wiral
All posts

Influencer marketing on X

How to Spot Fake Engagement on X Before You Pay

Maya Rao8 min read


A founder sent me a screenshot in March with the caption "this guy is huge." It was a post with 3,400 likes and eleven replies. He wanted to pay $1,800 for one post from that account.

I asked him to open the replies and read them out loud. He came back twenty minutes later and said "ok, they're all the same four sentences."

That's most of the job. You don't need a tool or a subscription. You need about nine minutes per creator and a list of things to look at in a specific order, and I'm going to give you mine, including the part where I got it wrong and killed a deal with a creator who was completely legitimate.

How to spot fake engagement on X in nine minutes

Open the replies and check who's replying and how fast. Compare likes to replies: ratios above roughly 150:1 on an account under 100k followers deserve a second look. Then ask for the analytics panel and compare impressions to profile clicks. Bought reach almost never produces profile clicks.

Everything below is the long version of those four moves.

Start with likes divided by replies

Likes are the cheapest thing on X to buy and the easiest thing for a pod to farm. Replies cost effort. So the ratio between them tells you something that neither number tells you alone.

I don't treat this as a pass/fail gate, because the healthy range moves with account size and with niche. But here's the band I work from, built from the accounts I've personally run campaigns with rather than from any published study:

A scale of likes-per-reply ratios showing that under 60 is usually a conversational account, 60 to 150 is normal, and above 200 is worth investigating for purchased likes. Likes per reply, accounts under 100k followers 0 60 150 300+ Talky Normal Look closer Account A (illustrative): 2,200 likes, 51 replies = 43:1 Account B (illustrative): 1,900 likes, 18 replies = 106:1 Account C (illustrative): 3,400 likes, 11 replies = 309:1
The square marker is the one from that March screenshot. Circles are accounts I've actually booked.

Account C is the one the founder wanted to pay $1,800 for. Three hundred people liking for every one who typed a word is not impossible, it's just rare, and when I see it I go read the replies before I do anything else.

One caveat that matters: this test degrades badly above roughly 200k followers. Large accounts get drive-by likes from people who will never reply, so ratios of 300:1 are ordinary up there. On a 12k-follower builder account, 300:1 is a question.

What a pod actually looks like in the replies

An engagement pod is a group chat where twenty people agree to like and reply to each other's posts within a few minutes of publishing. It isn't automated and it isn't technically against X's platform manipulation rules in the way bot farms are. It's just people. Which is why it's harder to detect with software and easier to detect with your eyes.

Three signatures, and you want at least two before you conclude anything:

  1. Timing compression. Every reply lands in the first four minutes, then nothing for six hours. Real posts get a first wave and then a long tail as the algorithm keeps showing the post to new people.
  2. The same faces. Open the creator's last five posts in five tabs. If eleven of the same handles appear in the replies of all five, that's a pod, not an audience.
  3. Content with no content. "This is so true." "Great thread." "Needed this today." Replies that would work under literally any post are replies written to hit a quota.
Two timelines of reply arrival times over one hour, showing a padded post receiving fourteen replies within the first four minutes and nothing afterwards, while an organic post receives replies spread across the full hour. When the replies arrive, first 60 minutes Padded post (squares) 14 replies, all inside 4 minutes, then silence Organic post (circles) 0 min 30 min 60 min Illustrative shapes, not measured data. The pattern is what I look for, not the exact counts.
You can do this test on your phone in the checkout line. Tap the timestamps under the replies.

X shows you the relative time on each reply. Tap into a reply and you get the exact minute. That's the whole method.

Go look at the replies under a post from a genuinely large builder account like @levelsio for the opposite pattern: hundreds of replies, wildly different registers, people arguing with him, replies arriving for two days. Nobody's coordinating that.

Follower curves that step instead of climb

Bought followers arrive in batches. Real ones arrive because of posts.

So a real growth curve is lumpy in a particular way: mostly flat, then a jump on a day the person posted something that traveled, then a slightly higher flat. The jump has a cause you can go find. A bought curve is flat, then a vertical wall on a Tuesday with no post attached to it, then flat again at the new level.

Two follower growth curves over twelve weeks. The healthy curve rises unevenly with small jumps tied to individual posts, while the padded curve is flat with three sudden vertical steps that have no posts attached to them. Twelve weeks of follower count, two accounts Healthy: solid line, jumps have posts Padded: dashed line, jumps have nothing Week 1 Week 6 Week 12 Illustrative curves drawn to show shape. Markers sit on each inflection point.
The dashed line is the tell. Nobody grows in three perfectly vertical Tuesdays.

You can get a rough version of this for free on Social Blade or a similar tracker, but honestly I mostly do it by hand: scroll the creator's timeline to the week of the jump and ask what caused it. If there's a post with 400 replies sitting right there, great. If the week is three quote-posts about coffee, ask.

The metric that catches everything else: profile clicks

This is the one I'd keep if I could only keep one.

Impressions can be inflated. Likes can be bought. Replies can be farmed. Profile clicks are the number of people who saw the post, cared enough to want to know who wrote it, and tapped the name. There is no cheap market for those, because the fake accounts that farm likes have zero interest in your creator's bio.

So I ask for the post analytics panel and I compute one number: profile clicks divided by impressions.

Post (illustrative)ImpressionsLikesProfile clicksClicks / impressionsRead
A142,0001,8501,6101.13%Real attention
B96,0002,4004300.45%Fine, low-intent audience
C310,0004,100380.01%Something is wrong

Post C is the shape that should stop you. Three hundred thousand impressions and thirty-eight people tapped the name. Either the impressions aren't real, or they came from an audience with no relationship to the account, which for your purposes is the same problem.

I don't have a published benchmark to hand you here and I'm not going to invent one. What I'll say is that across the campaigns I've run, the creators who drove actual signups sat somewhere north of 0.4%, and the ones who drove nothing sat under 0.1%. Treat that as my working rule, not as science.

Link clicks are the other half of this. If a creator has run sponsored posts before, ask for the link click number on one, and compare it to their impressions on the same post. The post analytics panel on X shows both, side by side, on the same screen.

What to ask for, and how to read what they send back

Here's the message I send. Copy it.

Hey, before we get to rates: could you send screenshots of the analytics panel for your last three posts, including one sponsored one if you've run any? I'm looking at impressions, profile clicks, and link clicks. Not judging the numbers, I just calibrate my expectations off real ones instead of guessing.

That last sentence does a lot of work. It's true, and it takes the accusation out of the ask.

What the reply tells you:

  • Sends all three within a day, including a bad one. This is the answer you want. Creators with real numbers are relaxed about them. Some of my best performers have sent me their worst post unprompted.
  • Sends only the follower count and the like counts. They're answering a different question than the one you asked. Ask again, specifically, once.
  • Sends a screenshot cropped to hide profile clicks. I've had this happen twice. Both times the uncropped version came back worse.
  • Refuses on principle. Legitimate position, and some established creators take it. Then you pay for one cheap test post and measure it yourself with a UTM link instead of arguing.

Check the screenshot itself for the boring things: does the status bar time match a plausible screenshot session, is the post date visible, does the impression count on the screenshot match the impression count publicly visible under the post right now. That last one catches edited images in about four seconds.

Things that look bad and aren't

I killed a deal in 2024 with a creator whose follower count had jumped 9,000 in a week. I decided it was bought. It turned out he'd been quote-posted by a much larger account and I could have found the quote-post in ninety seconds if I'd looked. He signed with someone else. That one still annoys me.

So, the list of things that are not evidence of fraud:

A low like count with high impressions. Plenty of accounts get shown to a lot of people who don't tap anything. That's an audience quality question, not a fraud question.

Bot followers the creator didn't ask for. Every account above about 5k followers accumulates crypto spam followers. Nobody buys those, they just arrive. Judge the engagement, not the follower list.

Replies from the same handful of friends. A pod is fifteen accounts with generic praise inside four minutes. Four friends who always show up and say something specific is a community. The difference is whether the replies could have been written without reading the post.

A big single-day spike. Go find the cause first. There usually is one.

A small account. Some of the highest-converting posts I've bought came from accounts under 8,000 followers with 300 people who genuinely trust them. Reach is the last thing I look at, not the first.

The uncomfortable truth is that the biggest waste of money in creator marketing isn't fraud at all. It's paying a real creator with a real audience that has nothing to do with your product. Fraud is maybe one in fifteen of the accounts I look at. Mismatch is most of the rest, and it costs you the same amount.

If you'd rather not run this checklist forty times, that's the job we do before anyone gets into the Wiral creator network. And if you already know what you're launching, send us the brief and we'll do the vetting on our side.

FAQ

How do you check if an X account has fake followers?

Look at the growth curve rather than the follower list. Bought followers arrive in flat batches, so the curve steps vertically on days with no post to explain it. Real growth is lumpy but every jump has a cause you can scroll back and find.

Are engagement pods against X's rules?

X's platform manipulation policy targets coordinated inauthentic activity, and reciprocal engagement schemes fall under it. Enforcement in practice is inconsistent because pods are humans in a group chat rather than automation. For your purposes as a buyer, the enforcement question matters less than the fact that pod engagement doesn't convert.

What's a normal like-to-reply ratio on X?

For accounts under 100k followers, I see most healthy accounts land between 30:1 and 150:1. Above 200:1 I go read the replies. Above 200k followers the ratio naturally rises, because large accounts collect passive likes from people who'll never comment.

Can I detect fake engagement without asking the creator for anything?

Mostly, yes. Reply timing, repeat repliers across five posts, reply specificity, and the follower curve are all visible from the outside. The one thing you can't see publicly is profile clicks, and that's the metric worth asking for.

Should I use a fake follower checker tool?

They're a starting filter, not a verdict. Most of them score follower lists, which is the weakest signal, because inactive followers accumulate on legitimate accounts without anyone buying them. I'd rather spend nine minutes in the replies than trust a percentage score.

Keep reading