Only 16.1% of employers tell candidates whether they may use AI to apply.
We read the careers pages and job postings of 299 technology and finance employers, the way a candidate would. 48 publish any statement at all about applicants using generative AI. Of those that address CVs and cover letters, most permit it on one condition: the substance must be yours.
The silent majority
83.9%
of employers publish nothing we could find about candidates using AI: 251 of 299, after two independent searches of every employer. 48 publish an official statement (16.1%); 43 of those put it where candidates look, on the careers site, a candidate FAQ or the job posting itself (14.4%).
Census of every active employer on the Morganson board at the freeze on 21 August 2026, pre-registered before collection. “Nothing found” is a statement about our search, not proof of absence; the completeness estimate below bounds it.
What the 48 statements say about CVs and cover letters
Employer-level stance on application materials. 31 permit AI-assisted materials outright, 4 permit them if the candidate declares AI use, 4 restrict them (3 explicit bans, 1 “red flag” discouragement), 9 address interviews or assessments only. Bars scaled to the section maximum.
75% of all statements carry the same condition in different words: AI may help you write it, but what you claim must be true and yours. The restrictions are the exception, not the rule: 4 of the 39 statements that address materials (10.3%), 2 of them in the boilerplate of individual job postings and 1in a graduate-programme FAQ. We do not write “no employer bans AI”: some do, and the number is on the page.
The interview is where the line is drawn
Employer-level stance on interviews and assessments. 35 of 48 statements bar AI during live or recorded interviews or timed assessments; 14 allow it for take-home tasks under stated rules; 27 explicitly welcome it for preparation. Bars scaled to the section maximum.
Published is not the same as findable
37.5%
of the statements (18 of 48) could not be reached from the employer’s careers site within two clicks or from its job postings. We found them only by web search. 25 sit in the careers navigation or one to two clicks from it; 5 live inside the job posting text.
Discoverability is coded per page from how our careers-site sweep reached it; web-search-only pages were found by the second, independent method.
Who publishes
Share of employers with any official statement, by number of live roles on the board. Descriptive only: no inference is claimed. Bars scaled to the section maximum.
By sector. The board is finance-weighted; the technology group is smaller and publishes more often. Descriptive only.
By the country the employer is seeded from on the board (a proxy for headquarters, not a measure of it). Descriptive only.
Say versus show
Surveys ask employers what they think. A University of York / Jisc survey (July 2025) found 7.7% of employers had an applicant-facing AI page and 12.8% said they prohibit AI in CVs and application forms. The Institute of Student Employers’ 2026 survey found two-thirds of employers believe graduates use AI to misrepresent themselves. Prospects’ 2026 survey found 73% of students and graduates used AI in applications. This census measures what employers actually publish where candidates can read it: 16.1% say anything, and restrictions on materials appear in 10.3% of the statements that address them. The gap between what employers say in surveys and what they show candidates is the finding.
If you are applying
Silence is not permission, but it is the norm: most employers have not told you either way. Where an employer has published guidance, it almost always converges on the same rule, so follow it even where nothing is written: use AI to sharpen how you say it, never to invent what you did. If a page asks you to declare AI use, declare it. And treat the live interview as yours alone; that is the one place nearly every published policy draws a hard line.
Methodology
Pre-registered. The protocol, codebook and frozen roster were committed to our public repository on 21 August 2026 before any data was collected; every later change is logged with its reason. Population. All 299 active employers on the Morganson board at the freeze (295 reachable by our site sweep, 299 searched): a census, not a sample, of technology and finance employers with machine-readable careers sites. Two independent searches of every employer.Web search with three fixed queries (accepting a page only if it appeared in the search’s own results), and a careers-site sweep from the careers root to two clicks deep plus the employer’s own job-posting text. 781 candidate pages were pooled; 147 were read in full by two independent model coders (different prompts, employer name masked, page order randomised, quotes extracted before any judgement and verified word-for-word against the archived page). Disagreements, every restriction or disclosure reading, and every unreadable page were adjudicated by hand from the page text. Reliability. Round-one agreement on whether a page was employer-owned was 98.6% but kappa was 0.66 (the prevalence paradox); the pre-registered rule fired, the codebook wording was clarified and all 147 pages were recoded: round two kappa 1 on ownership, 0.962 on the materials stance, 0.902 on the interview stance (Gwet’s AC1 0.975 and 0.933). Completeness. The two search methods found 39 and 33 publishing employers with 24 in common; a Chapman capture–recapture estimate puts the true total at about 53 (48–60), so the headline is 16.1–20.1% and “nothing found” is a lower bound on silence. Sensitivity. Counting only company-wide statements (excluding programme-only, country-site and posting-boilerplate ones) the share is 10%; resolving every adjudicated page the other way moves it between 12.7% and 16.7%. Archiving. Every statement page has a Wayback Machine snapshot or stored text; every statement was re-read within seven days of publication. Names. No employer is named here, in the data download, or in anything we post about this study.
Limitations.The frame is the Morganson board (finance- and technology-weighted, machine-readable careers sites), not “UK employers”. Published is not practised: an employer may hold an unpublished policy or not follow a published one. English-language pages only. A snapshot of 21 August to 31 August 2026; statements change, and one employer’s posting boilerplate was found in two contradictory versions within the window. Two model coders with human adjudication, not two human coders, and the adjudicator was not blind. Both discovery methods favour well-indexed sites. Morganson sells an AI-assisted application tool; pre-registration, the falsification gates and the published aggregates are our answer to that interest.
Data: aggregate CSV · protocol, codebook and changelog in the Morganson repository · questions and corrections: hello@morganson.co. Cite as: Morganson Research (2026), “Can you use AI to apply? A census of 299employers’ published guidance”, edition 2026-09.
The rule the policies converge on is the rule Morganson is built on.
Morganson drafts CVs and cover letters with AI from the experience you actually have, shows its working, and tells you when an employer has published AI guidance for the job you are looking at. See how we disclose AI and the rest of Morganson Research.
Get started free