How to Spot Influencer Fraud: What to Pull and What to Ask

The media kit arrives as a PDF. A lifestyle creator, 180,000 followers, a healthy engagement rate, most of the audience in your target market, three screenshots from Instagram Insights to prove it. Every number in that PDF can be bought, and every screenshot can be edited in a minute.

That is the problem with most advice about influencer fraud. It tells you to check the engagement rate, as though the engagement rate were evidence. It is one of the easiest numbers to fake. Detection that works doesn’t look at a number. It looks for patterns that real audiences don’t produce, across data the creator can’t easily control.

What you are actually looking for

Influencer fraud comes in a handful of forms. Bought followers inflate reach. Bought likes, views and comments, usually from automated accounts, inflate engagement. Engagement pods, groups of creators who like and comment on each other’s posts on a schedule, produce activity from real accounts that has nothing to do with audience interest. And some fraud happens after the fact: analytics screenshots altered before they reach you, or past brand partnerships that never happened.

In the US, part of this is illegal. The FTC’s rule on fake reviews and testimonials, in force since October 2024, prohibits buying or selling fake indicators of social media influence, such as followers or views generated by bots or fake accounts, where the buyer knew or should have known they were fake and used them to misrepresent influence for a commercial purpose. That gives you contract language to point to. It doesn’t do the detection for you.

Pull four data sets before you look at a single post

Fraud shows up in the relationships between numbers, not in any single number. So collect the data first and read it together.

The first is follower history over time. A public tracker such as Social Blade shows follower counts over months for many accounts. You want at least a year, because the patterns that matter are shapes, not snapshots.

The second is engagement per post across the last few months: likes, comments and, for video, views. Record who commented as well as how many did, because the names are where pods give themselves away.

The third is a sample of followers. Open the follower list, scroll to recent followers, and open a few dozen profiles one by one. It is tedious and it is the most honest check you have.

The fourth is the creator’s own analytics, and here the format matters. Ask for a live screen recording from inside the app, not screenshots: audience countries and cities, age range, reach over the last 90 days, story views, follower growth. A recording that scrolls through the platform’s own screens is far harder to doctor than a still image. Where a platform offers a creator marketplace with verified first-party data, use that instead.

What fraud looks like in the data

Each pattern below has innocent explanations. A viral post, a giveaway or a shout-out from a larger account can produce any one of them. The test is whether there is a cause you can point to, and whether several patterns show up together.

The staircase. Real follower growth curves. Bought followers arrive in steps: a sharp jump of thousands in a day or two, with no post that went viral and no mention by a bigger account to explain it. Then a slow bleed downward over the following weeks as the platform removes fake accounts. A chart with several steps and a gentle decline after each one is the most reliable single signal you will find.

Engagement that doesn’t move with the content. On a real account, a strong post outperforms a weak one. On an account with bought engagement, likes sit in a narrow band regardless of what was posted. Watch the ratios as well. Video views far below what the follower count implies, or story views that are a small fraction of followers, suggest many followers who never see anything. Stories tend to reach people who follow actively, so they are a good read on how much of the audience is alive.

Comments that didn’t read the post. Generic praise, strings of emojis, “great content” under a post about a funeral. Look at timing too: a burst of comments within minutes of posting, from the same handful of accounts, post after post. That is what an engagement pod looks like from the outside. The accounts are real. The interest isn’t.

An audience in the wrong place. A creator who posts in English about Los Angeles restaurants, with a large share of followers in countries where none of that content makes sense, is worth questions. Compare the audience geography in the recording with the language of the comments. When the two disagree, one of them is wrong.

Followers who aren’t people. In your sample, count the profiles with no posts, no profile photo, a handle made of random letters and numbers, thousands of accounts followed and a handful of followers. A few are normal on any account. A sample dominated by them is not.

Numbers that change between formats. If the screenshot in the media kit and the screen recording show different figures for the same period, you have your answer about the screenshot.

For any account you want to know more about, Instagram’s “About this account” panel shows when the account was created and, for many accounts, where it is based and how often its username has changed. An account that has been renamed repeatedly, or was created recently for a creator claiming years of history, is worth a closer look.

Fraud is rarely one account

Most influencer vetting treats each creator as an island. Fraud doesn’t work that way. Follower farms sell to many buyers, and pods involve dozens of accounts by design, so the evidence is shared.

The same commenters appearing in the first minutes of posts across several creators on your shortlist is a pod, and it means those creators’ engagement numbers are inflated together. Unusual audience overlap between creators who have nothing else in common can point to a shared pool of bought followers. The same analysis you would run to avoid paying twice for the same audience doubles as a fraud check when the overlap has no explanation.

The trail also runs off the platform. Link-in-bio pages, storefronts and websites carry analytics and ad identifiers, domains share registration details, and creators managed by the same operator often share the same setup. Tracing those connections is investigation work rather than influencer marketing, the same kind of network-level digging that uncovers fake storefronts and impersonation. It is what brand protection built on network investigation is for, and it catches what an account-by-account check misses.

What to ask for before you sign

Detection before the contract is cheaper than recovery after it. Ask for these, and treat reluctance as information.

1. A live screen recording of native analytics covering audience location, age, reach and follower growth over the last 90 days.
2. An explanation for any sharp jump in follower history. A real answer names the post, the mention or the event.
3. Results from past brand work with tracked links or codes, and the brand names so you can check the partnerships were real.
4. A clause stating the creator hasn’t bought followers or engagement, with the right to audit and to withhold or recover fees if that turns out to be false. The FTC rule gives the clause a legal anchor for US campaigns.
5. Part of the fee tied to outcomes you measure yourself, through links, codes or landing pages you control.

The last point does the most work, because it makes fraud expensive for the creator instead of for you. Bought followers don’t use discount codes. A creator paid partly on tracked sales has no reason to inflate reach, and a program measured that way is where influencer marketing run on tracked outcomes separates from influencer marketing run on screenshots.

It is the same principle that runs through every paid placement: attention without a conversion path underneath it is decoration, the argument we made in the sponsored article is the top of the funnel, not the funnel. Formats where the purchase happens inside the content make that even simpler, as with e.l.f.’s shoppable Twitch streams with Amazon Ads, where the sale is measured at checkout rather than inferred from likes.

What detection tools can and can’t see

Audience-quality tools such as HypeAuditor and Modash analyze public data on an account and estimate how much of the audience looks real. Trackers such as Social Blade chart follower history. They are fast, and they are useful as a first screen across a long list of creators.

They share one limit. They see only what is public. They can’t see private analytics, can’t see a pod coordinated in a group chat, and can’t tell you whether the screenshot you were sent is genuine. So use tools to shortlist and people to verify. A score from any tool is a reason to look closer, not a verdict.

For agencies vetting many creators at once

An agency checking dozens of creators a month has an advantage an individual brand doesn’t: memory. Keep a file on every creator you assess, with the follower history, the commenter sample and the recording, and keep it after the campaign ends. Fraud networks recur. The pod you found on one client’s shortlist will show up on another’s, and the file turns a fresh investigation into a lookup.

That file is the start of what L’Oréal is building at scale, as we described in L’Oréal’s creator data backbone: one place where every creator’s spend, content and sales outcome accumulates, so each decision draws on all the ones before it. It also belongs next to a wider view of where your brand appears online, which an online presence analysis maps, so a creator’s claims about past partnerships can be checked against the public record.

Fraud survives on the assumption that nobody will look past the first number. Look at four data sets together, read them for shape rather than size, follow the trail across accounts, and pay for what you can measure. Most fraud doesn’t survive the second look.

Frequently Asked Questions

How do you detect fake followers on an influencer’s account?

Look at follower history for sharp jumps with no viral post or mention to explain them, followed by gradual declines. Then sample recent followers by hand and count profiles with no posts, no photo, random handles and lopsided follow counts. Compare story views and video views with the follower count, since inactive followers don’t watch. Several of these signals together are far more reliable than any single engagement rate.

What are engagement pods and how do you spot them?

Engagement pods are groups of accounts that agree to like and comment on each other’s posts, often within minutes of publishing, to push content up the feed. The accounts are real, but the engagement doesn’t reflect audience interest. You spot them by the pattern: the same small set of accounts commenting early on post after post, often with generic comments, and sometimes appearing across several creators on the same shortlist.

Is buying followers illegal?

In the US, the FTC’s rule on fake reviews and testimonials, effective since October 2024, prohibits buying or selling fake indicators of social media influence, such as bot-generated followers or views, when the buyer knew or should have known they were fake and used them to misrepresent influence for a commercial purpose. Platform rules also prohibit it. Other countries have their own consumer protection rules, so check where your campaign runs.

Are influencer fraud detection tools reliable?

They are useful for screening, not for final decisions. Tools such as HypeAuditor and Modash estimate audience quality from public data, and Social Blade charts follower history. None of them can see a creator’s private analytics, pods organized off-platform, or whether a screenshot has been edited. Use them to narrow a long list, then verify the shortlist with screen recordings, follower sampling and tracked results from past campaigns.

How can agencies prevent influencer fraud across many campaigns?

Keep a record of every creator assessed, including follower history, commenter samples and analytics recordings, and reuse it across clients, since fraud networks recur. Require screen recordings rather than screenshots, write anti-fraud clauses with audit and recovery rights into every contract, and tie part of each fee to outcomes measured through links or codes you control. That makes fraud costly for the creator rather than the client.