Influencer marketing without getting scammed

BEFORE YOU START · 00 · 6 min

How to use this

One promise, one warning, and the number that started this course.

What this course is for

You are going to hand someone money in exchange for attention. This course is about checking, before the transfer, whether the attention exists.

It is not an argument against influencer marketing. Done well it is among the most efficient things you can buy. It is an argument against buying it the way most of the market still does: from a media kit, on a follower count, with no way to tell a real audience from a purchased one.

What it cannot do

It cannot certify anyone as clean. Detection is probabilistic, the people selling fake engagement improve their product every year, and no method here produces proof. What it produces is evidence strong enough to change a price, a contract or a decision — and that is the standard that matters commercially, because you are not prosecuting anyone, you are deciding whether to sign.

Anyone offering you certainty in this field is selling the same thing as the fraud: a number that feels reassuring and cannot be checked.

The shape of the four parts

Part one is what you are actually buying, and why the numbers in the deck are not the numbers you receive. Part two is detection — five ratios, a comment section, a growth curve, an audience map, and the questions that break a media kit. Part three is buying well: the contract clauses that make fraud expensive, and measuring the campaign rather than the creator. Part four is the uncomfortable one: what to do when everything is real and it still fails.

Every section ends in something you do. Your work saves in this browser, so you can close the tab and come back.

Run the scanner first

Before lesson one, take a creator you are considering, or one you already paid, and run their comment section through the scanner. Not to condemn anyone — to see what the instrument sees, and to have a baseline to compare against at the end.

PART ONE — WHAT YOU BUY · 01 · 15 min

What is sold, and what arrives

The media kit sells followers. The invoice charges for followers. Neither of them is the thing you needed.

Four different quantities, routinely confused

Almost every argument in this industry happens because two people are using one word for four things.

Money flows down that table and evaporates at every step. A follower count tells you almost nothing about the row below it, and nothing at all about the two below that. The first discipline of buying well is refusing to pay for the top row.

Why the top row is the one that gets sold

Because it is the only one that is visible from outside, stable over time, and easy to compare. A brand manager can put follower counts in a spreadsheet and sort. Nobody can sort by attention.

That convenience is precisely what created the market for fake followers. When a number decides who gets paid, that number becomes a product, and products get manufactured.

The one sentence to take into every negotiation

I am not buying your audience. I am buying a probability that some humans act, and I would like to see the evidence for that probability.

Say it once, politely, in the first call. The reaction tells you more than the media kit will. Creators who work seriously will open their analytics without being asked twice. The ones who deflect to follower counts and past brand logos have told you what they have.

What you do now

Take the last influencer invoice you approved. Write down what you were charged for, in the vocabulary of the table above. Most people find they paid for row one and reported on row two.

PART ONE — WHAT YOU BUY · 02 · 16 min

The arithmetic of reach

A creator with 300,000 followers does not have an audience of 300,000. Usually they have one of about fifteen thousand.

The number nobody puts in the deck

On most large platforms, a given post is served to a modest fraction of an account's followers. The exact fraction varies by platform, format, recency and the account's own recent performance, and no platform publishes it. What is stable is the direction: reach is a small multiple of engagement, not a large fraction of followers.

This matters commercially in one specific way. If you price on followers and report on reach, you are paying for a number that is roughly ten to thirty times larger than the one you receive. That gap is where the entire market's disappointment lives, and it exists even when nobody is committing any fraud at all.

Do the division before the meeting

The calculator below does the four sums that matter. Enter the numbers from a media kit and it converts them into cost per human who plausibly saw the thing, which is the only figure worth comparing between creators.

What the outputs mean

Engagement rate

Interactions divided by followers. Useful only as a comparison inside a tier: rates fall as accounts grow, so a 2% account with 500,000 followers may be healthier than a 6% account with 4,000. Compare like with like or the number misleads you.

Comment-to-like ratio

The most informative single number in this course. Likes are cheap to buy, cheap to automate and cheap to give. Comments cost effort, and plausible comments cost real effort. Genuine audiences produce comments in a fairly narrow band relative to likes. Purchased engagement is almost always likes-heavy, because that is what is cheap to supply.

Cost per thousand plausible viewers

Your fee divided by an estimate of humans reached, rather than by followers. This is the number to put next to your other media buys. It is frequently an unpleasant surprise, and it is the figure that turns an argument about vibes into an argument about arithmetic.

Always ask for the same three screenshots

1. Reach and impressions for the last ten posts, unfiltered — not a best-of selection.
2. Audience geography, top ten countries with percentages.
3. Follower growth for the last twelve months.

All three are two taps away in every creator tool. Refusal is not proof of anything, but it is information, and you are allowed to price it.

PART ONE — WHAT YOU BUY · 03 · 14 min

Where the fake supply comes from

Understanding who sells it, and what it costs them, tells you exactly which signals are cheap to fake and which are not.

Follow the cost, not the morality

Everything you can detect follows from one fact: different fake signals cost the supplier different amounts, and the cheap ones get oversupplied. You do not need to know how the industry works in detail. You need to know its price list, because the price list is the reason the patterns exist.

Read the table from the bottom up and you have the whole method. The signals worth trusting are the ones nobody bothers to counterfeit, because counterfeiting them costs more than the campaign is worth.

Three suppliers, three fingerprints

Bulk services

Sell followers and likes in blocks, delivered fast. Fingerprint: sharp vertical steps in the follower curve, engagement that does not rise with follower count, and comment sections dominated by short generic phrases and emoji.

Engagement pods

Groups of real creators who agree to engage with each other's posts, usually within minutes of publishing. Fingerprint: a burst of comments in a tight window right after posting, from accounts that are themselves creators in the same niche, with comments that are positive, generic and reciprocal. This is the hardest category, because the accounts are real people — which is lesson eight.

The creator themselves

Multiple accounts, friends and family, or simply buying a small top-up before a pitch. Fingerprint: a growth bump immediately before a commercial cycle, and engagement that is high on the posts a brand would look at and normal elsewhere.

Judgement, five rounds

Before the detection lessons, five decisions. Most people score two.

PART TWO — DETECTION · 04 · 17 min

The five ratios

Five numbers, computed from data anyone can see, that between them catch most of what is catchable.

Why ratios and not absolutes

Absolute numbers are what the supplier controls. Ratios are relationships between numbers, and keeping several relationships plausible at once is much more expensive than inflating any one of them. Fraud is usually detected not because a number is wrong but because two numbers stopped agreeing with each other.

1 · Engagement rate, within tier

Interactions over followers, compared only against accounts of similar size in a similar category. Both tails are informative: implausibly high suggests purchase, implausibly low for the tier suggests a follower count that outran the audience.

2 · Comments to likes

The workhorse. Genuine audiences sit in a fairly narrow band. Far below it means likes were supplied without the effort that produces comments. Far above it can mean a pod, a giveaway, or a genuinely conversational community — which is why it is a question, not a verdict.

3 · Saves and shares to likes

Requires the creator's own analytics, which is a good reason to ask for them. Saving and sharing are effortful and rarely counterfeited. A healthy account with a real audience shows meaningful save activity on useful content. Purchased engagement almost never does.

4 · Engagement stability across posts

Take the last fifteen posts and look at the spread rather than the average. Real audiences are volatile: some posts land, some do not, and the range is wide. Bought engagement is often delivered as a fixed package, which produces suspiciously flat numbers post to post. Consistency is a warning sign, not a good sign — this is the least intuitive idea in the course and one of the most reliable.

5 · Follower-to-following shape

Weak on its own and easy to game, but useful in combination. Accounts that grew by mass-following and unfollowing leave traces in the ratio and in the follower curve.

Compute all five

How to combine them without fooling yourself

Do not average the flags into a score and treat it as truth. Count them, and let the count set the action:

Note that none of the rows says "accuse". You are setting a price and a contract, not delivering a verdict, and that distinction keeps you out of arguments you cannot win and defamation you cannot afford.

PART TWO — DETECTION · 05 · 19 min

Reading a comment section

The comment section is the cheapest forensic surface you have, and almost nobody reads it properly.

What coordination looks like in text

You are not looking for bad comments. You are looking for comments that are too similar to each other, arriving too close together, from accounts that are too alike. Any one of those on its own is normal. Together they are the signature of supply rather than response.

Similarity

Generic praise dominates: fire emoji, "amazing", "love this", "so beautiful", "great content". Real audiences also say these things — the signal is the proportion. When the majority of a comment section could be pasted onto any post by any creator in any category, the comments were not about the post.

Timing

Bursts. A cluster of comments within the first few minutes of publication, followed by near silence. Real engagement decays over hours and days with a long tail. Pods and services deliver in a window.

Account similarity

Commenters with numeric suffixes, no profile photo, recently created accounts, or — in the pod case — accounts that are themselves creators of similar size in the same niche, appearing in each other's comments repeatedly.

The scanner

Copy the comments from a post, one per line, and paste them below. Nothing is uploaded; the analysis runs in your browser. Include the handles if you have them, in the form @handle: comment text, and the account patterns get checked too.

Reading the output honestly

The scanner reports four things: the share of comments that are generic, the share that are near-duplicates of another comment, the share that are emoji-only, and any patterns in the handles. It combines them into a band, and the band is deliberately coarse — unremarkable, worth asking about, hard to explain innocently.

The manual pass that beats every tool

Open the ten most recent comments and click three of the accounts. Look for: an account with no posts, an account whose own comments are the same phrase on twenty other creators, and an account created last month. Ninety seconds, and it catches things no text analysis will.

PART TWO — DETECTION · 06 · 15 min

The shape of a growth curve

Audiences accumulate. Purchases arrive. The difference is visible at a glance once you know the four shapes.

Four shapes

What each one is telling you

Organic

Compounding and noisy, with visible weeks where nothing happens. Growth accelerates gently because a larger audience produces more discovery. This is what you want to see and it is not dramatic.

Viral

A steep climb over days, then a partial fall back as the new followers who came for one thing drift away. Completely legitimate, but read the consequence: an account that grew mostly from one viral moment has an audience that never chose them. Engagement rate will be poor, and it is not fraud, it is mismatch.

Stepped

Flat, vertical jump, flat, vertical jump. Audiences do not do this. Deliveries do. Ask what happened on each date; a real answer exists for a television appearance or a collaboration, and no answer exists for a purchase.

The pre-pitch bump

A single jump shortly before a commercial cycle, then flat. The most common pattern in the small and mid tiers, because the return on a small purchase is high when a brand is about to sort a spreadsheet by follower count.

Play the shapes

How to get the curve

Ask for it. A twelve-month follower graph is two taps in every creator tool and the request is entirely normal. If you cannot get it, third-party tools reconstruct an approximation from periodic sampling — treat those as indicative only, because their sampling frequency is often too coarse to distinguish a step from a steep week.

PART TWO — DETECTION · 07 · 13 min

Audience geography and language

The fastest disqualifier in the whole course, and the one most often skipped because it feels rude to ask.

The mismatch

You are selling in Romania. The creator has 180,000 followers. Their audience is 34% Romania, 22% India, 11% Brazil and a long tail. You are about to pay for 180,000 and reach an addressable audience of roughly sixty thousand, before any of the arithmetic from lesson two.

This is not necessarily fraud. International audiences accumulate honestly, especially for visual content, English-language content, or anyone who once had a post travel. But you are paying for the whole number and using a third of it, and that is a price conversation you are entitled to have.

When it stops being innocent

Two patterns make an innocent explanation harder:

• Geography that does not match the content. A creator posting exclusively in Romanian, about Romanian places, with a plurality of followers in countries where nobody reads Romanian. Real audiences follow comprehension.
• Geography that shifted suddenly. If the audience was 80% domestic a year ago and is 40% domestic now with no change in content or language, something arrived that was not an audience.

The language check that takes one minute

Open the comments and read the languages. If a Romanian-language account has a comment section that is mostly English generic praise, the followers and the audience are different populations. This single check disqualifies more candidates than any ratio in lesson four, and it costs you sixty seconds.

The adjusted price

Recompute your cost per thousand using only the addressable share of the audience. Use the calculator with the geography percentage filled in — it does the adjustment for you, and the resulting figure is the one to bring to the negotiation.

PART TWO — DETECTION · 08 · 16 min

Pods, and the honest grey zone

The hardest category, because everyone involved is a real person doing something that is not against any law.

What a pod is

A group of creators who agree to engage with each other's posts quickly and consistently, usually coordinated in a private group. Everyone is real. Every comment is typed by a human. No money changes hands. And the effect on your campaign is close to the effect of fraud, because the engagement you paid to reach an audience was delivered by other creators who will never buy your product.

Why this is genuinely difficult

Because a supportive community and a pod look identical from outside, and because the ethical line is real but blurry. Creators help each other; that is not a crime and it is not even obviously wrong. What makes it your problem is narrower: you are paying for reach into a consumer audience and receiving reach into a professional one.

The three signals, in order of reliability

1 · Reciprocity

The strongest and most checkable. Take the ten most frequent commenters. Open their recent posts. Does the creator you are evaluating appear in their comments too? Reciprocity across several pairs is the signature. One pair is friendship; eight pairs is a structure.

2 · Timing concentration

Pods deliver fast because the platform rewards early engagement. A large share of comments inside the first ten to fifteen minutes, from the same set of accounts, week after week.

3 · Commenter composition

Ordinary audiences contain ordinary accounts. If the frequent commenters are overwhelmingly other creators of similar size in the same category, you are looking at a professional network, not a consumer audience.

What to do about it

Not accuse. Pods are not fraud and treating them as fraud will lose you a working relationship and possibly a lawyer's fee. Do three things instead:

1. Discount the engagement. Exclude frequent reciprocal commenters from your estimate of addressable audience, and price accordingly.
2. Move to outcome terms. If the audience is real but professional, a performance-linked fee costs you nothing and costs them nothing if they are right about their reach.
3. Say it plainly and kindly. "A lot of your engagement comes from other creators — which is fine, but it changes what I can pay for reach." Serious people are not offended by arithmetic.

PART TWO — DETECTION · 09 · 15 min

What a media kit hides

A media kit is a sales document. Reading it as data is the most expensive mistake in this field.

The four standard distortions

Best-of selection

The case studies are the top decile of past performance. Nothing in the document says so, and nothing in it is false. Ask for the median campaign, not the best one, and watch what happens to the numbers.

Aggregate reach across platforms

Instagram plus TikTok plus YouTube plus a newsletter, summed into one impressive figure, as though the same person following on three platforms were three people. They are largely the same person. Ask for per-platform figures.

Metrics without denominators

"2.3 million impressions" with no spend, no post count and no time frame. Any large number is available if the denominator is hidden. Every figure you accept must arrive with what it was divided by.

Screenshots without context

A cropped analytics panel with a good week visible and the date range cut off. Ask for the unfiltered last ten posts, not a selection.

The four questions

These break a media kit politely and completely. Ask them in this order, in writing.

1. What was the reach and engagement on your last ten posts, unfiltered? Not the best ten. The last ten.
2. What is your audience's top-ten country and age breakdown? Two taps, no excuse.
3. Which of your last five brand campaigns underperformed, and why? The most revealing question in the list. Every real practitioner has an answer. Anyone who claims all five succeeded is either new or not telling you the truth.
4. Would you accept a fee split between fixed and performance? You are not required to actually structure it that way. The answer tells you what they believe about their own numbers.

Build your request

The generator writes the message, in English or Romanian, with your specifics filled in. Send it before the price conversation, not after.

PART THREE — BUYING WELL · 10 · 18 min

A contract that makes fraud expensive

Detection is cheaper than litigation, but a contract is cheaper than both, because it changes the incentive before anything happens.

The principle

You will not catch everything. You do not need to. You need an agreement under which inflating the numbers stops being profitable, because payment is tied to things that cannot be bought cheaply.

This is also the kindest position to take. A creator with a real audience loses nothing under these terms. A creator without one declines, and you have learned what you needed to know without accusing anyone of anything.

Six clauses

1 · Access to analytics

The creator provides unfiltered screenshots or exports of reach, impressions, saves and audience breakdown for the campaign posts, within a stated number of days. Not a summary they typed.

2 · Payment split

A fixed portion covering production, and a variable portion tied to a delivered metric. Even a modest split changes the conversation, because it prices confidence.

3 · The metric is not followers

Tie the variable portion to something the creator cannot buy for less than you are paying: verified reach, saves, click-throughs on a tracked link, or code redemptions. Codes and tracked links are the cleanest, because they are counted by you rather than reported by them.

4 · No purchased engagement, defined

A plain-language warranty that the creator has not purchased followers, likes, views or comments for the account within a stated period, and will not do so for the campaign. Define the term rather than gesturing at it, and state the consequence: a refund of the variable portion, not a lawsuit.

5 · Content stays up

A minimum period before the post can be deleted or archived, with a remedy if it comes down early. Deletion after payment is more common than fraud and almost nobody contracts for it.

6 · Timing and disclosure

Publication window, and the advertising disclosure required in your market. This is a legal obligation for you as much as for them, and undisclosed advertising is a regulatory problem with your name on it, not theirs.

Generate your clauses

PART THREE — BUYING WELL · 11 · 16 min

Measure the campaign, not the creator

Everything so far was about choosing. This is about knowing, afterwards, whether it worked — which almost nobody does honestly.

The reporting trap

The standard influencer report says: reach, impressions, engagements, engagement rate, and a screenshot. Every one of those is a number the creator supplied about their own performance. None of them tells you whether anything happened that would not have happened anyway.

That last clause is the whole of measurement. The question is never "how many people saw it" but "what changed because they did".

Three levels, in increasing order of honesty

Level one — delivered

Did the agreed thing get published, on time, with disclosure, and stay up? Trivial, and routinely not checked.

Level two — counted by you

A unique tracked link and a unique discount code per creator. Counted on your side, not theirs. This single change does more for influencer measurement than every detection method in part two, because it replaces a reported number with an observed one.

Level three — incremental

Did sales rise beyond what the trend predicted? Requires holding something back: a region you do not activate, a period you leave clear, or a matched set of creators you run against a control. Harder, more expensive, and the only version that answers the question you actually asked.

The log

One row per creator, per campaign. Kept over several campaigns it tells you which creators actually move anything — which is a different list from the one with the best engagement rates, and that difference is the most valuable thing you will own.

What to do with the pattern

After three or four campaigns you will typically find that a minority of creators produced most of the measurable outcome, and that this minority was not predictable from follower count or engagement rate. Pay them more, work with them repeatedly, and stop treating creator selection as a fresh spreadsheet exercise every quarter.

PART THREE — BUYING WELL · 12 · 14 min

When everything is real and it still fails

The most common reason an influencer campaign underperforms has nothing to do with fraud.

Four honest failure modes

Fit

The audience is real, engaged and completely uninterested in your category. A beauty creator with an exceptional audience will not sell your accounting software, and no amount of reach fixes a mismatch. This is the largest single cause of disappointment and it is entirely your responsibility, not theirs.

Creative control

You wrote the script. It reads like advertising, the audience recognises it instantly, and the thing you paid for — a trusted voice — was the first thing your brief removed. The creators worth buying are the ones whose audience trusts their judgement, and that trust does not survive being scripted.

Timing and frequency

One post, once, against a purchase cycle measured in months. Influence accumulates; single activations mostly do not. Three posts from one credible creator over a quarter will usually beat one post each from three.

The offer

The audience saw it, liked it, clicked, and then met a landing page that did not deliver what the post promised. Attribution will blame the creator. The creator did their job.

The pre-mortem

Before signing, write down the answer to: "This campaign failed. Why?" Then check which of the four causes above your answer belongs to, and whether any of the work you have done addresses it. Most briefs, examined this way, turn out to have controlled hard for fraud and not at all for fit.

Where this ends

You now have five ratios, a comment scanner, four growth shapes, a geography check, a reciprocity test, four questions, six contract clauses and a measurement log. That is a complete apparatus for not being defrauded, and it is genuinely most of the job.

The rest is choosing people whose audience wants what you sell, and then letting them say it in their own words. No instrument helps with that.

PART THREE — BUYING WELL · ★ · 5 min

You are done

What you built, and the three instruments worth coming back to.

What you have

• Four quantities you can tell apart, and a rule against paying for the top one
• A calculator that turns a media kit into a cost per addressable human
• Five ratios and a flag count that maps to an action rather than an accusation
• A comment scanner, and the manual pass that beats it
• Four growth shapes and the question to ask about a vertical step
• A reciprocity test for pods that takes five minutes and no tooling
• Four questions that break a media kit politely
• Six contract clauses that make inflation unprofitable
• A measurement log that will eventually tell you who actually works

Come back for these

• The scanner and the calculator — every shortlist, before every negotiation.
• The clause generator — every contract, until your template has absorbed it.
• The campaign log — after every campaign. It is worthless for one and decisive after five.

Everything you entered is stored in this browser only. Nothing was uploaded anywhere.

One last scan

Run the same comment section you scanned in the first section. You will read the output differently now, and that difference is the course.

A closing caution

Everything here is a tool for pricing and contracting, not for public accusation. A pattern is not proof, detection is probabilistic, and calling someone a fraud in writing on the strength of a ratio is a legal exposure you do not need. Negotiate quietly, contract carefully, and let the terms do the work.