Engagement Authenticity Audit — Facebook Comments

Feed it the comments export from your own posts or campaign — any Facebook comments scraper's output — and get back per-account authenticity signals with evidence: timing bursts, duplicate-text clusters, coordinated reply chains, machine-regular cadence. Never a verdict. Never a guess.

Who it's for: advertisers asking "did my budget buy real engagement or a swarm?" · PR and comms teams checking whether a pile-on is organic · agencies auditing engagement quality before the client meeting.

What it detects — and the three tiers of fake

• Automated bots leak automation: machine-even timing, generated-name cohorts, mechanical cadence.
• Cheap operated fakes (real humans on fake accounts, paid per comment — the most common tier in real swarms) defeat automation signals, but their coordination leaks instead: timing bursts, near-identical talking points, cross-post repetition, mutual reply-chains.
• Professionally managed personas are indistinguishable from real people in engagement data by design — and the report says so, in a section titled "What this audit cannot detect." We state the limit; that is what makes the rest believable.

The honesty rules

• Every signal carries its evidence (the burst window, the shared text, the interval series). Levels: NO SIGNALS OBSERVED · WEAK · MULTIPLE · STRONG PATTERNS — plus INSUFFICIENT SIGNAL as a first-class answer (one comment is not enough to judge anyone, so we don't).
• The word "bot" appears nowhere as a conclusion. Signals describe behaviour in this dataset; they never characterize persons.
• Duplicates are removed before counting — and before billing.
• Absent data renders as "—". Coverage (rows in, window, dedup count, dateless rows) is printed into the report, not kept in a footnote.

What you get

A dataset (summary · per-account rows with signals+evidence · bursts · duplicate clusters) and a client-ready dark HTML report report_url in the summary) with the burst timeline, cluster examples, flagged accounts, the cannot-detect section, and the full coverage block.

Input

• Nothing → free synthetic demo: a believable comment section with a planted 40-second burst, a talking-point cluster, a generated-name cohort, a mutual-reply pair, honestly messy humans — and one single-comment account that correctly returns INSUFFICIENT SIGNAL. Watch it refuse.
• datasetId* → an Apify dataset from your comments scraper run.
• items* → comment rows pasted directly.

The field mapper states which mapping it used and refuses to analyze if too little of the format maps — a wrong analysis is worse than none.

What it deliberately does NOT do

• No scraping — it never contacts Facebook; it processes data you already have, and stores nothing beyond the run's own output.
• No "bot scores," no percentages of fake, no certified-human badges — those numbers cannot be honest from engagement data alone.
• No detection of professionally managed personas (stated above, stated in every report).

Price

The first 1,500 unique comments of every run are free — enough to audit one typical post end-to-end before paying anything.

$0.66 per 1,000 unique comments analyzed — deduplicated first, so you never pay twice for the same comment. Platform compute is included; the demo run is completely free. (A 1,000-comment post ≈ $0.66; a 30,000-comment campaign ≈ $20.)

Permissions, stated plainly

This Actor requests full permissions for one reason: to read the dataset ID you point it at. It writes only to its own run storage. It never modifies, deletes, or reads anything else in your account.

---

📖 Companion reading: The Facebook URL Zoo · How to compare Facebook pages without fooling yourself

Mentat Logic — analysis tools that state their coverage, show their working, and render the unknown as unknown.