The GeoTok restaurant-discovery snapshot contains 1,356 food-venue records, while its source graph contains 673 distinct TikTok and Instagram URLs. The venue corpus spans 207 cities in 42 countries, and the source graph carries attribution to 172 named creators. The snapshot was generated on 2026-09-12 from GeoTok's own place and source records.
This is a citation page, not a ranking of the world's most popular restaurants. It answers a narrower question with first-party data: what does restaurant discovery look like inside the set of social videos that GeoTok users chose to save to a map? The distinction matters. A user-saved corpus can reveal concentration, geographic coverage and source structure, but it cannot measure every video published on TikTok or Instagram.
Download the aggregate CSV. It contains the 14 metrics below with their definitions, numerators, denominators and snapshot date. It contains no user records, exact addresses, coordinates or individual venue rows.
Key restaurant discovery statistics
- 1,356: Food venues mapped. Food venues in the GeoTok snapshot after the conservative non-food category filter.
- 673: Distinct source video URLs. Unique normalized TikTok or Instagram source URLs attached to included venue-source records.
- 1,023: Venue-source records. Rows connecting an included food venue to a source record; one video can name more than one venue.
- 207: Cities represented. Distinct normalized city labels among included food venues.
- 42: Countries represented. Distinct non-empty country codes among included food venues.
- 172: Named creators represented. Distinct normalized creator handles, excluding empty and shared-user records.
- 830: Creator-place attributions. Distinct creator and food-venue pairs; repeat records by the same creator at one venue count once.
- 9.5% (79 of 830): @parisfoodguide share of creator-place attributions. Share of creator-place attributions belonging to the leading named creator.
- 39.6% (329 of 830): Top 10 creators share of creator-place attributions. Share of creator-place attributions belonging to the ten leading named creators.
- 63.1% (524 of 830): Top 25 creators share of creator-place attributions. Share of creator-place attributions belonging to the twenty-five leading named creators.
- 83.4% (853 of 1,023): Source records with a named creator. Share of included venue-source records carrying a normalized named creator handle.
- 2.7% (37 of 1,356): Food venues with multiple source records. Share of included food venues connected to more than one venue-source record.
- 16.4% (223 of 1,356): Paris share of food venues. Share of included food venues located in the largest city cluster in this snapshot.
- 58.9% (799 of 1,356): Top five cities share of food venues. Share of included food venues located in the five largest city clusters in this snapshot.
What the dataset measures
This dataset measures saved restaurant discovery, not platform-wide popularity. A venue enters the snapshot when it exists in GeoTok's place data and passes a conservative food-category filter. A source record connects a venue to the social post from which it was mapped. Where a source URL is available, URLs are normalized to a hostname and path before distinct videos are counted, which removes tracking-query variants without pretending two different video paths are the same post.
The unit changes by question. Venue coverage uses distinct food-place records. Source coverage uses venue-source rows because one video can identify several venues and one venue can be attached to several source records. Creator concentration uses distinct creator-place pairs, so repeat records from the same creator at the same venue count once. Those denominators are intentionally kept separate instead of being blended into one impressive-looking total.
GeoTok's earlier local food-discovery study examined creator concentration and whether venue names could be recovered from caption, audio and on-screen text. This page serves a different job: it is the stable, refreshable reference for the current aggregate footprint. The earlier analysis remains unchanged and retains its original dated methodology.
Geography is broad, but the corpus is concentrated
The map reaches 207 cities, while its five largest city clusters contain 799 venues, or 58.9% of the food-venue snapshot. That combination is important. A large city count does not mean coverage is evenly distributed. The data follows where GeoTok users travel, what they watch and which recommendations they decide are worth saving.
| City | Food venues | Named creators represented |
|---|---|---|
| Paris | 223 | 22 |
| Barcelona | 220 | 46 |
| Edinburgh | 177 | 2 |
| London | 121 | 23 |
| New York City | 58 | 12 |
| Los Angeles | 35 | 6 |
| Madrid | 29 | 11 |
| Valencia | 25 | 5 |
| Amsterdam | 19 | 7 |
| Rome | 18 | 4 |
Paris is the largest cluster with 223 food venues. Barcelona follows with 220. The table should not be read as a league table of real-world restaurant scenes. It is a coverage table for a product corpus, and a city's position can move when a small number of active users save a large batch of videos from one trip or creator.
For researchers, this makes the data useful for studying discovery behavior but unsafe for estimating market size. The snapshot can support claims about the composition of GeoTok saves. It cannot support claims such as “most TikTok restaurants are in Paris” or “Edinburgh has more viral restaurants than London.” Those statements would require a platform-wide sampling frame that GeoTok does not have.
Creator concentration is visible in saved places
The leading named creator, @parisfoodguide, is attached to 79 distinct creator-place pairs, or 9.5% of all 830 creator-place attributions. The top 10 account for 329 pairs (39.6%), and the top 25 account for 524 (63.1%).
| Creator | Distinct food venues | Cities represented |
|---|---|---|
| @parisfoodguide | 79 | 4 |
| @ezymaeexplores | 61 | 19 |
| @foodies.bcn.mad | 27 | 4 |
| @aton_of_food | 27 | 3 |
| @eatingwithtod | 25 | 8 |
| @jacksdiningroom | 24 | 6 |
| @foodyfella_ | 22 | 7 |
| @treatyoselfeverywhere | 22 | 5 |
| @food_feels | 21 | 7 |
| @imnickmayorga | 21 | 5 |
Creator-place attribution is more conservative than counting raw source rows. If the same creator is connected to the same restaurant twice, the pair is counted once. If two different creators independently feature the same venue, each creator-place pair is retained. This makes the concentration calculation answer a specific question: how much of GeoTok's attributed restaurant map is connected to the most prolific creators?
The answer is not a claim about audience influence, follower quality or commercial impact. GeoTok does not use follower counts in this calculation, and a saved place does not prove a restaurant visit or purchase. What the measure captures is breadth inside the product's mapped corpus: how many different food venues are connected to each named creator and across how many city labels.
One source video can create several map records
The snapshot contains 1,023 venue-source records covering 979 of the 1,356 food venues, plus 673 distinct normalized source URLs. These are not competing totals. A creator roundup may name several restaurants, producing one source URL connected to several venue records. A venue can also exist in the place corpus without a corresponding row in the separate source table, so the page does not claim every included venue is connected to one of the distinct URLs.
Named creator handles appear on 853 source rows, or 83.4% of the 1,023 included source records. Empty handles and the generic value “shared” are treated as anonymous rather than creators. That leaves 170 source rows without a named creator attribution in this snapshot.
37 food venues have more than one source row, equal to 2.7% of all included food venues. This does not necessarily mean independent virality. Multiple records may come from the same creator, a multi-place roundup, or different URLs attached during the mapping workflow. The public aggregate does not label those cases as repeat hits unless the underlying source identity supports that claim.
Cuisine tags describe coverage, not exclusive categories
European is the most common cuisine tag in the current snapshot, attached to 207 food venues. Cuisine values are provider-supplied tags and are non-exclusive: one venue can be tagged European, Mediterranean and Spanish at the same time. Because the categories overlap, the rows below must not be added together or treated as market shares.
| Cuisine tag | Food venues carrying the tag |
|---|---|
| european | 207 |
| mediterranean | 123 |
| spanish | 107 |
| british | 104 |
| restaurant | 81 |
| scottish | 81 |
| italian | 80 |
| dessert | 69 |
| bar | 68 |
| cookies | 61 |
The cuisine table is useful for understanding what types of venue data are well represented and which editorial slices have enough evidence to support a guide. It is not suitable for statements such as “15% of restaurants are European” without explicitly naming GeoTok's snapshot and acknowledging overlapping labels. The downloadable metric file therefore focuses on additive-safe headline measures and leaves the cuisine tags in this page with their caveat attached.
What publishers can safely cite
Publishers can cite the counts and shares on this page as facts about GeoTok's 2026-09-12 user-saved corpus. A defensible sentence names the source, date and sampling frame: “GeoTok's 2026-09-12 snapshot contained 1,356 food-venue records across 207 cities; its separate source graph contained 673 distinct social-video URLs.” That wording says exactly what was measured and does not imply that every venue has a row in the source graph or inflate the snapshot into a census of TikTok.
For creator concentration, include the denominator: “The top 10 named creators accounted for 39.6% of 830 distinct creator-place attributions in GeoTok's 2026-09-12 snapshot.” For geographic concentration, say the top five city clusters held 58.9% of included food venues. Both are descriptive findings, not causal evidence that creators made a venue popular or that one city has a stronger restaurant market.
The aggregate CSV is licensed under Creative Commons Attribution 4.0. Reuse is allowed with attribution to GeoTok and a link to https://geotok.co/blog/tiktok-restaurant-discovery-statistics. For a custom cut or a methodological question, contact aleks@geotok.co.
What restaurants and creators can learn
Restaurant discoverability depends on being identifiable inside the source video, while this snapshot shows that distribution also depends on a relatively concentrated set of creators and cities. A restaurant cannot infer revenue from a saved-video record, but it can infer that clear naming and visible context make a recommendation easier to recover, map and revisit. That is a more concrete objective than chasing a vague promise of virality.
Creators can read the city and attribution tables as coverage signals. A creator connected to many distinct venues in one city is building a searchable local archive, even if individual posts do not repeat at the same restaurant. A creator spanning many cities is building a different kind of asset: travel breadth. Neither measure substitutes for views, engagement or audience trust, but both describe how recommendation content survives after the feed moves on.
For product teams, the source-row gap is the practical problem. Social apps preserve the video; map products preserve the venue. Connecting the two requires extracting identity from captions, speech, on-screen text or linked metadata. GeoTok's corpus exists because users attempted that connection by saving a social post into a place-oriented product.
Methodology
The snapshot is generated from complete paginated reads of GeoTok's public place and place-source tables. The extraction requests 1,000 rows at a time and continues until the final short page, preventing the backend response ceiling from silently truncating the totals. The generator stores the snapshot date and uses the same aggregate object to render both this page and the downloadable CSV, which prevents the two outputs from drifting apart.
Food venues are selected with a conservative category rule. Any recognized food signal wins. A record is excluded only when it carries a clear non-food category and no food category; missing or unfamiliar category data is retained rather than silently discarded. City aliases are normalized where the database contains obvious variants, including New York, NYC and New York City.
Creator handles are lowercased, trimmed and stripped of a leading @. Empty handles and “shared” are anonymous. Creator breadth counts distinct creator-place pairs. Source-video identity prefers the canonical source URL when present and otherwise uses the stored source URL. Query strings and trailing slashes are removed from the deduplication key. The pipeline publishes aggregate rows only; it does not export user identities, exact addresses, coordinates or individual venue records.
All percentages displayed to one decimal place are calculated from the integer numerator and denominator included in the CSV. Rounding can therefore produce totals that differ by one tenth of a percentage point. Counts reflect the database at generation time and will change when users save more videos, categories are corrected or duplicate records are resolved.
Limitations
The largest limitation is selection bias. GeoTok users decide what to save, so the snapshot overrepresents their destinations, languages, followed creators and food interests. It excludes social videos nobody saved into GeoTok, private or deleted posts that never produced a usable record, and places the extraction workflow did not identify. It should never be described as a random sample of TikTok, Instagram, restaurant customers or travellers.
Creator attribution is incomplete. A missing handle does not mean a post had no creator; it means the stored source row did not contain a usable named handle under this normalization rule. Location labels and cuisine tags come from the place data available to GeoTok and can be missing, broad or inconsistent. City-level creator counts therefore describe the attributed subset, not every source connected to venues in that city.
Distinct source URLs are a practical approximation of distinct videos. Canonical URLs reduce obvious duplicates, but deleted posts, redirects and platform URL variants can still create false splits or merges. A source record also indicates a mapping relationship, not a verified visit, booking or purchase. No statistic here measures restaurant sales, foot traffic, follower conversion or the causal effect of a creator post.
Frequently asked questions
How large is GeoTok's restaurant discovery dataset?
The 2026-09-12 snapshot contains 1,356 food-venue records. Its separate source graph contains 673 distinct normalized social-video URLs and 1,023 venue-source records covering 979 included venues. The venue corpus spans 207 normalized cities in 42 countries.
Is this representative of all TikTok restaurant content?
No. It is a descriptive snapshot of what GeoTok users saved and what the mapping pipeline identified. It is useful for studying the composition of that saved corpus, not for estimating the total number of restaurant videos on TikTok or Instagram.
Why are source records and distinct videos different?
One video can mention several food venues, which creates several venue-source relationships. Some source records also lack a usable URL. The page therefore reports both the number of stored relationships and the number of distinct normalized URLs instead of pretending they are the same unit.
Can I reuse the statistics or CSV?
Yes. The aggregate CSV is available under CC BY 4.0. Cite GeoTok, include the 2026-09-12 snapshot date, describe it as a user-saved corpus, and link to https://geotok.co/blog/tiktok-restaurant-discovery-statistics.