How to spot purchased comments

The giveaway is usually the pattern, not the comment

A comment that says “Love this 🔥” proves almost nothing.

Real people write short comments. Real fans repeat themselves. A Reel can attract a burst of strangers in minutes. A creator with an international audience can suddenly receive replies in five languages. Even a profile with no posts may belong to a real person who only watches.

Purchased comments become easier to recognize when you stop asking whether one comment looks fake and start asking whether the comment section behaves like an audience.

That means checking several things at once: what the comments say, when they arrive, who posts them, whether the same accounts recur and whether any actual conversation develops.

The important shift is from spotting weird comments to spotting coordinated engagement.

First, define what “purchased comments” means

The phrase covers several different systems.

At the cheap end, a service may use automated or disposable accounts to post templated praise. Meta has previously taken legal action against services it said distributed fake likes and comments on Instagram through automation.

Other operations use people. In 2020, Meta described coordinated campaigns in which authentic and fake accounts were used to comment under other people’s posts to make particular views look more widely supported or criticized than they really were. In one case, the activity appeared to run in regular weekday work shifts.

There is also a newer version: comments that are generated at scale but are no longer obviously generic. In Meta’s H2 2026 threat reporting, one disrupted network used AI to mass-produce contextually relevant comments for automated accounts designed to boost engagement.

So the old rule — “bad grammar means bot” — is badly outdated.

A purchased comment can be:

• automated;
• written by a low-paid human operator;
• posted by a real account participating in an engagement scheme;
• generated by AI and then posted automatically;
• customized to mention something from the actual post.

That is why the audit has to look beyond the sentence itself.

Test whether the comments are actually about the post

The fastest manual check is simple: read the comments as language, not as a metric.

Ask whether each comment contains information that anchors it to the post.

A response such as “the third setup is easily the best” could only make sense under a post containing multiple setups. “Where did you get the jacket at 0:12?” refers to a specific moment. A disagreement with a claim in the caption is even harder to reuse blindly.

Now compare that with comments such as:

• “Amazing content”
• “So good 🔥”
• “Love this”
• “Keep it up”
• “Nice post 😍”

None of those comments is fake by definition. The problem appears when most of a large comment section is interchangeable.

A useful mental test is to move the comment to a completely different post. If it would work just as well under a gym video, a restaurant photo and a crypto chart, it contains almost no evidence that the commenter actually engaged with the content.

But specificity is no longer enough on its own. AI systems can generate comments that mention objects, captions and themes correctly. Treat relevance as one signal, not a verdict.

Look at the timing

Purchased engagement often has a delivery schedule.

The visible result can be a block of comments landing within a narrow window, followed by very little activity. Another suspicious shape is a delayed batch: an older post suddenly receives dozens of low-context comments at nearly the same time.

Organic attention is usually messier. A post may spike immediately because followers receive notifications, then receive another wave if the platform recommends it, then collect long-tail comments for days. Viral posts can produce enormous bursts that are completely legitimate.

So timing only becomes useful when you compare it with context.

Ask:

• Did the post go viral at the same time?
• Was it shared by a larger account?
• Did it hit a news cycle or trend?
• Did the creator run a giveaway or ask a question?
• Do other posts show the same tight delivery pattern?

A burst with an obvious traffic event is ordinary. A burst that repeats across otherwise unremarkable posts is more interesting.

Audit the commenters as a group

One empty-looking profile is weak evidence. Twenty profiles with the same behavioral pattern are different.

Open a sample of commenters and compare them.

Useful things to notice include:

• little or no original posting history;
• extremely high following counts paired with minimal personal activity;
• feeds that appear recently created or mechanically populated;
• bios, usernames or profile images that follow the same template;
• accounts whose visible activity is dominated by generic praise;
• the same accounts commenting on several of the creator’s posts;
• the same accounts appearing under multiple unrelated creators.

Do not turn any one of these into a hard rule. Plenty of genuine users have private accounts, no posts or ugly usernames.

The question is whether the accounts look like independent people with independent behavior or like a cohort built for engagement.

This distinction matters because automated accounts can be extremely active. A 2019 ICWSM study examining roughly two million Facebook comments found that the bot accounts detected in its dataset represented only a small share of commenting users but produced comments at a sharply disproportionate rate. The lesson is not that every prolific commenter is a bot. It is that behavior across an account can reveal more than the wording of one reply.

Check whether the same crowd follows every post

Real communities absolutely have regulars. Seeing familiar names is healthy.

What becomes suspicious is a nearly fixed cast appearing under post after post regardless of subject.

Imagine an account posts:

1. a product launch;
2. a family photo;
3. a political opinion;
4. a travel Reel;
5. a meme.

If the same 30 accounts arrive rapidly under all five and leave similarly positive, low-specificity comments, that looks less like topic-driven interest and more like a standing engagement mechanism.

This can happen through purchased services, engagement pods or coordinated groups. From the outside, you may not be able to tell which.

That distinction is important.

“Coordinated” is not automatically the same as “purchased.”

If the evidence only shows the same group boosting one another, say that. Do not upgrade a pattern into a financial claim without evidence of payment.

Real conversation has friction

Purchased engagement is usually optimized to create a number.

A real comment section does more than increment the counter.

People ask follow-up questions. Someone misunderstands the post. A friend tags another friend with context. The creator answers. Two strangers disagree. A joke turns into a thread. Some comments are enthusiastic, some are boring and some are mildly negative.

That irregularity is useful.

A section containing 300 comments but almost no replies, no questions, no disagreement and no specific discussion may be visually busy while remaining socially empty.

This is not a mathematical rule. Celebrity posts can receive thousands of one-way reactions. Some creators rarely reply. Some audiences communicate almost entirely through emojis.

Again, the strength comes from the combination: generic language + coordinated timing + suspicious commenter cohort + no conversational texture is more meaningful than any one of those alone.

Watch for comments that are too consistently positive

Purchased comments are often bought for social proof.

That creates a selection problem: the customer is usually not paying for fifty strangers to say the post was boring.

A suspicious section may therefore have high volume but almost no natural variation. Every reply is praise. Every product question sounds enthusiastic. Nobody asks about shipping, pricing, limitations or something that did not work.

Real audiences are not automatically hostile, but they are usually less perfectly aligned.

This is especially important when comments are being used as evidence that customers love a product, course, investment service or business.

A comment section is an engagement surface. It is not the same thing as independently verified customer feedback.

In the United States, the FTC’s current reviews and testimonials rule guidance separately addresses fake or false reviews and testimonials, paid reviews conditioned on sentiment and fake indicators of social-media influence. The exact legal treatment depends on what the comments are doing and how they are being used commercially.

Do not rely on a magic engagement ratio

There is no universal “real comments per 1,000 likes” number.

Different platforms, niches and post formats behave differently. A meme account, a beauty tutorial, a celebrity post and a small B2B founder account can have wildly different like-to-comment ratios while all being legitimate.

Ratios are most useful within the same account.

Compare similar posts over time:

• Does a normal post receive 20 comments, then a random post receives 400 without a reach spike?
• Do comments jump while likes, shares and visible views stay flat?
• Does every post seem to receive almost exactly the same number of comments?
• Does comment volume change in neat blocks rather than with content performance?

Those mismatches justify a closer look. They do not prove a purchase.

This is the same reason creator analytics need context. A metric becomes more informative when you can compare it with the surrounding system.

Platform rules confirm the behavior exists — but not who bought it

Major platforms explicitly treat artificial engagement as a manipulation problem.

TikTok’s current Integrity and Authenticity guidelines prohibit buying or selling followers or engagement for financial gain and say the platform may remove fake or inflated signals it detects.

YouTube says it does not allow activity that artificially increases views, likes, comments or other metrics, whether through automatic systems or other illegitimate methods.

Those policies are useful because they establish that artificial engagement is not a hypothetical category.

They do not give an outside observer access to the platform’s internal detection data.

If you are auditing someone else’s account, you usually cannot see device fingerprints, IP clusters, payment records, automation infrastructure or enforcement signals. You are working from public evidence.

That means your conclusion should stay proportional to what you can actually observe.

A practical five-minute audit

For a quick manual check, do this:

1. Open five to ten recent posts rather than the single most suspicious one.
2. Read at least 30 comments and mark which ones specifically reference the post.
3. Note whether comments arrive in repeated batches or unusually uniform windows.
4. Open a sample of commenter profiles and look for recurring behavioral patterns.
5. Check whether the same accounts repeatedly appear across unrelated posts.
6. Look for actual threads, questions, disagreement and creator replies.
7. Compare comment volume with the account’s own normal range and visible reach.

Then write the conclusion in risk language.

Weak conclusion: “These comments are definitely bought.”

Stronger conclusion: “This account shows several indicators of coordinated or inauthentic commenting: repeated generic replies, the same commenter cluster across posts and batch-like timing. Public data alone does not establish who paid for it.”

That wording may sound less dramatic.

It is also more accurate.

What does not prove purchased comments

Several popular “tells” are too weak to use alone.

Signal	Why it can look suspicious	Why it is not proof
Emoji-only replies	Easy to automate or purchase	Real fans use emojis constantly
One-word praise	Generic and reusable	Real people also write “nice”
Foreign-language comments	May not match the creator’s apparent market	Social audiences are global
Empty profiles	Disposable accounts can look this way	Many real users never post
Fast comments	Engagement services can deliver in batches	Followers and viral reach can arrive instantly
Repeating commenters	Pods and paid systems reuse accounts	Real communities have regulars
Perfect grammar	Older bots often failed here	Humans and AI can both write clean copy
Specific comments	Better than generic filler	AI can now generate contextual replies

The strongest evidence is not a checklist item. It is a repeating pattern that becomes hard to explain as independent audience behavior.

The comments are not the proof; the network is

Purchased comments used to be easier to caricature: empty accounts posting “Awesome pic!” beneath everything.

That still exists.

But the more useful model in 2026 is broader. Artificial engagement can involve bots, real workers, authentic accounts, coordinated groups and AI-generated text that correctly references the content.

The sentence can look real.

The account can look real.

The comment can even be written by a real person.

What matters is whether the activity behaves like independent people choosing to participate — or like a system designed to manufacture the appearance of participation.

When you cannot establish the difference, say that too.

A comment count is evidence that comments appeared. It is not automatically evidence of demand, trust, customers or a real community.