Make room for real people.
Review accounts, manage your block queue, and track results.
Open your workspace
Enter the workspace access key supplied by your administrator.
Detection scores are experimental. Accounts enter the queue only when you choose Block or Unblock. The browser worker must be connected to execute them.
Add a designated test account
Use an account you own or a volunteer. Read its numeric ID with the extension's profile inspector.
Research search
Quickly find collected comments by author or by words in the comment.
Exact author username, with or without @. Matches ignore letter case.
Searches the sampled comment archive. Missing results do not mean a person has never commented. Results appear in archive order, 25 per page.
Enter a username or keyword to begin.
Bot scoring
Choose which patterns should raise an account’s review score.
Higher score = more reasons to look closer. Matching rules add points. Points are not a percentage or proof that someone is a bot. You still choose who to block.
Username keywords
A username containing any saved keyword gets the Username match points above, once per account. For example, user_ matches user_123. Adding or removing keywords updates account scores immediately; point changes apply when you save scoring.
Image examples
Profile image examples
Compare account pictures against these examples. The closest match counts once.
Adding or removing an example updates available image comparisons immediately. Older accounts may need their images collected again. Point weights and the match threshold use the scoring settings above.
PNG, JPEG or WebP · up to 2 MB per image · up to 32 examples
Try a message. See the evidence.
Check promotion, toxicity, anger and message context. Add linked posting history to compare accounts and coordinated campaigns.
Examples are fictional. Review the input, then choose Analyze messages.
What this test runs
Runs content, account behavior, cross-account coordination, campaign and reply-thread checks. Rage-bait review looks for reaction-seeking or provocative wording alongside hostility; intent remains unconfirmed. Reply counts alone do not prove rage bait. Only supplied replies are analyzed, and this form does not fetch missing TikTok replies. Full analysis also uses local semantic matching, multilingual toxicity, anger, a specialized AI-content model and a language model for context. A stronger local reviewer checks up to 12 unique comments per test for overlapping abuse categories, including threats, insults and harassment. Larger tests prioritize context and uncertainty and include four audit selections. A prompt-token budget also limits processing. Other comments keep their baseline analysis; deeper review is marked when skipped. Disguised-spelling checks retain the original text. Context is supplied and unverified. The coverage panel shows which checks ran and what data is missing. Supplied profile observations add bounded support; missing fields stay unknown. AI-content estimates need at least 20 usable distinct texts for an account summary. Model outputs and review points are not calibrated bot probabilities. Quotation, satire, dialect and mixed languages can be misread. Political opinions, anger, profile appearance and coordinated human activity do not prove automation. Tests are temporary and do not change saved accounts or the block queue. Completed analyses expire from the analysis service after 15 minutes.