Editorial Standards
Report AI publishes statistics about artificial intelligence. A statistics publication is only as useful as its rules, so these are ours: how we choose sources, how we say how sure we are, how we keep figures current, how we correct mistakes, and how we keep the work independent.
Thank you for reading this post, don't forget to subscribe!
1. Sourcing
- Every figure has a named source and a date, shown next to the figure, not only in a list at the end.
- Primary sources first. We cite the study, filing, dataset or official statement that produced a number, not the article that repeated it. Where we could only reach secondary accounts, we say so and rate the figure accordingly.
- We state what a figure measures. A number is quoted with its definition, scope and sample where those matter. “Contains AI content” and “written by AI” are different claims, and we do not let one stand in for the other.
- When credible sources disagree, we publish the range and explain the disagreement rather than picking the most convenient figure. We do not build composite scores out of trackers that disagree.
- External links are citations. Where we rely on an outside report, we summarise it in our own words with attribution; we do not republish other publishers’ work.
2. Confidence ratings
Every figure carries one of four ratings. The rating describes how directly we could verify the figure, not how important or plausible it is.
| Rating | Meaning |
|---|---|
| HIGH | Stated directly in a primary source we read ourselves, with a date. |
| MEDIUM | Reported consistently by credible secondary sources, or from a primary source we could not open directly. A figure we have not read at source is never rated HIGH, however authoritative its origin. |
| DERIVED | Calculated by Report AI — an interpolation, conversion or check — with the inputs and method shown on the page. |
| CONTESTED | Credible estimates disagree too widely for a single figure; we publish the range. |
3. How figures are written
Statistics are quoted out of context constantly, by people and increasingly by AI systems that summarise the web. We write each key figure so that it still means the right thing when lifted out on its own: the number, its unit, what it measures, the source and the date in one sentence. “The FBI’s 2025 Internet Crime Report counted 22,364 complaints with an AI descriptor and $893 million in losses” — not “losses hit $893M”.
A figure that was correct at its date keeps that date. When a newer figure replaces it, we show the series rather than overwrite the old value, and where an old figure is the right denominator for a dated disclosure we keep it, labelled.
4. Freshness and review
Every index and report carries a review cadence, set by how quickly its figures can go out of date.
| Cadence | Review | Used for |
|---|---|---|
| Critical | Monthly | Figures that could be wrong within 30 days, live litigation or negotiations, and statutes about to take effect |
| Standard | Quarterly | Most statistics pages |
| Stable | Twice a year | Definitions and settled history |
An update is never just a new date. Each reviewed page shows a “What changed” block listing the previous value, the new value and the reason, with recent revisions kept visible. If a review finds nothing has changed, the page says “Reviewed — no change” with the date. The visible “Updated” date and the page’s structured data always agree.
5. Forecasts
We report forecasts as forecasts: who made them, when, and when they resolve. A projection is never quoted as a measured outcome, including after the year it was forecasting arrives. We score third-party forecasts against what actually happened on our predictions scoreboard, using a published scale that includes “unfalsifiable” for forecasts whose method is not disclosed. Where we publish our own probability estimates, they are labelled as editorial judgement.
6. Corrections
Errors are corrected as soon as they are found, not at the next review. The corrected page carries a visible note stating what it said before and why it changed; the figure is corrected everywhere it appears, including on other pages; and every correction is recorded permanently in our corrections log. We also maintain a fact-check register of AI statistics that circulate in error elsewhere. To report an error, use the contact page.
7. How we use AI tools
We use AI tools in our research, drafting and checking, including to search for sources, extract figures and test arithmetic. That does not change the rules above. Every figure must trace to a source, and its confidence rating reflects how directly that source could be verified: where a primary source could not be opened, the figure is rated MEDIUM and the page says why. AI tools are not cited as sources. Editorial responsibility for everything published rests with the Report AI editorial team.
8. Independence and disclosure
Report AI is published by Agile Agilist, a consultancy that offers AI and Agile transformation services. Report AI’s role is measurement: it reports what the data shows, including where the data reflects badly on AI adoption, consultancies or vendors.
- No sales copy. Report AI pages do not promote our publisher’s services.
- Disclosed links. Any link from a Report AI page to our publisher’s commercial pages carries a visible disclosure.
- No paid placement. Figures, rankings and ratings are not for sale, and companies we cover do not get to approve that coverage before publication.
- Advertising is separate. Where advertising appears on the site, it is served separately from editorial content and has no influence over it.
Frequently asked
What does a MEDIUM rating mean?
That the figure is reported consistently by credible sources, or comes from a primary source we could not open ourselves. It is not a judgement that the figure is doubtful; it is a statement about how directly we verified it.
How often are pages updated?
Monthly, quarterly or twice a year depending on how quickly their figures change. Each page shows when it was last reviewed and what changed.
Does Report AI use AI to write its reports?
We use AI tools in research, drafting and checking. Every figure must still trace to a named, dated source, and confidence ratings show how directly each source was verified. Responsibility for what is published rests with the editorial team.
Who publishes Report AI?
Agile Agilist, a transformation consultancy. Report AI pages do not promote its services, and any link to them carries a disclosure.
How do I report a mistake?
Through our contact page. Confirmed errors are corrected immediately and recorded in the corrections log.