Fake Comment Detector

You have the comments. You want to know which accounts are working together.

Feed it a comment export — from a scraper, a dataset, or your own pipeline — and it returns a verdict for every account with the evidence attached. The first 1,500 comments of each run are analyzed free.

This tool never calls anyone a bot. It shows you the pattern, attaches the evidence, and states plainly what it cannot detect.

How do you detect fake or coordinated comments?

Not by reading them. A coordinated campaign is built to look ordinary one comment at a time; the pattern only appears when you line up when each account posted, how often, and how similar the wording is across accounts that should have nothing to do with each other.

Four signals, computed across the whole export:

• Timing bursts - clusters of comments arriving in a window tighter than organic conversation produces.

• Machine-regular cadence - an account posting at intervals so even the variation is near zero. People are irregular; schedulers are not.

• Duplicate phrasing across accounts - the same distinctive phrases from accounts with no other connection.

• Reply chains - accounts that consistently amplify one another.

Each finding carries its evidence: the timestamps, the intervals, the shared phrases.

What does the answer look like?

Every account gets one verdict:

• MULTIPLE SIGNALS - more than one independent pattern points the same way.

• STRONG / WEAK PATTERNS - one pattern, graded by distance from normal.

• NO SIGNALS OBSERVED - nothing anomalous in this data.

• INSUFFICIENT SIGNAL - not enough activity to judge. A real answer, not a failure: an account with two comments cannot be assessed, and this tool will not pretend otherwise.

Plus coverage: how many rows arrived, how many could be read, and what share of the comments came from flagged accounts.

What this cannot detect

• Professionally run personas. A paid poster using a real, aged account with genuine history is indistinguishable from a real customer in engagement data. Any tool claiming otherwise is not being honest with you.

• Motive. Coordination is a pattern, not a confession. Employees, fans and volunteers coordinate too.

• Anything outside the rows you supply. No views, no reach, no private signals.

Signals describe behaviour in a dataset. They do not describe people.

What you feed it

Comment rows as JSON, or the ID of a dataset already in your account — typically the direct output of a comments scraper you just ran, so nothing needs exporting or re-uploading. Field names from the common Facebook, Instagram and TikTok comment scrapers are recognized automatically. Rows that cannot be read are reported as excluded, with the reason, never silently dropped.

Leave every input empty for the free demo: a synthetic comment section with a planted coordination burst, one account the tool must refuse to judge, and one unreadable row. Nothing is scraped, nothing is charged.

What it costs

$0.66 per 1,000 unique comments analyzed, deduplicated first, platform compute included. The first 1,500 unique comments of every run are free — enough to audit a typical post end to end before paying anything. The demo is free.

For AI agents

Inputs and outputs are schema-defined. Failures are explicit: a refusal returns status: refused with a reason string and exits cleanly rather than returning an empty success. Coverage is always reported, so a thin answer is distinguishable from a confident one. There is no LLM in the analysis — identical input produces identical output, every time.

Where the comments come from

If you do not have an export yet, the companion tool Facebook Fake Engagement Check takes a post URL, runs the scraper in your own account, and audits the result in one call.