ResumeAI — State of ATS 2026
A free ATS resume builder sitting on the only hand-verified dataset of which applicant tracking system 743 large employers actually use.
Founder self-reports ≈$20/mo — unverified. Only provider-verified revenue gets the green badge.
Read the full write-up →What it was
Two things in one codebase. 1) A consumer product: a free ATS-safe resume builder (9 templates, AI bullet generation, cover letters, interview prep, job tracker), a free ATS resume checker, and a Chrome extension. Free to build; the clean, watermark-free download is paid ($12.99 one-time pass, or $19.99/mo). 2) The asset that actually makes it unusual: State of ATS 2026 — 743 Fortune 500 / Global 2000 employers mapped to the applicant tracking system each one really runs, 707 of them verified by hand against the apply-URL host on their live careers portal (June 2026, fully re-verified July 2026). Plus 279 more machine-verified employers in an extended tier. It is MIT-licensed and free on purpose, mirrored to GitHub, npm, Kaggle and Hugging Face, and served through a public JSON API, an OpenAPI spec, and a remote MCP server for AI agents. On top of the data sits monitoring nobody else runs: ~550 of those employers get an automated weekly re-probe against live vendor infrastructure, detected changes go to human review before publication, and confirmed vendor migrations publish to a public changelog + RSS with a curl-able evidence host on every entry. The July re-verification found ~24% of the hardest-to-check records had rotted in a single month — acquisitions, renames, silent ATS migrations — which is the pitch to data buyers: every static ATS dataset is wrong within a quarter, and this one re-checks itself. Built for two audiences: job seekers who want to know which parser will actually read their resume, and companies (auto-apply agents, job-data vendors, unified-API platforms, HR-tech marketers) that need employer-to-ATS attribution to stay true.
To be straight about the status: this is NOT shut down. The site is live, the weekly verification job runs every morning, and there is one active $19.99/mo subscriber (renews Sept 3) plus a one-time pass sold this week. I'm listing it because I'm a solo founder taking a full-time job and I'd rather it go to someone who will run it than let it decay. The form only offered "shutdown" or "pivot" — neither is accurate. What actually happened, honestly: the product converts almost nobody, and I can prove why rather than guess. ~34 real humans a day (bot-filtered; raw analytics inflate ~2-3x), 153 registered users, 49 signups and 179 resumes built in the last 30 days, and $14.97 collected in lifetime revenue across 3 customers. Consumer resume tooling is a commodity with a brutal structural problem: the job search ENDS, so there is no retention to build on, and the incumbents win on a decade of SEO plus subscription billing that relies on people forgetting to cancel. I tested one-time pricing ($12.99), subscription ($0.99 first month), and reverse trials. All near-zero. What is genuinely valuable is the thing I built to support the consumer product: a portal-verified employer→ATS dataset with an automated re-verification loop. In July I re-checked all 743 employers and found ~24% of the hand-verified records had rotted in ONE MONTH — 14 acquired brands still listed as independent, 2 defunct, ~60 silent vendor migrations. Taleo alone fell from 39 employers to 18. That's the moat: any static ATS dataset is wrong within a quarter, and nobody else re-checks. A Harvard-affiliated health writer found the report through search and cited it unprompted; a WSJ reporter asked to see it; an auto-apply SaaS CTO took a call. Where the money didn't come from: ~120 cold B2B emails (0 sales), a $299 self-serve snapshot (0), a $199/mo feed (0). Where it did, twice in one week: a Microsoft Copilot recommendation that converted in 6 minutes, and a Bing visitor recovered by a lifecycle email. The AI-referral channel is small but real and growing (Copilot referrals 4 → 6 → 16/month).
Users didn't churn so much as never convert: they built a resume free, took the value from the on-screen preview, and left. Measured funnel over 30 days: 179 builds → 49 signups → 4 checkout starts → 1 payment. Experiments tried: one-time $12.99 pass instead of subscription (the current primary offer, and it did produce the two recent sales), $0.99 first month, 24-hour reverse trials (0 of 63 converted), a stronger download watermark, personalizing the paywall with the target company's verified ATS, exit-intent capture, and lifecycle recovery emails (this one works — it produced the subscription sale). The honest read: the gate isn't the problem, arriving-with-intent is. Visitors who came from an AI assistant recommendation converted; visitors who came from a search-engine listicle didn't.
Spending five months building a resume builder before testing whether anyone would pay for one. The dataset underneath it — the part that got cited by a journalist and pulled in an AI-assistant referral that converted in six minutes — was a side project I built to make the builder smarter. I should have started there and skipped the builder entirely.
1. Build the asset nobody can copy, not the product everybody has. The resume builder took months and converts at ~0%. The dataset was a side quest to make the builder smarter, and it's the only thing anyone ever wrote about, replied to, or asked for. 2. Distribution is the whole game and I confused "shipping" with "distribution" for months. Both sales came from surfaces I didn't control: an AI assistant recommending us, and Bing. Google never indexed us meaningfully (DA-15 sandbox). 3. Being the source AI assistants cite is a real, underrated channel in 2026. llms.txt, structured data, an OpenAPI spec, and an MCP server cost a weekend and produced the first full-price sale — a Copilot referral that paid within 6 minutes of landing. 4. Verify your own data before you sell it, then verify it again. My "hand-verified" dataset had 14 acquired companies listed as live employers a month after I verified it. Anyone could have falsified my central claim with one curl. Re-verification isn't hygiene; for a data business it IS the product. 5. Cold email into a market that hasn't budgeted for your category is a rounding error. ~120 sends, zero sales. One inbound citation from a journalist did more. 6. Price for how the customer's need is shaped. Job searches end, so subscriptions fight the user. But one-time pricing on a commodity is also just... a commodity. The mistake was assuming pricing was the problem when the real problem was that nobody arrived wanting it.
