SparkToro fake follower audit alternative: what still works for X in 2026

SparkToro's free Fake Followers Audit stopped working in 2023, after Twitter ended free access to its API, and SparkToro has not brought it back. No single tool replaced it: the X checkers left in 2026 either need you to sign in with X, charge per run, or, like TrustAudience, read the recent posts of a much smaller sample of newest followers instead of profile fields.

What SparkToro's audit measured

SparkToro launched the Fake Followers Audit as a free tool on 1 October 2018.⁠^[1] It estimated what share of an account's followers fell into four buckets: spam accounts, bot accounts, propaganda accounts and inactive accounts. The last bucket matters. An account that had simply stopped using Twitter counted the same as a spam bot, because the question SparkToro cared about was reach: would this follower ever see your tweets?

The method was profile metadata. The launch post lists the data points the model used: profile image, profile URL, verified status, language, tweet language, account age, bio length, follower and following counts, days since last tweet, number of tweets, list appearances, location and display name.⁠^[1] To train it, SparkToro bought 35,000 fake followers from three vendors in July 2018 and set them against a random sample of 50,000 ordinary accounts. A follower was labelled fake if it tripped 7 to 10 or more signals, and SparkToro said testing identified about 65% of spam, bot and inactive accounts with very few false positives.⁠^[1]

By 2022 SparkToro described the system as 17 signals, and said the public tool analysed "a sample (several thousand)" of an account's followers, not all of them.⁠^[2] It was deliberately conservative. SparkToro wrote that its method likely undercounted spam, and that a model built in 2018 from purchased accounts would miss more sophisticated ones.⁠^[2]

The celebrity numbers, with their dates

The figures that made the tool famous came from full runs over every follower, not the public sample:

• @realDonaldTrump, October 2018. SparkToro analysed all 54,788,369 followers and put 61.0% in its bots, spam, inactive or propaganda buckets. The same post gave Barack Obama 40.9% and Kamala Harris 24.4%.⁠^[3]
• @ElonMusk, May 2022. All 93,452,093 followers as of 14 May 2022: 70.23% were judged unlikely to be authentic, active users who see his tweets. Restricted to the 26.9 million followers who had tweeted in the previous 90 days, the figure fell to 23.42%.⁠^[2]
• @POTUS, May 2022. Press coverage reported SparkToro's figure of 49.3% of the account's 22.3 million followers.⁠^[4]
• Twitter as a whole, May 2022. In a random sample of 44,058 accounts active in the previous 90 days, 19.42% met SparkToro's conservative definition of fake or spam.⁠^[2]

The gap between 70.23% and 23.42% for one account in one week is the most useful lesson from that era. Most of the headline number was inactivity, not automation. Any follower checker's percentage depends on what it puts in the bucket, so read the definition before the number.

Why it went away

Twitter announced in early February 2023 that free access to its API would end on 9 February.⁠^[5] SparkToro's three free tools, Fake Followers, SparkScore and Trending, all depended on Twitter. In its June 2023 retrospective SparkToro said the new owner "made it impossible for us to operate any of our free tools".⁠^[6] Two months later it described the API changes as deprecating the data its following graph relied on, and announced a rebuild on other data sources.⁠^[7]

The old address, sparktoro.com/tools/fake-followers-audit, now shows SparkToro's audience research product, with no follower audit on the page (checked 4 October 2026).⁠^[8] Listicles that still recommend SparkToro's fake follower checker are sending readers to a page that no longer does that job.

What replaced it, and what didn't

Nothing replaced all of it. The SparkToro audit was free, needed no sign-in, worked on any public handle, and came from a company that published how it trained the model. Each X option in 2026 gives up at least one of those:

• Metadata checkers behind an X sign-in. TwitterAudit, integrated into Fedica in 2023, offers a free first audit once you sign in through X.⁠^[9] FollowerAudit audits up to 5,000 followers free, twice a day, after an X sign-in, with paid plans from US$39.99 a month.⁠^[10] Both read profile and activity fields, as SparkToro did.
• Management suites. Circleboom and Fedica fold follower-quality checks into paid account tools, built for cleaning up your own followers.⁠^[11]⁠^[12]
• Developer tools. A community actor on Apify scores up to 5,000 of an account's newest followers on profile signals such as a default avatar, zero posts and a lopsided follow ratio. It is priced from US$100 per 1,000 reports and capped at 25 followers a run on Apify's free plan.⁠^[13]
• Post readers. TrustAudience (/) needs no sign-in and works on any public handle. It reads up to 10 recent posts or replies from each of up to 100 of an account's newest followers and labels what they post.

Prices, limits and what each tool measures are laid out side by side in /guides/x-follower-checkers-compared.

How TrustAudience differs from the SparkToro method

SparkToro asked whether a follower was likely to see your tweets, judged from profile fields. TrustAudience asks what your newest followers actually post, judged from the posts. Each follower is labelled Substantive (posts with real discussion or information), Mixed (a mix of ordinary and promotional posting), Repetitive promotion, Repeated reply pattern, Silent (no posts of their own lately, posts too thin to judge, or a private account), Unavailable (suspended or deactivated) or Unclear (we couldn't read them, so they are left out of the score). The Audience score is the mean of fixed weights over everyone counted; /methodology lists the weights.

It is not a drop-in replacement, and the trade-offs run both ways:

• A smaller sample. Up to 100 newest followers, against SparkToro's several thousand. It describes who followed recently, not the whole following.
• A model's judgement. A language model assigns the post-based labels. They haven't been calibrated against a hand-labelled set, so treat the result as a careful read, not a measurement.
• No identity verdict. Labels describe posting behaviour. They can't establish whether an account is a person or automated, or who wrote a post, and the score does not verify audience quality.
• No blue-check data. SparkToro used verified status as a signal. TrustAudience doesn't capture it.
• Public reports. A check publishes an aggregate report on the public board. Individual followers are never shown.

What it adds is the part profile fields can't show: whether a complete-looking follower writes their own posts, repeats the same promotion, or leaves near-identical replies under other accounts.

Who should use what

• You want one percentage for a big account, like the old audit: FollowerAudit or TwitterAudit, if you're happy to sign in with X. Expect SparkToro's blind spot for accounts with complete profiles.
• You want to remove followers from your own account: Circleboom or Fedica, both paid.
• You want raw per-follower data to analyse yourself: the Apify actor.
• You want to see what your newest followers post, without connecting an account: TrustAudience. The first 1,000 eligible checks are free, then a fresh check is US$1.
• You only need a sanity check: do it by hand, as below.

Check it by hand

You can rebuild a rough SparkToro-style audit in about twenty minutes, and add the step it never had.

1. Open the account's Followers list on X. It generally shows the most recent followers first (X doesn't document the order), which is where a recent burst of follows shows up.
2. Take the first 50. For each, note SparkToro's strongest signals: no profile image, an account only weeks old, very few followers, no post in months, no list appearances, and a display name that looks generated.⁠^[1] Count how many show several at once.
3. Open 10 to 20 of the accounts that look complete and scroll their posts and replies. Note whether each writes its own posts, repeats one promotion, leaves near-identical replies on big accounts, or posts nothing.
4. Write down the date and your tallies. Repeat the same steps in a month and compare.

Step 3 is what a 2018 metadata model couldn't do, and it is where today's followers differ most from the ones SparkToro trained on. For what to look for in those posts, see /guides/how-to-tell-if-x-followers-are-bots.