ATS résumé screening in 2026 — sources for every number

Every claim in the video, with its source, its method and its weakness. Links go straight to the original — we are not the source, they are.

1. Our own measurement

Taken 26 August 2026 with the YouTube Data API v3: the top 10 results for how to write a resume, then each video's publication date and view count.

Figure	Value	Method
Median age of the top 10	66.5 months	search.list then videos.list, US / English, signed-out ranking
Videos under 12 months old	1 of 10	same
Median views per month	22,317	total views ÷ age in months
Search volume (Google, US)	22,200 / month	Google Ads Keyword Planner via DataForSEO

What is wrong with it. Views per month assumes a flat rate. That is false — a video decays. The error inflates older videos, which works against our own argument, so we kept it and said so on screen. YouTube's ranking is also personalised; ours is the signed-out US ranking, not yours.

2. The "75% of résumés are auto-rejected" claim

We could not find a study. The figure traces back to 2012 sales material from Preptel, a vendor selling résumé optimisation, which went out of business in 2013. No methodology, no sample size, no survey has ever been published.

Be exact about this. These are industry blogs, not studies. What they establish is a negative: nobody can produce the original research. We did not establish that the number is false — nobody has run that study either. A number no one can source is not a fact; that is all we claim.

What we deliberately left out. A survey reporting that 92% of recruiters do not auto-reject résumés — n = 25, published by a résumé builder. It has the exact flaw we are criticising, so we did not use it to win the argument.

3. What is actually documented

Hidden Workers: Untapped Talent — Harvard Business School with Accenture, 15 September 2021, Joseph B. Fuller and Manjari Raman. Survey of 8,720 workers and 2,275 executives across the US, UK and Germany.

Figure used in the video	Where we read it
99% of Fortune 500 companies use an automated screening system	Harvard Gazette
27 million qualified Americans those systems do not surface	Harvard Gazette
Half of US companies filter out anyone not employed in the last six months	Harvard Gazette

Weaknesses. Published in 2021. Co-authored by Accenture, which sells recruitment consulting. The 27 million is an estimate, not a count.
And one figure we did not use: the widely quoted "88% of employers admit their system filters out qualified candidates" appears only in secondary coverage. We did not read it in the report itself, so it is not in the video.

4. The "7.4 seconds"

Weaknesses. n = 30 recruiters. Published by a job board — a company with something to sell. Neither the roles being screened nor the instructions given are disclosed. Screening for a warehouse shift is not the same act as screening for a vice president. Useful as an order of magnitude. Not a fact.

5. What actually changed: the law

Date	Rule	Source
5 Jul 2023	New York City Local Law 144 — annual independent bias audit of any automated employment decision tool	Deloitte
1 Jan 2026	Illinois HB 3773 — bans AI producing discrimination, and requires notifying the applicant	National Law Review
2 Aug 2026	EU AI Act — recruitment classified high-risk: risk management, bias testing, logging, human oversight	Crowell & Moring · Hunton

6. And the half that rarely gets mentioned

Fact	Source
A proposal (the Digital Omnibus) would push the EU high-risk obligations from 2 August to 2 December 2026	DLA Piper
Colorado moved its AI act twice — Feb 2026 → Jun 2026 → Jan 2027 — and scaled the requirements back	Alston & Bird · Skadden
New York State Comptroller audit (Jul 2023 – Jun 2025): across the same 32 published bias audits, the city regulator identified 1 issue of non-compliance; the Comptroller identified at least 17 potential ones. And 75% of test calls to the 311 complaint line never reached the department.	DLA Piper

Nuance. A comptroller's job is to find problems — that is the function of the office. "17 potential issues" is not "17 proven violations", and the video says potential.

Raw data

The YouTube measurement is a JSON file: keyword, volume, CPC, and for each of the top 10 the title, channel, publication date, view count and derived views-per-month. Ask for it in the comments and we will publish the file here — we would rather host it than have you take our word for it.