Translate your menu to English (or any language) without losing the soul

“Polpo alla griglia” becomes “grilled octopus”. Fine. But “ropa vieja” becoming “old clothes”? Less fine. Menu translation is one of the things AI is genuinely great at, provided you brief it like a human translator, not a dictionary.

When to translate (and when not to)

Translate when you have meaningful tourist or expat traffic, or when you list on delivery apps that index in multiple languages. Don’t translate if your menu’s identity depends on the original language (a tiny ramen-ya in Tokyo doesn’t need English; the experience IS the Japanese menu). When in doubt, translate once and keep it side-by-side, not as a replacement.

The two big mistakes

Literal translation. Word-for-word. The most common mistake. Produces things like “ants climbing a tree” (Mǎyǐ Shàng Shù), which is delicious and not actually ants — but a tourist sees that on a menu and orders the steak.
Over-Englishing. Stripping every regional name and replacing it with a description. “Cacio e pepe” becomes “Pasta with cheese and pepper”, and now your menu reads like a chain restaurant.

The right answer is in the middle: keep the original name + a short appetizing description in the target language.

The translation prompt

📋 Prompt — copy & paste v1

You're a bilingual restaurant menu translator. Translate the following menu
from [source language] to [target language].

Rules:
1. KEEP the original dish name in the original language. Do not translate proper
 names of dishes (e.g. "ropa vieja", "cacio e pepe", "tom kha gai" stay).
2. Below each original name, add a short descriptive line in [target language]:
 12–22 words, sensory, mentions main ingredients and preparation method.
3. For ingredients that won't be familiar (e.g. "guanciale", "labneh", "harissa"),
 add a brief 3–6 word clarification in parentheses inside the description, ONCE.
4. Do not invent ingredients. If you're unsure what's in a dish, mark it
 [[VERIFY]] and I'll fill it in.
5. Avoid clichés in the target language: no "tantalizing", "exquisite",
 "to die for", "mouth-watering".
6. Match the restaurant's voice: [casual / refined / playful / traditional].

Menu:
[paste original menu]

Works in: ChatGPT · Claude · Gemini · Claude tends to handle culinary nuance best; GPT-4 close behind

The verification step

Always run a second pass — translate each line back to the source language and read it. If something comes back wrong, that’s an item AI got confused. Common trouble spots:

Animal cuts. “Skirt steak” vs “flank” vs “hanger” vs “vacío” vs “entraña” — all related, none identical, AI conflates them.
Regional cheeses and cured meats. “Caciocavallo” is not “mozzarella” — but AI will sometimes flatten them.
Spice levels and sauces. “Pico de gallo” is not “salsa”. Don’t let AI substitute.

What to do with the result

Side-by-side layout. Original on top, description below in italic or smaller type.
One language at a time. Don’t bilingual every line in 4 languages. Keep two: the original + your primary tourist language.
Print and live with it for a week before reprinting all your menus. You’ll catch awkward lines reading them across a table.

Bonus: the same prompt for delivery apps

Delivery platforms (Uber Eats, DoorDash, Glovo) often want a short description per item. Reuse the descriptive line from the translation — just trim to 90–120 characters. One workflow, two outputs.

When to hire a human

If your restaurant’s identity is tied to its origin culture (a regional Mexican spot in Texas, a Hokkaido-specific izakaya, a Lebanese institution), spend $200–500 on a human translator for the final pass. AI gets you 90% of the way there in 10 minutes; a human translator catches the remaining nuance that matters most when authenticity is your differentiator.