Do fake followers hurt your reach on X? What the open-source algorithm shows
Fake and inactive followers don't trigger a direct penalty in the code X has published for its For You feed. They hurt reach more quietly: ranking is driven by how likely real people are to engage, so followers who never read or interact add to your count but not your distribution. To see how much of your own audience looks low quality, scan your followers with Revealed.
The answer in full
X has published the code behind its For You feed on GitHub, and it answers the follower question better than guesswork. Every feed is built for one viewer at a time. Posts come from accounts the viewer follows and from accounts they don't, and a model called Phoenix predicts how likely that viewer is to like, reply, repost or dwell on each one. Those predictions are combined into a score, and posts from accounts the viewer doesn't follow are discounted. That makes your followers your main distribution channel, and followers who never engage give the system nothing to work with. The published code shows no rule that penalises an account for having fake followers, though some of X's spam rules are not public. X also scores account credibility with PageRank over the follow graph and engagement, so follows from low-credibility accounts carry little weight. The practical goal is an engaged audience, not a bigger number.
Step by step
- 01
Check your labels with Under the Hood
X is piloting a transparency tool called Under the Hood. It shows aggregate stats on labels applied to your account and posts that can limit visibility. It is still rolling out, so you may not see it yet.
- 02
Scan your own followers
Open your profile on X and start a Revealed scan. It reads your public follower list in your own signed-in browser, and the data stays in Chrome extension storage on your device. Scanning is available on the Free plan. X restricts automated access, so keep scans modest and stop if X shows you a challenge.
- 03
Review audience quality (Pro)
Open Audience quality to see how complete your followers' profiles are and how many carry caution signals such as no bio, a default avatar, very few posts or a high following ratio. Revealed also highlights bios that look copy-pasted. These are prompts to look closer, not verdicts on any single account.
- 04
Stop the inflow
Don't buy followers, and avoid follow-for-follow schemes. Both add accounts that are unlikely to read or engage with what you post.
- 05
Grow the engaged core
Use what the scan shows about who follows you to decide what to post and which accounts to engage with. Engagement from real people is what the ranking system is built around.
At a glance
| System | What it does | Why it matters for followers |
|---|---|---|
| user-cred-v2 | Runs PageRank over the follow graph and engagement to give each account a credibility score | A follow from a low-credibility account passes on very little weight |
| bdsm | Reads the sequence of actions an account takes over time to spot inauthentic behaviour | Bot-like accounts are identified from how they behave, not just how they look |
| agatha | Labels accounts by blocks and reports relative to likes, plus spam-suspension labels | Spam accounts collect labels that feed into visibility filtering |
How the For You feed picks posts
The code is public at xai-org/x-algorithm. Each time someone opens For You, X gathers candidate posts from two places: accounts the viewer follows, and accounts they don't. It filters out posts older than 48 hours, posts the viewer has already seen and posts from accounts they have blocked or muted.
A model called Phoenix then predicts how likely this particular viewer is to like, reply, repost, dwell on or click each post, and how likely they are to mute, block or report it. Those predictions are combined into one score. The scoring is personal: it asks whether this person will engage, based on their own recent behaviour.
Why inactive followers drag reach down
Posts from accounts the viewer doesn't follow are multiplied by a factor below 1. Reaching strangers is harder than reaching your own followers, which makes your follower list your main distribution channel.
A bot or abandoned account never likes, replies or reads, so it produces no engagement for the system to work with. Two accounts with 10,000 followers can see very different reach if one has mostly active readers and the other has thousands of accounts that will never see a post.
Myth: one report cancels out 468 likes
This claim spread after the code was published, and X added a correction to its README. The final score is the sum of each action's weight multiplied by the predicted probability that the viewer takes that action.
The weights apply to predicted probabilities for one viewer, not to the raw count of likes or reports a post has. A report carries a large weight, but the chance that any one viewer reports an ordinary post is tiny. Raw counts of anything, including followers, matter less than most people assume.
What to keep in mind
- Revealed shows who follows you. It does not measure reach, engagement or how X ranks your posts.
- Revealed does not label individual accounts as fake. Audience quality signals are prompts to look closer, not proof.
- This guide describes the For You code X published on GitHub. X can change its systems, and some of its rules are not public.
Common questions
01Does X penalise you for having fake followers?
The code X has published shows no direct penalty for fake followers. They hurt reach by adding no engagement, and follows from low-credibility accounts carry little weight in X's credibility scoring.
02Is the X algorithm really open source?
Yes. The For You feed code is on GitHub at xai-org/x-algorithm under the Apache 2.0 licence. Some files, such as certain spam rules and LLM prompts, are held back to stop people gaming the system.
03Does buying followers help with the algorithm?
No. Ranking is based on how likely real viewers are to engage. Bought followers are unlikely to engage, and a follow from a low-credibility account passes on little credibility.
04How long do posts stay eligible for the For You feed?
The published code filters out posts older than 48 hours before ranking.
05Can Revealed tell me which of my followers are fake?
Revealed does not label individual accounts as fake. Its Pro Audience quality view shows how complete your followers' profiles are, how many carry caution signals such as no bio or very few posts, and which bios look copy-pasted, so you can see whether part of your audience is worth a closer look.
06Does Revealed upload the followers I scan?
No. Scans run in your own browser and the audience data (profiles, bios, analytics and exports) stays in Chrome extension storage on your device. Revealed’s servers only handle your account, plan and billing.
07Can scanning with Revealed get my X account suspended?
It's possible. Revealed reads follower lists through your own signed-in X account, and X restricts automated access to its service. X can challenge, restrict or suspend accounts it believes are automated, including linked accounts. Revealed only reads data and paces scans to reduce the risk, but it can't control X's decisions. Keep scans modest, stop if X shows you a challenge, and avoid scanning from a business or client account.
Sources and further reading
xai-org/x-algorithm on GitHub
The For You feed source code and README, including the scoring weights and X's note on how they work.
Under the Hood
X's pilot tool showing labels on your account and posts that can limit visibility.