X follower audit: check who really follows an account

An X follower audit takes some of the accounts that follow a handle and sorts them by what the evidence shows. Done well, it tells you what a slice of the audience looks like today: how many of those followers post anything, how many repeat the same promotion or the same reply, how many are gone. It cannot tell you who is a person, who paid for a follow, or what the whole following looks like. Treat any score as a read on a sample, and ask what was sampled before you trust it.

What a follower audit measures

Every audit makes two choices: which followers to look at, and what to look at on each one.

On the first choice, some tools try to fetch an account's entire follower list, which gets slow and costly as accounts grow. Others read a sample, usually the most recent followers, because they are the easiest to reach and show what the account is attracting now. A whole-list percentage sounds more complete, but it still rests on whatever signals the tool checks for each follower.

On the second choice, there are two kinds of evidence:

• Profile signals. Avatar, bio, account age, the ratio of followers to following, post count. These are cheap to read, so they scale to thousands of followers, and they are good at spotting empty or abandoned accounts. They are blind to a finished-looking profile that posts junk, and X's own authenticity policy describes manufactured identities built from stock, stolen or AI-generated photos and copied bios.⁠^[1]
• Posts. What the follower writes and who they reply to. Reading posts costs more per follower, so samples are small, but repetition, copy-paste replies and plain silence show up in a way a profile never reveals.

Which one a tool uses decides what its number means. A "twitter follower checker" built on profile fields and one built on posts can give two very different answers about the same account, and both can be honest. How X follower checkers compare sets the main ones side by side.

What an audit can tell you

• What recent followers post. The mix between substantive posting, ordinary posting mixed with promotion, and repeated text is the core of any posts-based audit.
• How quiet the audience is. Followers with no posts of their own are common on large accounts. In the 15 public TrustAudience reports on the current method, between 11 and 33 of the newest 100 followers were Silent on small accounts, against 54 and 72 on the two large ones.⁠^[2] Our guide to silent followers on X explains what that does and doesn't mean.
• Repeated patterns. The same promotion again and again, the same reply under many posts, or follow-for-follow language in bios. These describe behaviour you can check for yourself.
• Change. Run the same method twice and compare. A sample method that stays fixed is what makes two results comparable.

What an audit can't tell you

• Whether a follower is a person. An operator can fill in every profile field, and a human can post the same promotion daily. That is why an honest audit sorts followers by behaviour and stops there.
• Whether followers were bought. It can show what a batch of followers looks like, not how they arrived. Our guide to what bought followers look like covers the shapes that raise the question.
• What the whole audience looks like. A sample of the newest followers is not a random draw from the full list. Older followers may look different.
• Who reads your posts. Someone who opens X every day and never posts has nothing for an audit to judge, but may read everything.
• Anything permanent. Follower lists move. X removes locked accounts from follower counts, so a count can change without anyone unfollowing.⁠^[3] An audit is a snapshot with a date on it.
• Whether an account broke a rule. A pattern in posts is not a finding about intent.

Why "real or not" is the wrong question

People search for how to check if X followers are real, and the question deserves a straight answer: no outside audit can settle it. A real person may post nothing but promotion, and an account run by an operator may look entirely ordinary. What an audit can do is describe what followers do in public, and let you decide what that means for your purpose, whether that is a sponsorship, a launch or a curiosity about your own account.

That is also why the better audits give you counts, not just one number. A score of 64 means little until you can see how many followers were read, how many had posts, and how many were Silent or repetitive.

How a TrustAudience follower audit works

Anyone can request a check of a public X handle without signing in. The steps are the same every time, and they match the method page.⁠^[4]

1. Sample. We read up to 100 recent followers and up to 10 public posts or replies from each. Account, follower and post data comes from TwitterAPI.io. We don't assume the provider's ordering is random, so we never extrapolate the sample to the full audience.
2. Sort. A model sorts each selected follower with posts into a type, from their public posts and nothing else. We set two types ourselves. A follower with no posts of their own, or too few to judge, is Silent, and so is a private account, which we count without reading its posts. A suspended or deactivated follower is Unavailable.
3. Score. There are six types, each with a weight: Substantive 100, Mixed 60, Repetitive (repeated promotion 25, repeated replies 0), Silent 40, Unavailable 25, and Unclear, which is not scored. The Audience score is the mean of the weights over every follower except Unclear ones.
4. Limit. A score needs at least 5 counted followers. If more than 20% of the selected followers are Unclear, there is no score and we ask you to try again.

Every report shows the sum behind its score, so the arithmetic can be checked, plus how many followers were selected and readable and how many posts were read. You can see how a finished report reads on @pewdiepie or @supabase, and the audience board on the home page lists the others.

Repetitive describes behaviour in the posts we read. It is not a finding that an account is automated, and ordinary networking or self-promotion can be legitimate. The score can't establish whether an account belongs to a person, whether a post was written by AI, or how good the whole audience is, and it does not verify audience quality.

What a check costs

Existing public reports are free to view. The first 1,000 checks are free, one per browser and per handle. After that, a fresh check is US$1, and US$5 buys a 30-day pass that refreshes one handle every day. Both are one-time payments, not subscriptions.⁠^[5] The pricing page shows the current allowance. A check publishes an aggregate report for the handle on the public board, and the page asks you to acknowledge that before it starts; it never shows an individual follower.

Reading a result without fooling yourself

• Start with the counts. Look at how many followers were read, then how many had posts. A result built on 40 readable followers is thinner than one built on 90.
• Compare like with like. In our early study of 15 accounts, the two large accounts had far more Silent followers than the small ones, which pulled their scores down without implying anything sinister.
• Look for what you can verify. If a report says a follower repeats the same reply, open the account and see.
• Keep the date in mind. A score describes the followers read on the day of the check.

Check it by hand

You can run a small follower audit with no tool at all, in about half an hour:

1. Open the account's followers list on the web and write down the first 25 handles, newest first.
2. Visit each profile and give it twenty seconds. Is there any post of its own, or only reposts? Is the account private, suspended or gone?
3. For anything with posts, read the last ten. Sort the follower into one of four piles: ordinary posting and real replies; a mix of ordinary posts and promotion; the same promotion or reply repeated; or nothing of its own to read.
4. Count each pile, plus the private and gone accounts. Write the date at the top of the page.
5. Repeat on a different day from the top of the list. If the two counts differ a lot, your sample is too small to say much, and that is itself worth knowing.

Twenty-five hand-sorted followers carry wider error than a hundred, and the same caveat applies to any sample: it describes recent followers, not people in general.