Algorithmic taste in 2026: when every city's TikTok food feed looks identical

Algorithmic taste in 2026: when every city's TikTok food feed looks identical

I spent the first week of May 2026 doing something almost embarrassing on company time: I opened TikTok on 8 different burner accounts, one per city, watched 30 minutes of the food For You Page on each, and screen-recorded everything. Tokyo. Madrid. Mexico City. Bangkok. Istanbul. Seoul. Lisbon. New York. Then I lined the recordings up side by side on a single monitor.

They were the same video.

Not literally — different chefs, different languages, different storefronts. But the visual grammar was identical. Slow-motion cheese pull at 0:03. Dough slap on the counter at 0:07. Overhead pour shot of something glossy and dark at 0:11. Neon back-bar, usually pink or that very specific orange-red. A handheld push-in on the final plate. The same 4 sound-on hooks looping in 6 languages. Tokyo's tonkatsu sandwich and Mexico City's pambazo were edited to the exact same rhythm, by people who have never met, for an audience that has been trained to scroll past anything that doesn't conform to it. That, I want to argue in this post, is the tiktok algorithm food problem in 2026, and it is bigger than a content gripe — it is a quiet erasure of local visual culture happening at the recommendation layer.

This is the part of the tiktok fyp food trends story nobody is being loud enough about. The platform did not flatten taste by accident. It optimized for retention and got a global food aesthetic out the back end, the way a factory that optimizes for throughput gets identical bolts.

How the algorithm got us here, and why "engagement" is the wrong word

The mechanism is not a secret. TikTok's Creative Center publishes ad-side trend data that any marketer can pull, and the academic side has been mapping it for a couple years now. Marcus Gilroy-Ware has written about algorithmic homogeneity as a feature of attention markets, not a bug. NYU's Center for Social Media and Politics has been tracking how recommendation systems narrow the variance of what gets surfaced even as the absolute volume of content explodes. The pattern is consistent across platforms, but TikTok is the most extreme case because the FYP is the entire product.

Here is what I think people get wrong. They talk about "the algorithm rewards engagement" as if engagement were a neutral signal. It is not. The model rewards a specific shape of engagement: long average watch time on short clips, with sound on, with a high replay rate in the first 3 seconds. There are only so many edit patterns that hit all three. Cheese pull works. Dough slap works. Overhead pour works. Knife through a soft thing works. You can run that grammar through any cuisine on earth and it will outperform a 45-second static shot of a grandmother making her actual recipe.

So creators converge. Not because they are unimaginative — most of them are obsessive craftspeople. They converge because the alternative is 200 views and a flat learning curve. I have watched friends in Lisbon and Bangkok independently arrive at the same 11-second edit length, the same first-frame composition, the same caption-on-screen timing. They did not copy each other. They were both fitted to the same loss function.

The result is what I am calling algorithmic taste convergence: a measurable narrowing of the visual and rhythmic variance of food content across geographies, with the strongest convergence among creators between roughly 50k and 2M followers. Below 50k, you still see local idiosyncrasy because nobody is watching closely enough for the algorithm to discipline you yet. Above 2M, a few creators are large enough to bend the format back — think @keith_lee for the U.S., @bayashi.tiktok for Japan's home-cooking corner, @cocoa_cookie for Korean street food. The middle is where the homogenization is brutal.

The takeaway: when we say the FYP "shows us what we like," we should be saying it shows us what watched well in the training data. Those are not the same thing, and the gap between them is where local food culture is being quietly compressed.

The 6 visual tells that now show up in every city's feed

I pulled out my notebook from those 8 sessions and counted. Across roughly 240 individual food clips — 30 per city, all surfaced by the FYP without me searching anything — there were 6 visual tells that appeared in more than half of the clips in every single city.

The cheese pull, or its cuisine-appropriate cousin. In Madrid it was a tortilla half-set, the yolk doing the work cheese does in Naples. In Bangkok it was a stretched mozzarella roti, which is a 2024 invention nobody in Bangkok asked for. In Tokyo it was the obligatory raclette-over-yakiniku stunt restaurants that opened in Shibuya in the last 18 months. The point is, even in cuisines where stretchy fat is not native, creators are inventing a stretchy-fat moment because the algorithm has taught them they need one.

The dough slap. Pizza, naan, Korean hotteok, Mexican tlayuda — same hand motion, same camera angle, same audible thud. There are 4 sound effects that recur. I counted.

The overhead glossy pour. In Mexico City it is salsa macha. In Istanbul it is pekmez over kaymak. In Seoul it is gochujang base on tteokbokki. In New York it is honey on a hot honey chicken sandwich that exists because of TikTok and only because of TikTok. The lens is the same focal length, the pour is the same speed, the surface tension does the same thing.

The neon-pink or orange-red back-bar. This one is so specific I almost laughed when I saw it in 7 of 8 cities. Lisbon was the only outlier, and only because the place I happened to get served has a 1960s azulejo wall they would not paint over. Even there, the bar light is now a strip of warm-pink LED added in 2025.

The handheld push-in on the final plate. 0.5-1.0 second push, no cut. It is now so ubiquitous I think people no longer notice it. It is one of the things @bayashi.tiktok built his channel on in 2021 and the format has metastasized to every food creator under 5M followers worldwide.

The 4-second "first bite" reaction shot. Eyes close. Slight head tilt. No verbal reaction in the first 2 seconds. Cut to mouth movement. Cut to caption. The reaction shot used to be cultural — Italians did it differently from Koreans, Mexicans did it differently from Japanese. Now it is one shot, one rhythm, six languages of subtitle.

"Same hook every time. If I change the open I lose 40 percent of my first-3s retention. I'm not going to do that to my channel." — paraphrase of a 600k-follower Bangkok food creator I spoke to on background, May 2026

That paraphrase is the whole problem in one sentence. The creators who are best at this know exactly what they are giving up. They are not in denial. They are just rational actors inside a system that punishes deviation in the only currency that matters to them, which is reach.

The takeaway: the 6 tells are not a coincidence and they are not a trend. They are the surface area of a global food aesthetic that the algorithm has converged on because that surface area watches well. Watching well is now the only filter that matters, and every other filter — geography, cuisine integrity, regional style — has been downgraded behind it.

What this means for how you find a real meal in 2026

I am not going to pretend I have a clean solution. The algorithmic taste problem is structural and it is not getting fixed by a vibe shift in the comments section. But I do think there are practical things a person who eats out can do in May 2026 that they could not really do 2 years ago.

One: stop trusting the FYP as a discovery layer for food in a new city. It is now a discovery layer for content about food in that city, and those are different products. The places that have learned to make algorithmically optimal videos are over-represented. The places where the chef is 67 and does not own a tripod are under-represented by something like an order of magnitude. The food is not better at the first kind of place. Usually it is worse.

Two: pay attention to the comment sections of small creators, not big ones. The 5k-50k follower local food accounts are still the layer where the algorithm has not fully disciplined people yet. Their captions are weirder, their edits are worse, their food is better. When I am traveling I scroll past anyone who looks like they could be in any city and stop hard on accounts whose B-roll could only be the city they are in.

Three: cross-reference. If a place only shows up on the FYP and never in a local language search, in local creator captions, or in non-TikTok writing about the city, that is a signal — usually not a good one. The most overrun place I ate in this year was a Madrid spot with 4.2M cumulative TikTok views and a 3.6 Google rating from people who actually went. The next street over had a 9-table room with no TikTok presence and the best escabeche I have had this decade.

This is the reason we built GeoTok in the first place. The map shouldn't be a leaderboard of who edited the cheese pull best. It should be a record of which places real people walked into, sat down at, and came back to — with TikTok video evidence attached but not in charge. We index roughly 11,000 places across the 8 cities I sampled this week, and the ones with the longest retention in the GeoTok app are not the ones with the highest TikTok view counts. They overlap maybe 18 percent of the time, by our internal measure. The other 82 percent of the good map is invisible if you only watch the FYP.

The takeaway: in 2026 the FYP is a marketing surface, not a map. Treating it as a map is how you end up eating the same 11-second video in 8 different cities and going home thinking you had 8 different meals.

I do not think TikTok is going to undo any of this. The economics of recommendation will not let them. What I do think is that the 12-24 months ahead are going to be defined by a small but real backlash among the people who eat for a living — chefs, food writers, the kind of traveler who plans a trip around lunch — toward places that opt out of the format entirely. May 2026 is early days of that backlash. By the end of the year I would not be surprised if "no TikTok" becomes a positive signal in serious food circles, the way "no Michelin star" was for a stretch in the late 2010s.

Until then, the global food aesthetic is going to keep eating local. The least we can do is notice. — Aleks, GeoTok, May 2026