How it works
How PureFeed decides
what to hide
PureFeed uses three mechanisms: fixed rules that remove AI answer boxes from Google and Bing, the AI labels platforms already add, and 0-100 scores for each feed post. A post is hidden only when a score crosses the limit you chose - and it always says which one.
Three mechanisms
Rules, labels and scores
Rules inside the page
On Google and Bing, PureFeed recognises AI Overviews, the AI Mode button and Copilot answers by their labels and layout, and hides them. Nothing is sent anywhere and there's no limit.
Platform AI labels
When Pinterest marks a pin as AI or YouTube marks a video as altered or synthetic, PureFeed hides it right away, without scoring and without counting toward the daily limit.
Scores for every post
Everything else on LinkedIn, Reddit, X, YouTube and Pinterest is scored by a language model from its public text. Scores are cached by a content hash, so a post is scored once for everyone.
The scores
What each post is scored on
Every score runs from 0 to 100. A hidden post shows the score that hid it, for example "Likely AI 92" or "Neg 84 over your 70 limit".
Likely AI
How strongly the text shows the patterns of AI writing: templated structure, filler, generic hooks and closers. It measures style, not topic.
Negativity
How toxic, hostile, angry or bleak the tone is.
Sensationalism
Clickbait, ragebait and manufactured outrage - writing built to provoke before you think.
Usefulness
Practical value to a reader. Strict also hides posts with very low usefulness.
Scam
Get-rich-quick schemes, fake giveaways, MLM and "DM me" patterns.
Topics
Categories such as politics, crypto, celebrity and self-promotion, used by the topic filters in Pro.
Strengths
Where the line sits
Each strength is a set of limits, and a post is hidden when any of its scores reaches one. Balanced is the default and is deliberately cautious with the AI score: letting some slop through is better than hiding a person.
| Strength | Likely AI | Negativity | Sensationalism | Usefulness |
|---|---|---|---|---|
| Light | 80 | 85 | 88 | - |
| Balanced (default) | 65 | 70 | 75 | - |
| Strict (Pro) | 50 | 38 | 40 | Below 35 |
Mark a hidden post as a wrong call and PureFeed nudges your limits. Open Advanced to set each slider yourself.
Data
What leaves your browser
Sent, for feed posts only
- The post's public title or text
- The public author name and post URL
- Over HTTPS, with no user identifier
Never sent
- Google and Bing searches or results
- Messages, DMs, drafts or passwords
- Browsing history outside supported feeds
See it work on your own feed
Install it, open LinkedIn or Google, and watch the slop fold away with the reason shown.
FAQ
Common questions
How does PureFeed detect AI-generated posts?
How accurate is AI detection?
What data leaves my browser?
Which AI model does PureFeed use?
Why not run the model locally in the browser?
Keep reading
By the PureFeed team · Last updated