Guide

TikTok food discovery isn't global — it's 10 people per city

One in six viral food videos never names its restaurant: 266 of 1,518 mapped venues are unresolvable from caption, audio and on-screen text combined. And the top 10 creators account for 43.4% of everything attributed.

In short

We mapped 1,518 real venues out of short-form food video across 255 cities. Two findings. Discovery is concentrated: the top 10 of 163 creators account for 43.4% of every creator-attributed venue, the top 25 for 67.0%, and @parisfoodguide alone for 112. And discovery frequently fails outright: 266 venues (17.5%) can't be identified from caption, audio transcript and on-screen OCR combined, while 277 of the 1,438 videos with audio (19.3%) never say the restaurant's name aloud.

By AleksUpdated

Everyone talks about "the algorithm" deciding which restaurants blow up. Our data says something narrower and stranger: in any given city, a handful of named humans decide.

Update — 4 September 2026

The corpus has grown since this was first published, and we've since measured something we hadn't looked at in August: whether a video makes its restaurant identifiable at all. Current figures:

  • 1,518 venues across 255 cities, from 1,185 source videos and 163 creators; 1,055 venues carry a creator attribution.
  • Top 10 creators: 43.4% of every creator-attributed venue. Top 25: 67.0%. The single largest, @parisfoodguide, accounts for 112 venues on its own.

A note on denominators, because they differ from the August tables below: this refresh counts all mapped venues, where the original counted food venues only (1,150) and expressed concentration as a share of videos rather than venues. Both are stated as measured; neither supersedes the other. The August finding — half of all creators appear exactly once — still holds.

Finding 4 — one in six viral food videos is a dead end

We read every video three independent ways: the caption, the audio transcript, and OCR of any text on screen (menu boards, signage, receipts). Then we asked a blunt question: does the restaurant's name appear in any of them?

For 1,252 of 1,518 venues, yes. For 266 — 17.5% — no. Not in the caption, not in the audio, not on screen. One in six viral food videos shows you a dish you now want and gives no machine-recoverable way to find where it is.

SignalVideos where it existsTimes it carried the name
Audio transcript1,4381,161
Caption1,143871
On-screen text (OCR)17391

Three things fall out of that table:

  • Speech is not the reliable channel people assume. Of the 1,438 videos where we have the audio, 277 (19.3%) never say the restaurant's name aloud. Nearly one in five.
  • On-screen text is rare but decisive. OCR existed for only 173 videos, but it was the only signal that resolved 14 of them — venues that survived solely because a sign or menu board happened to be in frame.
  • More signal doesn't guarantee an answer. For the 163 venues where all three signals were present, 14 (8.6%) were still unresolvable. Even with caption, audio and on-screen text together, roughly one in twelve stays anonymous.

Of the 266 unresolvable venues, 80 had no readable signal whatsoever; the other 186 had at least one signal that simply never contained the name.

What this means if you run a restaurant: the bottleneck isn't reach, it's legibility. Your venue wins by being recoverable from footage someone else shot, on their terms, without your involvement. A visible sign, a name said once out loud, a menu header in frame — each is worth more than it sounds, because for one video in six none of them happened.

What this means if you're a creator: saying the name once, on camera, is the difference between a video that sends people somewhere and a video that only makes them hungry.

GeoTok users save TikTok and Instagram food videos to a map. That produces something no one else has — a record of which creator's video sent a real person to a real restaurant, at scale. This is what 846 of those videos look like when you count them.

Method, and what this is not

  • Corpus: 1,303 venues mapped from user-saved social video, of which 1,150 are food venues (bars, restaurants, cafés, bakeries — attractions and landmarks excluded).
  • Creator attribution: 846 of those videos carry an identifiable creator handle, spread across 153 distinct creators.
  • What this is not: a census of TikTok. It is a census of what GeoTok's users chose to save, which skews toward the cities they travel in and the creators they already follow. Sample sizes vary a lot by city, and small cities are noise — we flag them below rather than dressing them up.

Finding 1 — half of all creators appear exactly once

Of 153 creators, 76 appear a single time. The median creator accounts for 2 videos. The distribution is a hard power law:

CreatorsShare of all mapped videos
Top 19%
Top 527%
Top 1042%
Top 2569%

Ten accounts out of 153 drive nearly half the map. That is not "the algorithm surfacing the best food" — that is a small number of prolific people whose output happens to be saveable.

Finding 2 — the concentration is per city, and it's more extreme

This is the part we didn't expect. Aggregate concentration is a power law; city-level concentration is closer to a monopoly.

CityVideos mappedLeading creatorTheir share
Paris116@parisfoodguide66%
Barcelona113@foodies.bcn.mad15%
London78@aton_of_food31%
New York38@imnickmayorga32%
Los Angeles36@nate_eatz47%

Every creator column links to that creator's own map on GeoTok — the underlying per-venue evidence for their share of the city, not a summary of it.

Two-thirds of every Paris restaurant in our corpus traces back to one account. In Los Angeles it's nearly half. Barcelona is the outlier in the other direction — a genuinely competitive market with no dominant voice.

Smaller cities show even higher concentration (Leeds is 19/19 from a single creator), but at those sample sizes that reflects one enthusiastic user's saving habits, not a real market structure. We're not counting those as a finding.

Finding 3 — virality doesn't repeat

The most-shared single venue in the entire corpus is Konoba Scala Santa in Kotor, Montenegro, with 7 videos mapped to it — rated 4.4/5 across 981 reviews. After that the tail collapses immediately: Quimet & Quimet in Barcelona (4.5/5, 2,173 reviews) and a small cluster at 2 videos each, including L'Antica Pizzeria da Michele in Naples (4.1/5, 22,600 reviews) and Brigit Bakery in London (4.5/5, 4,048 reviews, known for afternoon tea).

The implication for restaurants: a TikTok hit is almost never a second TikTok hit. Nearly every venue in the corpus was mapped from exactly one video. Whatever "going viral on TikTok" does for a restaurant, it is a single event, not a channel.

What this means if you run a restaurant

You are not optimising for an algorithm. You are optimising for the five to ten people in your city who make food video consistently — and in most cities that list is short enough to write on a napkin. Barcelona is the exception where breadth matters.

What this means if you're a creator

Your city position is worth more than your follower count. @parisfoodguide is functionally the index for Paris food TikTok inside our corpus. That's a defensible position no follower count alone confers.


Data: GeoTok. Findings 1–3 collected to 17 August 2026 — 1,150 food venues and 846 creator-attributed videos, cross-verified against TripAdvisor ratings and review counts; city-level figures given only where the sample exceeds 30 videos. Finding 4 and the update block recomputed 4 September 2026 across all 1,518 mapped venues, 1,185 source videos and 163 creators. Name-matching normalises for case, diacritics and generic words (restaurant, café, bar) before testing each signal. Happy to share the underlying per-city and per-creator breakdown with anyone writing about this — aleks@geotok.co.