Menu translation that does not lose the dish

I spent sixteen years in restaurant operations, a good stretch of it inside Korean-American businesses, and menu translation was a recurring argument in every one of them. Somebody always wants to make the menu easier for new customers. Somebody else points out that the regulars will not be able to find anything.

Both are right. The resolution is structural, not editorial.

Why is replacing the name so costly?

Because the name is doing three jobs and the translation only does one.

The name identifies the dish. It also carries a history the customer already knows, and it is the string they type into Google. An English approximation handles identification badly and the other two not at all.

The search consequence is concrete. Customers type dish names. If the words on your site are "spicy tofu soup" and the customer types sundubu, soondubu, or soon tofu, you are not in that result. The name your customers use is a search term, and deleting it deletes the traffic.

The phone tells you the same thing from another direction. Maple's analysis of 1.2 million calls across more than 1,000 restaurant and local business locations found menu questions make up 15% of all calls, behind reservations at 38% and hours or location at 22%. Every menu question that reaches your phone is a question your menu page did not answer, arriving during a rush, because 68% of those calls land in the lunch and dinner windows.

Cover every spelling your customers actually use. There is usually more than one romanization for the same dish, and each one is a separate query. This is the most common gap I find in Korean-American restaurants, and Korean-language search rewards getting the spellings right more than almost anything else on the site.

How do I count what I am actually missing?

What should the description actually say?

The practical things, in fifteen words or fewer.

Menu descriptions try to be evocative and end up useless. What a customer deciding between two unfamiliar dishes needs is narrow:

1. What is in it. The two or three defining ingredients, not the full recipe. 2. What it is served with. Rice, sides, bread, nothing. This surprises people more than ingredients do. 3. Whether it is shared. A dish meant for the table ordered by one person is a bad first experience. 4. Spice level, in real terms. Not icons. "Hotter than most kimchi jjigae" tells a returning customer something. "Spicy" tells nobody anything. 5. Allergens and restrictions. Pork, shellfish, dairy, nuts, gluten. This is the part with actual consequences.

Everything else is decoration. Nobody reads a paragraph while a server is standing there.

Fifteen words times ten dishes is 150 words of writing. That is one afternoon, and it is the highest-return afternoon on this list.

What do I do about dishes with no equivalent?

Describe them, do not equate them.

The failure mode is the false equivalent: calling something a pancake, a burrito, a dumpling, or a stew because those words exist in English. Each of those substitutions tells the customer something confidently wrong, and the customer finds out at the table.

Describing instead is easy and takes the same space. Say what it is made of, how it is cooked, and how it is eaten. A customer who reads three accurate facts orders with more confidence than one who read a comforting but wrong single word.

Where this breaks down is worth naming, because the rule is not universal. A dish that genuinely has an accepted English name in your market should use it. Nobody searches for the Korean word for fried chicken in San Jose. The test is not whether an English word exists, it is whether your customers use it. Ask three regulars and three first-timers the same question and you will have your answer in five minutes.

This is also a respect question, not only a clarity one. A menu that flattens every dish into the nearest American analogue is telling your core customers something about how you see their food. Getting this right is part of a larger practice, and serving a community you are not from starts with details exactly like this one.

Should the menu be bilingual on the page, or two separate menus?

One page, both languages, with the original name leading. Not two menus.

Two separate menus drift apart. The prices change on one and not the other, an item gets dropped from one, and now you have two versions of the truth. Whichever one is wrong is the one a customer will read.

On the page, the pattern that works:

Original name first, in its own script if you use one. Then the romanization on the same line. Then the short description in the second language underneath, smaller.

That layout lets a regular scan the names and skip the descriptions entirely, while a new customer reads down. Both get served without either one being slowed by the other.

One caveat, because it looks like a contradiction. Google's multilingual guidance asks site owners to help it determine a page's language "by using a single language for content and navigation on each page, and by avoiding side-by-side translations." My reading, and it is mine rather than Google's, is that an item name plus a short description is a name and its explanation rather than two translations of one paragraph. I would not extend that reading past a menu. Your service pages and your About page should be one language each. Do not build those as two columns.

Photos do more than either language. A clear photo of the plate resolves the entire question of what this is and how much of it there is, in less time than any description.

And do not lock your menu inside a PDF or an image alone. The text has to be readable as text for search to use it, which is one of the basic requirements behind how restaurants actually get found on Google.

Do keyword-stuffed image filenames help my dish photos?

No, and this is where a menu translation project usually gets padded with billable work.

The claim is that naming your photo sundubu-korean-tofu-soup-san-jose.jpg, and writing the same string into the alt attribute and the image metadata, will get you found for those terms. It is easy to sell because it is easy to demonstrate in a deliverable. It is also, by the assessment of the practitioners who do this for a living, close to worthless for local ranking.

In Whitespark's 2026 Local Search Ranking Factors survey, in which 47 local search experts scored 187 factors from zero to five, the image factors sit at the bottom of the entire list. Presence of keywords in image metadata fields scored 30. Keywords in image filenames scored 34. Keywords in image alt attributes scored 43. For scale, primary Business Profile category scored 227 and proximity to the point of search scored 225.

That survey is expert opinion rather than test data, and its author says so plainly. But when 47 practitioners independently place something at the floor of a 187-item list, that is a strong signal about where your afternoon should not go.

Write alt text anyway. Write it for a person using a screen reader, describing what is in the photo in plain language. That is what the attribute is for, it is an accessibility obligation rather than a ranking tactic, and the honest version reads better than the stuffed version in every case.

The photo itself still matters enormously. It is the description that needs no translation.

Who should write the descriptions?

Whoever cooks it, edited by whoever writes.

Ask the cook to describe the dish out loud to someone who has never had it. Record it. That gives you the real ingredients, the real serving format, and the detail nobody thought to write down. Then edit it to fifteen words.

Do not hand your menu to a translation vendor with no context. What comes back will be grammatically sound and full of false equivalents, because a translator without the dish in front of them has nothing to work from but the words.

Have a customer read the final version, not a staff member. Staff know too much to notice what is missing.

Once the menu is right, the people who serve it do the rest of the work. Servers who can describe a dish in either language sell more of it and get mentioned in reviews. If you want to encourage that, do it inside the rules, because there is a compliant way to incentivize staff around reviews and several ways that violate Google's policy.

What to do this week

Run the spelling matrix. Ten dishes, every spelling, three surfaces, and write the fraction down with today's date.

Get the missing spellings into your online menu item names and into your profile's products and services.

Rewrite the descriptions for the five dishes your servers get asked about most. Fifteen words, covering ingredients, serving format, and spice.

Photograph those five dishes. A phone is fine if the light is good.

Then check that your online menu is real text and not only an image, and that it is readable on a phone without zooming.

Events are a good place to test whether the descriptions work, because you are explaining dishes to strangers all day. Cultural events double as an acquisition channel and a focus group if you pay attention to what people ask.

Questions about your menu?

Email me at eric@seod.com with a link to your online menu and I will send back the dish names where you are losing search traffic to a translation, plus any false equivalents I spot. Korean I read myself, and our team covers Spanish.

I answer these personally. If your menu is already doing the job, I will say so rather than invent work.

Or keep reading more on multilingual and community marketing.