Technolgy & Science

The Tourist, the Local, and the Lying 4.7‑Star Average

Tourist “uplift” in restaurant ratings is a clean example of how mismatched user motivations warp data, mislead algorithms, and quietly push locals out of trusting and participating in platform feedback systems.

If you've ever been happily full in some new town, your thumb just itching to hit that five-star button, then you probably already get what this study figured out.

When I'm traveling, I tend to be generous. A little jet-lagged, not quite sure where I am, maybe a bit tipsy, I'll still tip generously and appreciate the effort a staff puts into making a place welcoming.

But locals, they don't quite see things the same way. They're probably wondering if the soup's always that salty on a Tuesday, or if the prices have gone up again. They consider if it's a spot they can count on when they're rushing between work and picking up their child. Basically, the same meal just lands differently depending on your life.

A study in the journal Information Systems Research recently put numbers to this difference. They looked at about 71,000 restaurant reviews from a big Chinese rating platform, comparing how people reviewed places when they were away from home versus when they were local. According to a summary by INFORMS, tourists were at least 13.4% more likely to give higher ratings than locals. Also, their comments were shorter, more emotional, and often included photos, unlike the locals’ more practical, detailed reviews.

When publications like Phys.org labeled this a "tourist bias," it might sound like a minor point. But then you remember that ratings do more than just express a thank you. Algorithms use them, and they control whether businesses get noticed, earn money, and sometimes, even stay open in certain neighborhoods.

So this is the thing that keeps nagging at me: what happens when a system sees all ratings as proof of one underlying "quality," but the people leaving those ratings are actually looking for totally different things?

A simple model might clarify this. Picture each restaurant with at least two real qualities: how good it is for tourists, and how good it is for locals. Let's call these T and L. Tourists, often in a holiday mindset, largely care about T—things like the atmosphere, how unique it is, or if the meal makes a good story later. Locals, though, mostly prioritize L—like consistent quality, the price point, how noisy it gets on a Tuesday, or if they can get a table without booking ahead when their train is running late.

Now, picture a platform that can't tell T from L. It only gets the noisy ratings that both tourists and locals leave. This platform just assumes every rating comes from the same hidden "quality" variable, Q. So, it simply averages everything out and ranks places. A five-star rating from a tourist and one from a local? They just get thrown into the same group.

If tourist ratings are consistently higher, which that Information Systems Research study hinted at, then the platform's idea of Q leans more toward T. The algorithm, without anyone really noticing, gets really good at showing restaurants that appeal to visitors, and not so good at highlighting the spots that quietly serve the local community.

But the system doesn't actually tell you any of this. Instead, it just presents the average like it's some kind of objective truth.

You probably see where this is going. Most platforms these days don't just show ratings; they actually build their whole system around them. They plug those numbers into ranking algorithms, recommendation engines, and even promoted ads.

Then, the whole motivation changes. If you're running a restaurant in a trendy area, you're not just focused on making the lunch rush happy anymore. Your goal shifts to satisfying the platform's idea of "Q," which, let's be honest, is totally geared towards what tourists like right now. You start redoing the inside so it photographs better. You simplify the menu. You emphasize dishes that grab you instantly, rather than those subtle flavors that take a few visits to truly appreciate. You start treating every single table like it's a potential five-star review machine, instead of a regular customer you hope to see again next week.

Now, from the model's perspective, everything looks great. People are engaging more, ratings stay high, and tourists are happy. The algorithm has basically figured out how inputs like location, food, and decor lead to outputs like stars, photos, and bookings, and it's just getting better at it.

But if you live there, the city center begins to feel like it's full of a certain kind of restaurant. Everything seems a bit themed, maybe too brightly lit, a little too ready for that perfect shot of someone eating noodles. The average rating becomes useless for figuring out where you'd actually want to eat on, say, a random Wednesday. You stop relying on it. You might still use the map feature on the app, sure, but you start hitting up coworkers for recommendations instead of scrolling through reviews.

See, here's what just happened to the data. The people who see the bias most clearly—the locals who actually depend on these spots—they quietly stop using the rating system. They write fewer reviews. The platform's view of the world ends up being even more swayed by tourists and their generally positive vibes. The exact people who could help balance things out opt to leave, because they've realized the system isn't really set up for them.

Then, the real issue isn't just "tourist bias in ratings" anymore. It’s a much bigger problem with who’s participating. If only certain people, those having a very specific kind of experience, are still talking, the platform can't really claim it represents "what people think." Instead, it just shows "what this particular group of people, feeling a certain way, decided to say publicly."

The Chinese research gives us a clear example of something we see repeatedly in the digital world. When users have different motivations, it messes up the overall data. Nobody's necessarily lying, but people are trying to solve individual problems while all contributing to the same big tally.

Tourists rate memories; locals rate tools.

When you get that, asking, "What's the average rating?" becomes obviously incomplete. Average for who? In what mood? Compared to what other choices? Still, our interfaces rarely give you any of that background. You just get a number with a glossy half-star next to it.

Now, there are some things to keep in mind. The restaurant study only looked at data from one country and one particular platform (they didn't say which one). So, we shouldn't assume every little detail applies to every city out there. Tourists might not always be the optimists, for instance. Sometimes, visitors might even have higher expectations than locals, maybe because fancy travel magazines hyped things up for them. And let's be real, locals aren't always super objective either. If you've ever seen a favorite local spot get trashed after new owners took over, you know hometown loyalty can really skew opinions, good or bad.

You could also look at tourist bias in a positive light. Sometimes, visitors pick up on things that locals, who are used to them, might overlook. That dish you ate all through childhood? It seems pretty normal until you watch an outsider's face light up with their first taste. So, in a way, those higher tourist scores might actually balance out local skepticism. A site that only paid attention to the regulars who've seen it all might end up not recommending some truly great spots.

But, the main structural issue is still there. When an algorithm tries to optimize for a measurement that blends various user motives, it’s going to consistently favor the motive that shows up most in the data. In busy city centers everywhere, that motive often boils down to "I'm visiting for a short time and I want something memorable to talk about later."

Once you realize this, you can begin advocating for improved systems rather than just better tourists. A rating platform, hypothetically, could differentiate between tourists and locals instead of lumping them together. It already has a general idea of where you reside and where you're traveling. It could display two distinct scores for each restaurant: one based on local ratings and another reflecting visitor opinions. Plus, it could reveal the breakdown by reviewer type if you hover over the stars, helping you decide how much to trust the reviews.

You can even tweak your own routine without needing any elaborate product updates. As you glance through reviews, pay attention to the tone. Does the person sound like they're a regular, eating there twice a month, or more like they're sending a postcard from vacation? Adjust how much weight you give their review based on that. And if you mainly leave reviews when you're traveling, maybe think about occasionally rating the spots you frequent at home. It could help smooth out the path others have to follow a little bit.

This doesn't make that 4.7-star rating a perfect judge of anything, of course. But it does bring home the point that every combined score tells a kind of story. It's about who actually showed up, who didn't say anything, who was on holiday, and who just needed a meal on a Tuesday.

It's not that the algorithms are inherently wrong for trusting the data they receive. The mistake lies in treating that data as though it originated from a single type of person living an identical day. The rest of us need to keep this in mind. When we gaze at those stars shining above a bustling entryway, we're not truly observing a measure of quality. Instead, what we're seeing is merely an impression of someone else's feelings, moments captured on a screen and then averaged into the decisions we make.

The Tourist, the Local, and the Lying 4.7‑Star Average

Sources

Read next

Dan Houser, AI, and the Boring Middle of the Hype Curve

Anonymous Reviews, Invisible Power