Epistemic Status: Full Power

Related: Surgeons Should Not Look Like Surgeons, An Economist Gets Lunch

Follow-Up: Restaurant Guide 2: Pizza

Conceptually Related More Than You Might Think: Short-Termism

I have spent a lot of time optimizing restaurant choice, and what to do when you get there. A lot of this is based on my personal preferences, but a lot also generalizes to others. A lot also generalizes to things that are not restaurants.

The basic thesis is that a restaurant will choose what signals to send based on what type of place they want to be and what they care about, so those signals are remarkably reliable. It would not benefit a place to send a misleading signal to someone using my criteria, because it would hurt them otherwise.

I can’t find his old posts on this, but Scott Adams had experience with it when he owned a restaurant. He didn’t know the rules, so he had a signal mismatch. The most stark example was when he made his place look ‘too nice.’ People would walk in, decide the place was not right for them, and leave, so he got better results by making the place superficially appear worse. A signal mismatch drives people away. Even when people don’t consciously understand what is going on, they can feel it.

The other basic principle is that the more a place is optimizing for things you don’t want, the less they are optimizing for the things you do want. Attention and focus are limiting factors, and (except perhaps on the very high end) if the place is focusing on the symbolic representation of the thing, it won’t get you the thing.

You want the thing. Restaurants tell you what they are. Believe them. The trick is to know how to read them.

This post covers finding candidate restaurants, and considerations for evaluating those restaurants that are mostly independent of cuisine. Intent is for part 2+, to the extent I do finish this series, is to cover different cuisine types, chains, how to order, how to balance exploration/exploitation, min/max considerations, group dynamics, delivery, and location recommendations in New York City.

Where to Look

There are seven procedures I know of to decide what places to consider. Each has its place.

Option 1: Walk Around

This is the default method in New York, since walking around is a free action. You can get better data from looking in person than you can any other way.

Option 2: Google Maps close up

If you zoom in close, it will show you most but not all the places in the area. You can then click on them to get more information.

Option 3: Google Maps or Yelp search

I find searching unhelpful in areas you know well, but useful while travelling if you have strong preferences on the type of food you want. Avoid this when you can.

Option 4: Zagat or similar

Zagat cares about signals I do not care about, but incentive to send those signals correlates with the food, so the numbers are useful. Michelin ratings, on the other hand, seem to convey no additional information and provide no useful search.

Option 5: Ask People

This works quite well. Even if the person is random, what they choose for you is better than the average choice they make for themselves. You are already ahead of the curve. Where you live, ask for favorite places and for types of cuisine you can’t find good options for. Where you are visiting, keep it as general as you can given your group, and ask what is good.

Option 6: Keep an Eye Out

Whenever people mention a restaurant in another context, that is a sign you should investigate. The context does not have to be praise, it only has to not be them complaining about the place.

Option 7:  Read Professional Reviews

If I find a review of a place that sounds interesting, I find it to be worth looking. The details they give you are often great. If you know what details you value, you can see if those are described. Their advice on what to order is trustworthy. Trust the details, distrust the rating.

Best Methods

I try to get as much as I can from #1 (where practical) and #6, and opportunistic #5 and #7. #2 is worth doing in your area and in any place you will stay for more than a day or two. If you know you are short on options for a trip, you should first put marginal work into #2. #3 and #4 should be kept to a minimum, as should additional #5 and #7 that require non-trivial work.

How to Use Google/Yelp Style Reviews

Google ratings are noisy but useful. If a rating is below par on 20+ reviews, unless you have a good story why, avoid. Full stop. Exceptional ratings on large reviews counts are also reliable. Yelp ratings are more noisy, but still have value, and are good mostly for negative selection.

Different areas have different average ratings for a place of similar quality, so look for the local baseline, especially in a similar price range, before making judgments. In New York City, a (20+ review) 4.4 or higher is reliably good, 3.9-4.3 is still promising, and anything at 3.8 or lower is probably bad. Also, high volume of ratings is itself an endorsement. Exceptional places often have very high rating counts.

Reading the reviews themselves is worthwhile if you are considering going.

If the place is great, a large percentage of the reviews will use words like ‘best’ and ‘amazing’ and use exclamation points. Reviews will be enthusiastic endorsements. Here is an example of what you want to see.

Look for specific praise. Watch for dishes that are mentioned as very good, which is a good sign in general and a guide on what to order.  The more detail about the food you see from positive reviews the better.

Bad reviews require careful attention. What was wrong? If staff was rude, delivery was poor and you are not ordering delivery, or other trivialities, disregard that. Overpriced can mean different things. Sometimes it is code for small portions. Sometimes it is code for ‘not fancy enough for the price point.’ Sometimes, however, it is a warning they are trying to signal a tier or two better food than they serve, and if this is clearly what the person means, that should be taken seriously. Also worry about unambiguous bad experiences. If there are a lot of unjustified bad reviews, the review average is lower than it ‘should’ be.

Bad reviews focused on the food signal trouble. Look at what they ordered. Some places do multiple things and are only good at one of them (e.g. many Italian places have great pasta or great main dishes, but not both, and if you order the wrong way you will be sad). So if all the bad reviews mention pasta, place could still be good. Just avoid the pasta.

Menu Search

Once a place has your attention, start with the menu. Find cuisine type and where it stands on the low-end to high-end scale, and look for what you might order. If this combination does not appeal to you, or there is no dish that appeals to you, there is no reason to continue.

Next, some important details.

Menu: Prices

Cents are a key story. If prices mostly end in .95 (or worse .99) this is very bad. It means they are trying to look cheaper than they are. If the prices below $20 end in .50 sometimes, that is acceptable, or .25 and .75 below $10, but less is better. If prices are bizarre, e.g. $6.38, they are round post-tax numbers and these are the resulting pre-tax numbers, and that is fine.

Different things should cost different prices, and those prices should be a bit ‘off’. If everything costs the same amount, or the differences are exactly the differences you expect, everything is generic. Look for things that are more expensive than they ‘should’ be. If  a few dishes are relatively expensive, that is a good sign for the restaurant and that those dishes are special.

Overall price level is similar. Ideal is either somewhat more, or somewhat less, expensive than you expect after controlling for location, setting/decor and cuisine. Unusually high prices means higher upside. Places at a small discount are paying attention to detail and often doing simple things well, so they too can be quite good, but be wary when prices go too low. Half-price offerings are optimizing for price first, using fast procedures and sub-par ingredients, and will likely be bad to inedible.

Menu: Size and Uniqueness

The smaller the menu, the better. Duplicate or unnecessary offerings are very bad, as is a lack of unique offerings. Every item should have a reason to exist, filling a role nothing else fills. There is some slack for things that it would be expensive to not have (e.g. the standard cuts of steak at a steakhouse) but it would still be a good sign if some of them were missing. If there is something that is conspicuously missing (e.g. a curry they don’t offer with a meat they do offer elsewhere) that is also great.

The menu should tell you what to do. It should highlight their best and most unique dishes. Having a signature dish is a big plus. The first time you go you should probably order it.

The typical ‘generic Chinese’ menu is the prototype of what to avoid. On a typical ‘generic Chinese’ menu there are 80+ items, so no one will look for something and fail, but has zero items that surprise. The majority of the 80+ items have no reason to exist, and take focus away from quality and originality. If something was special, how would you know? If your answer was ‘Chef’s Specials’ then I say that this is another terrible sign. They are always the same. You specialize in General Tso’s Chicken? What a coincidence! If the specials were genuinely different from typical specials, then that would be very good. They are not.

The flip side of the bad menu is something like this that indicates that the restaurant is going something different, but has decided it must also include the standard menu. Note the section title “American Chinese Food.” It would be better if they did not feel this compromise was necessary, but you can be confident real things are happening in that kitchen. Even if you do order off the generic menu, things will likely go well (and, in fact, the place is quite good).

The ‘great middle’ inoffensive menu, and the huge diner menu, are also places to avoid. They focus on being inoffensive, and giving everyone something they can order, which prevents the focus necessary for quality offerings.

Detailed descriptions of dishes are a plus, especially descriptions vary. Same principle. The exception is if they are telling the story that they are healthy or locally sourced or fresh. This is signaling, protests too much, and is bad. Damn hipsters.

Pictures of the food are helpful, but correlate very badly with restaurant quality, which is unfortunate. Huge negative. I wish it was otherwise. Putting the menu in plastic is slightly bad.


Once you control for other factors, not offering lunch is only a small positive. What is a strong positive is being closed for entire days of the week, or hours that make no sense. Something other than profit is being maximized.

Location and Looking Inside

The more obscure the location, the better. Side streets offer better value. Quiet signs are better than loud. Even better is almost no sign at all. If it is hard to notice a restaurant without looking closely, that is almost a guarantee of quality.

Before looking, listen. Loud music means bad food. Quiet music is good, give or take whether you want to listen to it.

Once you are looking inside, the first question for low and medium-end places is how full they are. More people is better. The lower end a place is, the more this matters; a strangely empty low-end restaurant is almost always awful. This goes double if you can contrast with a full similar place nearby. At busy times, the best low-end places are packed.

At high-end places popularity means less. Being full often means you are ‘trendy’ which is net negative. Many of my favorites are empty at the times I go.

Second, consider what types of people are inside. You want people who match the restaurant. Every group knows their own ethnic food best.

Third, look at the dishes. See if they look good, and what people actually order. One thing you can measure is the bread basket, which matters a lot to me. Good free additions are great value and indicate great other things.

See how many people have food so you know if service is slow.

Fourth, look at presentation and overall vibe. As with everything else, you want something unique. Almost anything that you don’t normally see is a plus. What matters most depends on the cuisine type. One good universal: offering higher quality cutlery than you would expect from other signals (non-disposable chopsticks, real forks/plates/cups and so forth) is a great sign. Offering disposables when you would have expected non-disposables is neutral to small positive. Treys are a remarkably bad sign.


Mentioned in
New Comment

New to LessWrong?