How we know who uses what
Every line on usedby links a company to an AI tool and points to a public page that shows them together. When a vendor stops showing a customer, the line is hidden after repeated checks.
usedby holds 19,662 public lines on 595 tools: 6,599 confirmed, 1,844 declared by the company, 10,650 observed and 569 signals.Latest check Nov 22, 2026: 19,406 of 19,662 public lines checked.
No public page, no line.
A vendor’s word, a pitch deck or one employee’s profile is not a line. 19,658 of the 19,662 public lines link to a public page anyone can open. 53 older signal lines link to a page we have not checked yet. A line the company declares itself is the one exception: it has no page to open and is marked Declared.
Proof is never for sale.
No one can pay usedby to add a line, raise its proof level, move a tool up a ranking or hide a customer. Lines, levels and rankings come only from public pages, the companies’ own statements and the rules on this page, the same for every vendor and every company.
Four levels of proof
Each line carries one level, and every level counts as a customer: in totals, rankings, top tools and shared customers alike. The level shows next to each line, and lists show the strongest proofs first, so you can weigh it. A signal is the weakest proof.
The customer is named in a published story
A case study, a named quote, a press release, or the company’s own public account of using the tool (its own site, or a post by one of its executives, VP or above) names the company and the tool together.

Unilever uses Pigmentpigment.com, captured Nov 22, 2026The company declares it itself
The company tells usedby which tools it uses, after proving it manages its own domain: a message from an address on that domain, or a DNS record we ask it to add. We keep the message and its date. A vendor the company lists as a sub-processor on its own legal page is Declared too, labelled “listed as a sub-processor by the company”: the page states a data-processing relationship, not how the tool is used, so it never counts as Confirmed. Marked Declared on every line.
The vendor shows the company on its own pages
A logo wall or a customers page shows the company. Most of these lines come from a logo or a name matched to one company domain.

Microsoft uses Snorkel AIsnorkel.ai, captured Nov 22, 2026A job post names the tool
The weakest proof: a job post shows that the company’s employees use the tool. Counted as a customer and marked Signal on every line.
Kayak uses FloQastjobs.ashbyhq.com, captured Nov 22, 2026From a public page to a line
589 of 607 tools go through these steps today; older lines came from earlier imports and most keep their source link.
Find the pages
We open the vendor’s customer pages, case studies and logo walls, and public job boards for job posts.
Capture them
A browser takes a full-page capture of each page and stores it with its date.
Read the names
A model reads the logos and names on the capture, together with the page’s own text: image labels, titles and links.
Match the company
Each name is matched to one company domain. A name we cannot match to a domain is left out.
Check again
Twice a day, we capture again the tools whose newest capture is more than 10 days old and compare, so each tool is due again about every 10 days. Once a week, tools not checked for 25 days are checked again. A failed or partial capture never counts as a miss.
19,406 lines have a dated check; the latest was on Nov 22, 2026.

Nvidia uses TrueFoundry
- Proof
- Confirmed, case study
- Source
- truefoundry.com
- Captured
- Nov 18, 2026
- Latest check, all sources
- Nov 18, 2026
When a line leaves
A line leaves usedby only when 3 checks in a row no longer find it and its last sighting is at least 14 days old.
A line backed by a case study or a named quote is never removed automatically: it stays visible.
When 3 checks in a row no longer find a company on the vendor's customer pages and its last sighting is at least 14 days old, usedby states the date of that last sighting next to the proof. It does not conclude that the customer left.
Departures are not published. A logo leaving a vendor page does not prove the company stopped using the tool, so the line is hidden, never listed as lost.
Job posts
A job post that shows employees using a tool makes the employer a customer at the signal level, the weakest proof. It counts wherever a customer counts: totals, rankings, top tools and shared customers, with ties going to the stronger proofs. It is never a case study or a vendor listing, so it never shows as newly listed. Lists and short summaries show the stronger levels first.
516 of the 569 signal lines went through the job-post method: posts on public job boards (Greenhouse, Lever and Ashby), kept only when a classification step finds wording that points to real use. Their date reads as the month of the job post, never as a listing date.
The other 53 signal lines come from an older import whose source is not a job post we have checked. They are being checked again and carry no publication date.
Dating a line from archives
A date on usedby is the first public proof, not an adoption date. We read archived copies of vendor pages (Internet Archive): a line is public “on or before” the first copy that shows it. Archive dating runs twice a day on about 10 tools per run, so many lines have no date yet.
We write “first listed between” two dates only when an earlier usable copy of the same customers page does not show the line and the two copies are at most 186 days apart. Wider than that, only the later date is kept.
Today 9,907 lines are dated from archives, 756 of them with a range.
Facts before you buy
Pricing, certifications, data training, data residency and other facts are read on the vendor’s own pages. Code, not the model, decides what is published: the quote must be found on the cited page, and the page must be of the right kind for the fact. Coverage varies by tool and by topic.
Certifications, data training and data residency carry legal weight: they show only at high confidence or after a human review.
1,833 facts are published on 577 tools, latest check Nov 22, 2026.
How we track changes
The first complete check of a tool’s pages is a baseline: the customers already there are never shown as new. After that, a company that appears is a newly listed customer. A company that comes back after leaving is not new.
Change tracking runs on 551 of 607 tools since Oct 2, 2026. See newly listed customers
What usedby does not claim
Some readings of the data look natural and are wrong.
Questions about the method
Where can I check which companies use an AI tool?
Open the tool’s page on usedby, choose a company line, then open its public source. Read the proof level and date together: Confirmed is a published story, Observed is a vendor listing, and Signal is a job post, the weakest proof. All three count as customers in usedby. A line does not establish a purchase, company-wide deployment or current use by every team; a first-public-proof date is not an adoption date.
What is a line on usedby?
One company and one AI tool, with the public page that shows them together. usedby holds 19,662 public lines today.
Does a line mean the company adopted the tool?
No. It means a public page shows the company with the tool. The date shown is the first public proof we found, not the start of use.
Why does usedby not show companies that left a tool?
A logo leaving a vendor page does not prove the company stopped using the tool. The line is hidden after 3 missed checks in a row and at least 14 days without a sighting, and is never listed as lost. A line backed by a case study or a named quote is never removed automatically.
Are job posts counted?
Yes. A company whose job post shows its employees using a tool counts as a customer, at the signal level, the weakest proof. Each of these lines is marked Signal, and counts in every total and every ranking. usedby holds 569 signal lines today.
How recent is the data?
19,406 of 19,662 public lines have a dated check. The latest was on Nov 22, 2026.
Is a count on usedby a market share?
No. Counts are of the companies on usedby, found on public pages. A tool with more public customers is not proven to have more customers.
See a line that is wrong?
Write to us with the page and the line. We check it against its source. Every correction we made