Accuracy and scope

Every score is a measurement of observable patterns in public data: when an account posts, whether its text repeats, how its follower and following counts relate, how much history it has, and what its posts say. We report what those patterns resemble. We are not asserting that an account is a bot, that anyone was paid, or that any rule or law has been broken. We publish these readings so that patterns in public conversation can be examined openly rather than guessed at.

We make no claim about who owns or operates an account, whether money changed hands, whether an account belongs to any organisation or campaign, or what anyone intended. None of that is in the data we analyse, and any report that reads as though it were should be reported to us using the link on it.

This is automated analysis and it makes mistakes. A high score can be produced by an account that is entirely genuine, and a low score does not certify that an account is authentic. Some ordinary account types resemble the patterns we look for, including:

• businesses, brands, shops, and creators, whose posting is promotional by nature
• fan, hobby, and supporter accounts, which are repetitive and enthusiastic by nature
• news, sports, and aggregator accounts, which post on a schedule
• people who genuinely just joined, and so have few followers, few posts, and no bio
• accounts that were dormant and became active again
• private individuals with a small footprint, which is most people
• people writing in a second language or a non-Latin script, whose phrasing and handles can superficially resemble automated text

Our analysis is instructed to recognise these and to require several independent indicators before returning a high score. It still gets things wrong, which is why the review process below exists.

Treat a score as a starting point for your own judgement, not a conclusion. The written note beside each account is meant to name the specific, checkable facts behind the number, so that you can go and look at the account yourself and disagree with us. If a report makes a claim you cannot verify against the account it describes, that is a defect and we want to hear about it.

Every public report carries a “Request a review” link. Use it. You do not need an account with us, and you do not need to explain who you are. Tell us which account the report names and what it gets wrong, and a person will check it against the evidence the score was based on.

We can and do withdraw published reports. A report that we cannot stand up is removed, and the public link stops working immediately.

We analyse information that is already public on the platform: profile metadata, public posts, and public comments. We do not access private messages, private accounts, deleted content, or anything behind a login that is not our own. See our privacy policy for what we retain and your rights over it.

If a report is inaccurate, misleading, or being used to harass someone, tell us. Use the review link on the report itself, which reaches us fastest because it carries the report reference, or write to privacy@omisphere.ai.