Methodology

Counts and percentages

A detector badge such as “11/14 Human” means 11 of 14 posts in the sample were classified as Human. Percentages are calculated from those counts: 1/2 becomes 50% AI · 50% HUMAN, and 2/2 becomes 100% HUMAN.

The simplified “AI” group includes every result not classified as Human, including mixed and AI-assisted results. These percentages describe the share of posts in the sample, not the amount of AI text or a probability of AI use.

For numeric results, we group accounts as HUMAN at 66% human posts or above, MIXED from 33% to below 66%, and AI below 33%. We use the original counts before rounding. Displayed percentages are rounded to whole numbers that add up to 100%. Tiny nonzero shares use <1% and >99%, reserving 100% for an entirely uniform sample.

Reading a result

Check the sample size and date alongside each result. A result based on one post describes that post, not the whole account. When a detector result supplies a MIXED classification without a human/AI breakdown, we show MIXED without a percentage.

Results describe a dated sample and may change as more information becomes available. They do not represent an account’s entire history.

Community reports

You can submit an X account, a HUMAN, AI or MIXED assessment, a supporting link and an explanation. We review every report before it contributes to the public record. Approval is a moderation decision, not verification of AI use. Reports are not counted as analyzed posts or converted into detector percentages.

For accounts without a detector result, agreeing reports produce HUMAN or AI; a MIXED report or conflicting assessments produce MIXED. The badge says “Community report” and shows the number of approved reports. These may cover overlapping posts and are not counts of unique people or unique posts. We count the same evidence link for an account only once.

When a profile has both a detector result and community reports, we display them separately. Community reports do not enter category detector averages. Rejected, withdrawn and deleted reports never contribute to public counts. We keep evidence and explanations private.

Limitations and corrections

Detector results can be wrong. Community assessments are opinions. Neither proves AI use. If a result needs correction or an account should be removed, use the correction or removal request on its profile. We review these requests privately.