We checked five AI statistics we were about to publish. Three had no source.

Every AI agency page carries numbers. Ours did too. In September 2026 we made somebody find the primary source for each one before it shipped.

Two survived intact. Here is the whole audit, including the one that was on our own homepage.

1. “A 1,445% one-year surge in demand for multi-agent systems, says Gartner”

Verdict on 7 September: misattributed. Removed. On 24 September: Gartner’s, but not what we said. Still removed.

This was on our homepage. When we first chased it back we found it in a press release from Belitsoft, a software development company, found no Gartner publication carrying it, and called it a vendor’s number wearing an analyst’s name.

We were wrong about that. Gartner’s own article on multiagent systems, published on 18 December 2025, says it plainly: “Gartner reports a 1,445% surge in MAS inquiries from Q1 2024 to Q2 2025.” Belitsoft had quoted Gartner correctly. Our search had failed, and we blamed the messenger.

What was wrong was our sentence. Gartner counted inquiries to its analysts, over five quarters; we had written a one-year surge in demand. That is a smaller claim dressed as a bigger one, and the second mistake, on top of it, is the one to remember: a search that finds nothing is a fact about the search.

2. “40% of enterprise apps will run task-specific AI agents by 2026, says Gartner”

Verdict: true, but misworded. Corrected and dated.

This one is genuinely Gartner’s: press release of 26 August 2025, predicting 40% by 2026, up from less than 5% in 2025. But Gartner published “feature”, and we had written “run”.

One verb. It is the difference between quoting an analyst and paraphrasing one, and only one of those is checkable. It now matches, and carries the date.

3. “Desktop computer-use scores in the low 80s on OSWorld”

Verdict: stale, and it had stopped making its own point. Replaced.

We recorded the leaderboard’s best as 86.1%. Read again on 24 September 2026, the best model shows 86.0% and no entry reads 86.1%, so we cannot reproduce our own figure; the point survives it. The human baseline on OSWorld is 72.36%, so agents now score above humans on that benchmark.

We had been using “low 80s” to imply a ceiling. At 86% against a 72.4% human baseline it implies the opposite. The number had drifted past the argument it was hired to make.

4. “Across 500,000+ cold emails, AI personalisation performed 25% worse than sending nothing”

Verdict: no source found. Contradicted. Struck.

We could not locate any study matching this claim. What we did find was Hunter’s State of Email Outreach 2026, drawn from 31 million emails, reporting that two custom attributes lift reply rate by 56%: 5.6% against 3.6%.

That is not a small discrepancy. It points the other way, from a dataset sixty times larger than the one our claim cited.

5. “One human running two AI seats books 1.9× more meetings per dollar”

Verdict: no source found. Struck.

Nothing. No study, no dataset, no vendor even claiming it. It had presumably been absorbed from a conference slide.

What replaced them

The method, so you can do it to us

1. Find the primary source, not the article citing it. Most “according to Gartner” claims resolve to a blog citing a blog citing a press release.
2. Check the date. Benchmarks move. A true 2024 number can be a false 2026 one.
3. Check the wording against the original. “Run” for “feature” is how a checkable claim becomes an unfalsifiable one.
4. Ask who profits. A vendor dataset is not worthless (Hunter’s 31 million emails are worth more than most academic samples) but it must be labelled.
5. Write the date you checked it, so the next person knows how stale it is without repeating the work.

Every page on our site now lists the sources behind its figures and the date they were last checked, and our build fails if a page has none, or if a source goes more than thirty days past its re-check. That is not a promise. It is a test that runs.

What this does not tell you

How reliable AI statistics are in general. Five claims is not a survey: we checked these because we were about to publish them, so three of five failing is our rate on our own drafts, not an industry rate. It may say more about where we had been reading than about the field.

“No source found” means none we could reach with reasonable effort, not proof that none exists. If you hold the primary source for any of the three, send it: we will correct the note and say what changed, which is the same promise the register makes.