Skip to content

How we work

Methodology: how figures are found, checked and corrected

Every figure on this site comes from one data file in which each row holds the source link, the source type and the date we read it. We quote sources in their own words and currency, never model or adjust a number, leave out what we cannot read at the source, and correct errors when someone shows us a better source.

Page updated 19 Oct 2026. Figures checked 18 Oct 2026 at their sources.

A fact-checker highlighting printed web pages and ticking a checklist at a desk with a laptop

What is tracked

21

programmes and channels

15

deals with a disclosed value

23

reported unit prices

18

vendor retention rules

Current set checked 18 Oct 2026.

1. One data file, four tables

Everything numeric on the site is read from a single file with four tables: programmes that pay for data, disclosed deals, unit prices, and SaaS retention rules. Pages, the price CSV and the comparison tools all read the same file, so a figure cannot say one thing on one page and another elsewhere. Each row records the source URL, the kind of source and the date it was checked.

2. What counts as a source

In order of preference:

  1. The organisation's own page: a programme's terms, a vendor's help centre, a press release.
  2. Regulatory filings and company reports, such as annual reports and stock-exchange announcements.
  3. Reporting by a named outlet. Where the original blocks reading, we use a faithful republication and name both (for example, a Reuters story read through a syndication partner).

We do not use figures that appear only in secondary commentary, in social posts, or behind a paywall we have not opened. When a figure was dropped for that reason, the row's notes say so.

3. How we quote

  • Wording stays as published, in quotation marks where it matters, so the firmness of a claim is visible.
  • Currency stays as published. We do not convert, average or adjust for inflation.
  • Floors stay floors: “$10m+” or “more than $250 million” are never rounded into totals.
  • Offers, estimates and settlements are labelled as such and kept apart from closed licences.

4. Dates

Each figure shows the date we last read its source. Each page also shows the date its text last changed. The two differ on purpose: a page can be rewritten without its figures changing, and the reverse.

5. What gets a page

A programme gets its own profile only when its rows hold at least 4 sourced facts, out of: what it takes, published figures, payment model, exclusivity, referral terms and other notes such as eligibility. Below that it stays in the directory, so there are no thin pages. Comparisons are built only between two profiled programmes.

6. What the tools do and do not do

The payout range filter and the offer check line up figures that programmes publish. They contain no valuation model of our own. The retention day counter adds a vendor's published windows to a date you enter. All three run in the browser and send nothing.

7. Conflicts and independence

Who publishes the site, and the conflict that comes with it, is set out on the about page. Programmes do not pay to be listed, and listing is not an endorsement.

8. Corrections

Send the page, the figure and a better source to hello@aitrainingdata.broker. Confirmed corrections go into the data file, the row gets a new checked date, and the change is listed in the next monthly update.

Keep reading

FAQ

Frequently asked questions

01Where do the figures on this site come from?

From the programme's or vendor's own pages first, then company filings, then named news outlets. Every figure on the site links to the page it came from and shows the date we last read it.

02Do you convert currencies or adjust figures?

No. Figures stay in the currency and wording the source used, nearly always US dollars. We do not convert, average, inflate or model them.

03Why are some well-known claims missing?

Because we could not read them at the source. Claims seen only in secondary commentary, behind paywalls we did not open, or on pages that blocked our reading are left out or marked as such.

04How often are figures re-checked?

Every row carries its own checked date; the current set was checked on 18 Oct 2026. Rows are read again before each monthly update, and changed figures are noted in that email.

05Where do corrections go?

Email hello@aitrainingdata.broker with the page, the figure and a source. We correct confirmed errors and update the checked date on the row.

06Does a programme pay to be listed?

No programme is listed for payment. A programme is listed when it has a public page offering to pay for or broker data for AI training that we could read.

Stay current

Get the monthly price update

Once a month: deals that disclosed a value, programmes that changed their published figures, and corrections to this site. Free, and one click to leave.

One email a month: new deals, changed programme figures, corrections. See the data protection notice.

Holding an offer already? Compare it with published ranges.