How to spot a fake TikTok restaurant review in the comments (2026 patterns)

How to spot a fake TikTok restaurant review in the comments (2026 patterns)

I have spent the last six months at GeoTok reading TikTok restaurant comment sections the way an entomologist reads ant trails, and I am going to argue something specific: fake comments cluster by syntax. The tell is not the words people use — it is the shape they take. Short bursts of three to five words, all-caps enthusiasm with no punctuation, and identical emoji patterns landing inside a 30-minute window. That signature, more than any single phrase, is what comment-pod activity looks like in May 2026.

Comment pods are not a conspiracy theory. The New York Times wrote them up in late 2025, The Verge ran a follow-up about agencies selling packages of 200 comments per video for under three hundred dollars, and a half-dozen smaller outlets have shown receipts since. What none of those pieces did was give you, a person scrolling on the train trying to decide whether the noodle place is actually good, a way to tell. So I want to fix that.

This piece is for people who use TikTok the way I do — as a restaurant filter. If a video gets three thousand comments and you cannot trust any of them, the platform stops being useful as a discovery tool. The point of GeoTok is to surface places creators actually visit, not places that bought their comment section. The same patterns I am about to describe are what we screen for on the backend.

The 30-minute burst is the giveaway

The single most reliable signal is timing. Real comments on a food video arrive in a long tail — a spike in the first hour, then a slow trickle that lasts for days as the video keeps getting served. Comment-pod activity looks nothing like that. It looks like a brick wall: 40 to 80 comments inside a window so tight you can see the seams.

I have looked at hundreds of pod-suspected videos and the pattern repeats. The first 90 seconds after publish are normal — friends, regulars, a few "where is this." Then between roughly minute four and minute thirty, you get a wave of three-to-five-word approvals: "this looks fire 🔥," "need to try asap," "adding to my list." They are not responding to each other. They are not responding to anything anyone else said. They are responding to a brief.

You can verify this yourself in about eight seconds. Tap a comment that looks suspicious. TikTok shows you that user's recent activity if their account is public. Pod accounts have a very specific footprint: 600 to 1,200 followers, follows 2,000 plus accounts, posts maybe four videos that have 200 views each, and their comment history is a parade of one-liners across food, fashion, and gym content posted in the same hour. A real diner has none of those traits. A pod account has all five.

The corollary takeaway: if you cannot click into three commenters at random and see at least one of them with normal-looking activity — a comment history that includes actual conversations, not just one-line stamps — you are looking at a planted comment section. Trust the video less than the view count suggests.

The emoji repetition fingerprint

The second signal is denser and weirder, and it took me a while to see it. Comment pods reuse emoji combinations like fingerprints. The exact same two-glyph pair — fire-plus-eyes, or 100-plus-tongue, or pleading-face-plus-fork-and-knife — will appear in eight or twelve different comments, attached to slightly different five-word templates.

The reason is mundane. Pod operators give their workers a copy block that says "use one approval phrase and one or two of these emoji." The workers comply because that is what they are getting paid for. So you end up with comment sections where the same emoji pair appears far more often than chance would predict.

I have watched videos with 1,800 comments where 11 percent of them used the exact fire-plus-drool combination. That is not how people text. People text inconsistently. Real fans of a place use the chef's-kiss emoji, or no emoji, or a heart, or whatever was on their last keyboard suggestion. The variance is the signal of authenticity, and pods strip it out by accident because they are running a script written by one person.

You will also see weirdly clean punctuation absence. Real food comments are a mess — typos, partial sentences, replies to other replies, autocorrect disasters. Pod comments have an eerie smoothness. No typos. No "lol." No "wait is this the place near—." Just complete declarative fragments. That smoothness is itself a tell once you start looking for it.

The corollary takeaway: count emoji pairs in the first 30 comments. If the same two-glyph combination appears more than three times, the video has been juiced. Real enthusiasm does not converge on a single emoji choice, because real enthusiasm is not coordinated.

The all-caps approval pattern, and what to do about it

The third pattern is the one most people already half-notice but cannot name: short, all-caps enthusiasm that does not engage with anything specific in the video. "OMG NEED THIS." "GOING THIS WEEKEND." "ADDING TO MY LIST." Capitalized, three to five words, never references the dish by name, never asks about the location, never argues with another commenter.

Compare that to how a real fan comments on @keith_lee or @foodwithsoy or any of the big food creators who built audiences before the pod economy got industrial. Real fans get specific. They name the dish. They say "the brisket sandwich is the one I always get" or "I went last March and the line was 45 minutes but worth it." Specificity is what unbought enthusiasm looks like. Generic approval is what bought enthusiasm looks like, because the buyers do not have time to brief workers on the actual menu.

The economics here are simple. You can pay people to type, but you cannot pay them to remember the brisket.

That is what someone who runs a small comment-detection tool told me last month, and it is the single best summary I have heard of why this scam is fundamentally detectable. Specificity costs money. Comment pods sell volume. The two are at war and specificity always loses when the budget is fixed.

So when you are evaluating a TikTok restaurant video in 2026, here is what I do, and what I would suggest you do. First, scroll past the top 20 comments — those are sorted by engagement and pod accounts game engagement aggressively. Go to the most-recent tab. Read 30 comments in a row. Count how many of them name a dish, reference a previous visit, ask a question, or argue with the creator. If that number is under five out of 30, something is off.

Second, look at the timestamps. If 40 comments landed inside one 30-minute window and the rest are sparse, you are looking at a pod. The platform shows you "2h ago" granularity which is enough to spot the brick wall.

Third, click three random commenters. If two of them have the pod footprint — low followers, high following, four low-view posts, scattershot comment history — discount the video entirely. It does not mean the restaurant is bad. It means the video is not evidence that the restaurant is good.

The corollary takeaway: stop trusting comment counts as a quality signal. Comment volume is the easiest metric to buy, and in May 2026 it is being bought constantly. Trust comment specificity instead — what people actually say about the food.

Why this matters for how we discover restaurants now

The reason I care about this enough to write 1,800 words on it is that TikTok displaced Yelp for restaurant discovery somewhere around 2023, and we did not get a replacement for the trust infrastructure Yelp at least pretended to provide. Yelp had fake reviews too, but Yelp also had a fraud team and a public-facing filtered-reviews shelf. TikTok has neither. The comment section is the de facto review system and it is being manipulated at scale, in public, with no countermeasures shipping from the platform itself.

That is why we built GeoTok the way we did. We index places that creators visit on camera, not places that show up in comment sections. The video itself is the evidence — you can see the dish, the room, the line, the actual human eating actual food. Comment sections are downstream of money. Footage is downstream of someone physically being there.

If you have a place you saw on TikTok and want to verify whether it is real before you spend an hour traveling to a queue, opening GeoTok and searching for the spot will tell you fast — how many separate creators have actually filmed there, when they filmed, and what the rest of the neighborhood thinks. We do not run on comment counts because comment counts are no longer information.

The deeper point is this. The TikTok food economy in May 2026 is bigger than the Yelp economy ever was, and the manipulation is more sophisticated than the platform's defenses. Until that changes, you are going to have to read comment sections like a forensic accountant — look at timing, emoji entropy, capitalization, and most of all, specificity. The fakes cluster by syntax. Once you see the shape, you cannot unsee it. And the next time you scroll past a video with 4,000 generic approvals and zero conversation, you will know exactly what you are looking at.

I am going to keep writing about this monthly because the patterns shift — pods get more sophisticated, platforms ship halfhearted countermeasures, both sides keep moving. If you spot a pattern I missed, the GeoTok team reads every email, and we add new screening rules to the backend roughly every two weeks. May 2026 is the snapshot. The arms race continues.