# How many X followers are bots? What we found reading the newest followers of 15 X accounts (early look)

By Jonah Ease (https://trustaudience.com/about) · Published 4 Oct 2026 · Last checked 4 Oct 2026

Public posts can't prove who is a bot, so this study doesn't count bots. It shows what the newest followers of 15 X accounts actually post. Of 1,325 newest followers counted on 3 and 4 October 2026, 527 posted substantively, 394 were Silent, and 161 showed repetitive promotion or a repeated reply pattern. Audience scores ran from 47 to 76, with a median of 63.[1] Treat it as an early look: we'll update it as more accounts are checked.

## What we read, and when

Every figure here comes from TrustAudience's public reports, read through the public API on 4 October 2026.[1] We saved the API responses and the script that computes each number from them, and we'll share the script and data with anyone who asks. Each account's counts are also public on its report page.

- **Who:** every account on the public board with a report on the current method. That is 15 accounts, all checked between 3 October and 4 October 2026 (14 on the 3rd, one on the 4th).
- **Method:** `audience-context-2026-10-03-v3`, model gpt-6-luna. For each account we take up to 100 of its newest followers, as our data provider returns them, and read up to 10 recent public posts or replies from each. The model labels followers who have posts. We set Silent and Unavailable ourselves. The full method is on [the Method page](https://trustaudience.com/methodology).
- **Size of the read:** 1,325 followers counted, 984 of them with readable posts, and 9,081 follower posts analysed. Twelve reports read a full 100 followers. The other three accounts had fewer than 100 followers, so we read all or nearly all of them: 9, 49 and 67.
- **Exclusions:** no report on an older method was used, and none was needed, because all 15 board reports are current. No follower was Unclear (unreadable), so nobody was dropped from a score.
- **Disclosure:** one of the 15 accounts, [@JonahDevs](https://trustaudience.com/u/JonahDevs) (116 followers, score 69), belongs to the author. Leaving it out doesn't change any range reported here; the median score would move from 63 to 62.5, and the small-account median from 64 to 63.5.
- **Score:** the Audience score is the mean of each counted follower's weight: Substantive 100, Mixed 60, Silent 40, Repetitive promotion 25, Unavailable 25, Repeated reply pattern 0.

A follower who follows two of these accounts would be counted twice. We publish only aggregates and per-account scores already shown on the board, never anything about an individual follower.

## Finding 1: most Audience scores landed in the 60s

Nine of the 15 scores fell between 60 and 69. The lowest was 47 and the highest 76, with a median of 63.[1]

**Chart: Audience scores, grouped by range**

| Label | Value |
|---|---|
| 40-49 | 1 report |
| 50-59 | 3 reports |
| 60-69 | 9 reports |
| 70-79 | 2 reports |

*n = 15 current-method reports (audience-context-2026-10-03-v3), checked 2026-10-03 to 2026-10-04. Source: TrustAudience public reports.*

No account scored 80 or more, and none scored below 40. The narrow band partly follows from the weights. A follower base made mostly of Substantive and Silent followers lands between 40 and 100 by construction, and almost every account here had both in large numbers. A score does not verify audience quality. It summarises what the newest followers' recent posts looked like.

## Finding 2: about four in ten newest followers posted something substantive

Pooled across all 15 reports, here is how the 1,325 followers were labelled.[1]

**Chart: How 1,325 newest followers were labelled**

| Label | Value |
|---|---|
| Substantive | 527 followers |
| Mixed | 243 followers |
| Repetitive promotion | 86 followers |
| Repeated reply pattern | 75 followers |
| Silent | 394 followers |
| Unavailable | 0 followers |

*Pooled over 15 current-method reports (audience-context-2026-10-03-v3), checked 2026-10-03 to 2026-10-04. Source: TrustAudience public reports. Unclear: 0.*

Substantive (527) and Mixed (243) together made up more than half. The two repetitive labels, Repetitive promotion (86) and Repeated reply pattern (75), were the smallest groups apart from Unavailable. No follower in any report was suspended or deactivated. These are newest followers, so accounts that were removed earlier are unlikely to appear among them. Read that zero as a fact about this sample, not about X's enforcement.

## Finding 3: the two large accounts had far more Silent followers

Silent means no posts of their own, posts too thin to judge, or a private account. We cover the three reasons and what they mean for reach in the [silent followers guide](https://trustaudience.com/guides/silent-followers-on-x). Here is the share for each report that read a full 100.

**Chart: Share of the newest 100 followers who were Silent**

| Label | Value |
|---|---|
| @pewdiepie (658K) | 72% |
| @supabase (211K) | 54% |
| 920 followers | 15% |
| 479 followers | 24% |
| 399 followers | 31% |
| 292 followers | 33% |
| 286 followers | 23% |
| 267 followers | 11% |
| 254 followers | 32% |
| 158 followers | 20% |
| 149 followers | 14% |
| 116 followers | 12% |

*The 12 reports that read a full 100 newest followers, largest account first; each bar is out of 100 followers. The ten small accounts are shown by follower count only. Method audience-context-2026-10-03-v3, checked 2026-10-03 to 2026-10-04. Source: TrustAudience public reports.*

The ten small accounts ranged from 11 to 33 Silent out of 100. [@pewdiepie](https://trustaudience.com/u/pewdiepie) had 72 and [@supabase](https://trustaudience.com/u/supabase) 54.[1] Silent means no posts of their own, posts too thin to judge, or private; it says nothing about automation. The same gap shows in how many followers had posts we could read at all: 31 and 49 out of 100 for the two large accounts, against 69 to 94 for the small ones. Silent counts 40 in the score, so it largely explains why the large accounts scored 53 and 57, while the 13 accounts under 1,000 followers had a median of 64.

The comparison rests on just two large accounts, and the sample has no account between 1,000 and 100,000 followers. We can't tell from this data whether size, the kind of account, or how each audience was built drives the gap. It's the finding we most want to test as the dataset grows.

## Finding 4: repeated replies were scattered; follow-exchange language clustered on small accounts

**Repeated reply pattern** is the label for a follower with at least two replies that are near-duplicates or interchangeable and add little to the topic. Replying often doesn't earn it. It appeared in 13 of the 15 reports, 75 times in all. In the ten small full-sample reports it ranged from 1 to 16 of 100. The two large accounts had 1 and 4.[1] We also track the pattern as a separate indicator over readable followers, which counts it even when another label won. On that measure it showed up in 91 of 984 readable followers. For why this behaviour matters on X in 2026, see [AI reply bots on X](https://trustaudience.com/guides/ai-reply-bots-on-x).

**Follow-exchange language** means an explicit follow, follow-back or connect invitation in a follower's posts or bio. It showed up for 103 of 1,295 followers whose profiles we could check. It was present in 9 of the 10 small full-sample reports, ranging from 0 to 24 per report, with denominators of 96 to 100. Both large accounts had none (0 of 92 and 0 of 95).[1] A follow-back culture that targets small accounts would explain that. So would chance, with two large accounts.

Neither indicator says an account is automated. Repeated replies and follow requests are behaviours that people, schedulers and bots all produce.

## What this early look cannot tell you

- **It isn't random.** All 15 accounts are on the board because someone chose to check them. Thirteen have fewer than 1,000 followers, and the other two have more than 200,000.
- **It reads newest followers only.** We read up to 100 per account, in the order the provider returns them. Older followers may look quite different.
- **It's small.** Fifteen reports can show a pattern worth testing. They can't give a reliable rate. We report counts and ranges rather than precise percentages for that reason.
- **The labels are a model's judgement.** They are uncalibrated and can misread context or language. They cannot establish human identity or AI authorship.
- **It's one method on two days.** Earlier methods counted followers differently (before 3 October, Silent was weighted 25, not 40), so their reports aren't comparable and aren't included. Figures from future method versions will be reported separately.
- **It says nothing about X as a whole.** Please don't quote these numbers as "X's bot rate".

We'll update this page as more current-method reports come in. We especially need accounts between 1,000 and 100,000 followers, and more large ones. Each update will carry its own n, method version and dates.

## Check it by hand

You can reproduce the spirit of this study on any account without TrustAudience. Pick the account, open its followers list on the web and take the first 25. For each, open the profile and put it in one bucket: mostly original posts or real replies; a mix of ordinary posts and promotion; the same promotion or the same reply again and again; or nothing of its own to read. Note any follow-back requests separately. Then compare your buckets with the label mix above, and remember that 25 hand-checked followers carry even wider error than our 100. To check our numbers rather than your own, the per-account counts are public on each report page on the [board](https://trustaudience.com/).

## How to cite this study

Please cite the n, the method version and the dates with any figure:

> TrustAudience (2026). *How many X followers are bots? What we found reading the newest followers of 15 X accounts (early look).* n = 15 public reports, 1,325 newest followers, method audience-context-2026-10-03-v3, checked 3-4 October 2026. https://trustaudience.com/guides/x-newest-followers-study-2026

When this page is updated, the "Last checked" date and the n will change. If you need a fixed figure, quote the version you read and its date.

## Sources

1. TrustAudience public reports: 15 reports on method audience-context-2026-10-03-v3 (model gpt-6-luna), checked 2026-10-03 to 2026-10-04, read through the public leaderboard and profile API - [TrustAudience audience board](https://trustaudience.com/), accessed 2026-10-04

---

TrustAudience is independent. It is not affiliated with, endorsed by or sponsored by X Corp. Other tools and companies are named for comparison only and have not reviewed this guide.

Source: https://trustaudience.com/guides/x-newest-followers-study-2026