AI methodology and disclosure

Last updated 1 October 2026

The short version

Our five critics — Alex, Mia, Noah, Riley and Sam — are AI personas: consistent virtual editorial voices with published priorities. They are not real people. They have not played the games they review. They evaluate a shared research dossier built from public evidence.

How a review is produced

1. Research once. Our software collects a limited set of public sources for a game — official store listings, developer pages and patch notes, independent reviews and player discussions — while respecting robots.txt and access restrictions.
2. Extract evidence. Store data is read directly by code. An AI model turns relevant passages into short, cited evidence statements; each must match text in its source or it is discarded.
3. Build one dossier. A model summarises the evidence into a structured dossier covering gameplay, progression, monetization, performance, accessibility, updates and open questions, citing evidence throughout. Facts such as price, version and developer are filled in by code from official store data.
4. Five critics, one dossier. Each persona scores the dimensions in its published profile and explains its view with evidence. Overall critic scores are calculated by code from the persona’s weights.
5. Editorial composition. A final pass writes the article, keeping real disagreements between critics instead of averaging them away.
6. Checks and human review. Automated fact, originality, policy and image-rights checks run before a human editor reviews and approves publication.

What the critics will never do

• Claim or imply first-hand play (“I played…”, “the controls felt…”).
• Invent facts, prices, versions or quotations.
• Change a score because of advertising or a publisher relationship.

If human play-testing is ever added, those contributions will be labelled with who tested, on what device and for how long.

Known limitations

• Reviews rely on what sources document. Aspects nobody has written about may be missing or marked as unknown.
• Player reports describe individual experiences and devices; we present them as reports, not universal truths.
• Live-service games change; check the reviewed version and date on each review.
• AI systems can still make mistakes. Our checks reduce that risk but cannot remove it.

Reporting a problem

If something is wrong, please tell us via contact. Confirmed errors are corrected and logged on the corrections page.