Auditing a creator before you pay: what fake engagement actually looks like

Fraud in creator marketing is not a fringe problem. Published 2026 audits put suspicious activity somewhere between roughly a third and over 40% of the profiles they examine, depending on methodology and sample, and a World Federation of Advertisers survey found the large majority of marketers had run into it within the previous year. The estimates vary because "fraud" covers everything from a creator who once bought 5,000 followers to industrial bot networks — but the direction is consistent enough to justify a standing check.

The good news: most of the cheap fraud is visible from public data, before any money moves.

Check 1 — the ratio triangle

Look at followers, views and engagements together. Each pair tells you something, and fraud usually breaks one relationship while faking another.

None of these is proof on its own. A creator posting genuinely passive content — satisfying loops, ambient footage — can legitimately show low comments. The point is to notice the anomaly and ask.

Check 2 — the timing curve

This is the strongest public signal and the hardest to fake convincingly.

Genuine engagement tracks distribution: as views accumulate, likes and comments accumulate roughly in proportion, then both taper. Purchased engagement arrives as a step — a few thousand likes appear within a short window, often long after the view curve has flattened, or immediately at publication before views could plausibly have supported them.

You need timestamped readings to see this. One snapshot cannot show shape. Two readings a day for the first week of a video is enough to expose most bought engagement.

Check 3 — read the comments

Two minutes of manual review catches a surprising amount:

• Generic praise with emoji, repeated across many posts by different accounts ("🔥🔥 amazing!!").
• Comments unrelated to the content of the video.
• The same handful of accounts commenting on every post within minutes.
• Comment threads with no replies from real conversation.

Bot comment quality has improved with generative text, so the absence of obvious junk is weaker evidence than it used to be. Repetition patterns across posts remain informative.

Check 4 — history, not just the highlight

Ask for the last 15 posts, not the best three. Look for:

• A sudden step change in follower count with no corresponding jump in views.
• Wildly inconsistent view counts between adjacent posts — a genuine account varies, but a 50× spread on similar content is worth a question.
• Deleted underperformers. If every visible post is a hit, the archive has been curated.

Check 5 — ask for a screen share

For meaningful budgets, ask the creator to walk you through their own analytics live: audience geography, follower growth chart, traffic sources. This is normal professional practice, takes ten minutes, and refusal is itself informative. Screenshots are not a substitute — they are trivially edited.

What to do with a red flag

Not every flag is a rejection. A proportionate response ladder:

1. 1Ask. Many creators inherited a bought-follower problem from an early agency and will say so.
2. 2Restructure the deal. Move from a flat fee to a capped per-view model so you pay for delivered reach rather than claimed audience.
3. 3Shorten the settlement window so a burst of purchased views after the fact does not increase the invoice.
4. 4Walk away when the timing curve shows bought engagement and the explanation does not hold.

Make it a standing process, not a one-off

Vetting at contract time catches the account as it is today. Accounts change hands, agencies change tactics, and a creator under pressure to hit a milestone bonus has a fresh incentive. The cheap defence is continuous: track every campaign video from publication, keep the reading history, and let anomalies surface against the creator's own baseline rather than against an industry average that describes nobody.