Buying 50,000 followers is cheap, obvious, and largely a solved problem. Any audit tool catches it: the engagement rate collapses, the follower graph has a cliff in it, and half the audience has no profile picture.
The fraud that costs brands money is the careful kind. Followers added slowly, from aged accounts with real photos and plausible bios. Engagement topped up in a pod, or bought at a ratio tuned to look normal for the niche. The account looks fine on every surface metric. It is the accounts in this middle band, not the obvious fakes, that survive a shortlist and absorb a budget.
Here is what still holds.
1. Engagement against the niche, never against a global average
The single most common analytical error is comparing a creator's engagement rate to a sitewide number. Sitewide averages are close to meaningless: engagement varies more between niches, and more between follower tiers, than it does between honest and dishonest accounts.
Published industry studies for 2026 put typical Instagram engagement somewhere near 1% for static posts and meaningfully higher for Reels, TikTok around 4% measured against views, and YouTube in the low single digits. Those numbers are useful only as an order of magnitude. They mix every niche and every account size together, which is precisely the mixing that hides fraud.
What matters is the comparison within a peer group: same platform, same niche, same follower tier, same market. A 1.8% rate can be excellent for a large finance account and poor for a small beauty one. If your tooling cannot make that comparison, the number it shows you is not evidence.
The suspicious pattern is not "low". It is flat. Real engagement is volatile: posts hit and miss, a viral one runs many multiples of the median. An account whose last thirty posts all land within a narrow band is describing a purchase, not an audience.
2. Comment quality, read by a human
This takes four minutes and outperforms most automated checks.
Open the last five posts and read the comments properly. You are looking for:
- Generic praise with no referent. "Amazing!", "So nice 😍", "Great post" repeated across unrelated posts. Nothing in the comment could only have been written about that post.
- Comments that answer nothing. A creator asks a question in the caption and the replies do not respond to it.
- The same handles on every post, in the same order, within minutes of publication. That is a pod.
- Language mismatch between the caption and the bulk of the comments, in a way that does not match a diaspora explanation.
Real audiences argue, ask logistics questions, tag friends, and complain. A comment section with no friction is a comment section that was assembled.
3. The shape of the growth curve
Follower growth over time has three recognisable shapes.
Organic compounding. Gradual, slightly accelerating, with visible steps where individual posts performed. Steps are followed by a plateau slightly above the previous level, not a return to it.
A genuine viral spike. A steep rise over days, then partial decay as the incidental followers drift, settling well above the pre-spike baseline. Crucially, engagement volume rises too and stays partly elevated.
A purchase. A steep rise with no corresponding rise in engagement volume, followed by either a flat line or a slow bleed as the platform removes accounts. The tell is the divergence: followers moved, engagement did not.
That last comparison is the useful one. Never read a follower curve without the engagement curve beside it.
4. Cross-platform consistency
A creator with 200,000 Instagram followers and 1,400 on TikTok and YouTube is not necessarily fraudulent. Plenty of people genuinely only work one platform.
But audiences that were built are usually built on one platform, because that is what was paid for. Audiences that were earned tend to leak across, even weakly. A large gap is not proof; it is a reason to look harder at the other signals.
5. The ratios inside a platform
Within Instagram, compare Reels views to follower count, and Story views to follower count where a creator will share them. Bought followers do not watch Stories. A creator with 100,000 followers and 800 Story views has an audience that is not present, whatever the follower number says.
On TikTok, compare view counts against followers. TikTok distributes to non-followers by design, so a healthy account often shows views well above follower count. The inverse — consistently far fewer views than followers — is a strong signal.
What none of this can tell you
An honest audit has to state its limits, and most published ones do not.
- Pods are not fraud in the platform's eyes and can be indistinguishable from a loyal community. A tight group of real people who reliably engage looks like exactly what you want.
- Engagement can be real and worthless. An audience that likes everything and buys nothing passes every check here.
- Follower quality is not audience fit. A completely genuine audience in the wrong country is the more common and more expensive failure. That is a different check entirely.
- A single bad number is not a verdict. Every signal here has an innocent explanation. Two or three together, in the same direction, are worth acting on. One is worth a conversation.
- Nobody outside the platform sees the real data. Every third-party estimate, including ours, is inference from public surfaces. Treat confident percentages with the suspicion they deserve.
A ten-minute check for a shortlist
Run this on the final five, not on a list of two hundred.
- Pull engagement rate and compare it to the median for that niche, platform and follower tier. Note whether it is unusually consistent, not just unusually high or low.
- Read fifty comments across five posts. Judge whether a human wrote them about that post.
- Put the follower curve and the engagement curve side by side. Look for divergence.
- Check the same creator's other platforms for a gap that the content does not explain.
- Check Reels or Story views against follower count.
- Ask the creator for a screenshot of their own analytics. What they send, and how fast, is data.
If steps 1 to 5 are clean and step 6 comes back promptly, you have done more diligence than most campaigns ever get.
Where a tool helps and where it does not
Software is good at the comparison in step 1, because it requires a peer group you cannot assemble by hand: the distribution of engagement for that niche, in that market, at that account size. It is good at plotting two curves together. It is good at doing this across two hundred creators instead of five.
It is not good at step 2. Reading a comment section is judgement, and it stays yours.
Lyren scores engagement against the niche median rather than a global average, and shows growth and engagement together for that reason. It does not claim to detect fraud, because from outside the platform nobody can. It gives you the comparison that makes the question answerable.