited as 1,722 in some articles and 1,908 in others, which looks like somebody got it wrong. Neither is wrong. 1,908 people started the survey and 1,722 completed it. Publishers pick whichever number their source used, and almost none of them explain the difference. This is a small thing, but it is a useful tell: a roundup that reports a sample size without knowing which one it is reporting usually did not open the original report.
Where 87% comes from
The 87% figure is not an organizational adoption rate. It traces to a Resume.org survey of 1,399 US workers. That is a sample of employees describing what they believe their employer does, not a sample of employers describing what they have implemented. It is a legitimate measure of worker perception. It is not interchangeable with SHRM’s 27%, and stacking them in the same paragraph as if they describe one trend is the most common error in this category of content.
Why credible surveys disagree by 45 points
Published estimates of AI adoption in recruiting run from 27% to 72%, and the spread is mostly definitional rather than a measurement failure. HireVue’s February 2025 report, based on more than 4,000 respondents, found 72% adoption, but it asked individual HR professionals whether they personally use AI. SHRM asked organizations whether AI is adopted in a named function. One counts people who have touched a tool; the other counts companies that have operationalized one. Both are accurate answers to different questions. Before you use an adoption number, establish four things: organizations or individuals, recruiting only or all of HR, generative AI or any automation, and which population was sampled.
The number with no source at all
"99% of Fortune 500 companies use AI in hiring" appears across a large number of 2026 articles with no citation trail. We could not trace it to a named study, sample, or methodology. That does not prove it is false. It does mean nobody quoting it can tell you what it measured, which makes it unusable in anything you would defend to a hiring manager or a compliance reviewer.
A statistic whose attribution changes by publisher
The finding that roughly 26% of applicants trust AI to evaluate them fairly is credited by some publishers to a Gartner survey and by others to a Greenhouse candidate report. One of those attributions is wrong, or they are two separate studies that landed on a similar figure. Either way, a claim you cannot attribute confidently is a claim you should not put in a hiring-manager deck. When two sources disagree about who produced a number, go to the named study or drop the number.
Numbers in this space that do hold up
Verifiable, attributable, and useful: Korn Ferry’s 12th Annual Talent Acquisition Trends, based on more than 1,670 global talent leaders, reports 84% planning to use AI in recruiting in 2026 and 52% planning autonomous agents. A Gartner survey of 114 HR leaders in October 2025 found 88% saying their teams had not yet seen significant business value from AI tools. Those two sit together honestly and describe the actual 2026 condition better than any single adoption percentage: intent is high, realized value is not yet there. Note the Gartner sample is 114 leaders, which is small. Knowing that is the point.
How to check a statistic in about four minutes
Find the original report, not the blog quoting it. Record the publication date, the population sampled, the denominator, and the exact question. If any of the four is missing, treat the number as unusable rather than approximately right. If the original is a vendor report, note that vendors publish numbers that favor their category, which does not make them false but does make them worth pairing with an independent source. This is the same discipline SourcingOS applies to candidate evidence: a claim without provenance is not a weaker claim, it is an unverified one.
Why this matters for sourcing specifically
Recruiters get asked to justify tooling decisions with market data. A hiring manager or a procurement reviewer who checks one of these numbers and finds it does not say what you said it says will discount everything else you brought. Using fewer, better-sourced numbers is not a rhetorical weakness. It is the difference between an argument that survives scrutiny and one that does not.
Copy-paste starting strings
"State of AI in HR 2026" SHRM site:shrm.org
"AI recruiting" statistics 2026 -site:*.vendor.com filetype:pdf methodology sample
"99% of Fortune 500" AI hiring original study source
FAQ
Which AI adoption number should I actually use?
Use the one whose question matches your claim. If you are describing how many companies have operationalized AI inside recruiting, SHRM’s 27% is the defensible figure, with the caveat that enterprise adoption runs closer to 60%. If you are describing how many recruiters have personally used an AI tool, an individual-level survey is the right instrument.
Are vendor statistics unusable?
No. They are often the only data on a specific workflow, and many are methodologically sound. Attribute them clearly as vendor-published, note the sample, and pair them with an independent source where one exists.
Why does SourcingOS care about this?
The same rule governs the product: a claim without provenance is unverified, not merely weak. We apply that to candidate evidence, and we apply it to our own market claims.
Keep reading in Evidence & review
What a public source can actually prove, how provenance is recorded, and where human verification is required.