AI discoverability score on a menu item

When you write a description for a dish in the menu item editor, a small card below the long-description field shows a number from 0 to 100 and a coloured dot — red, yellow, or green. The number is the AI discoverability score, and it tells you, in real time, how likely AI search engines are to recommend this dish when a diner asks one. The number ticks up as you type. The score is for the chef, the owner, or the marketing manager refreshing the menu — anyone whose words on the dish page decide whether a stranger finds you.

Why this page exists

AI search engines do not read the kitchen’s mind. They do not taste your dish, ask your nonna for the recipe, or know what you meant. They rank what is written on the page. A menu row that says only Spaghetti carbonara — 380 baht gives the AI nothing to work with, so when a customer asks “where can I find handmade pasta in Bangkok?” the AI lists three other restaurants and skips you.

A dish written out properly — sixty to eighty words, two or three named ingredients, a sensory word, a note about where the recipe comes from, an allergen line — gives the AI everything. When the same customer asks the same question, that dish surfaces with your restaurant attached. The same kitchen, the same dish, dramatically different visibility. The only thing that changed was the writing.

The score makes that gap visible. Before, there was no way to know whether a description was “enough”. The chef wrote what felt right, the menu went live, and the dish lived or died in AI search results nobody could see. Now, the score is a live coach: write a line, watch the number; add an ingredient, watch the number; mention the allergens, watch the number turn green. It is a feedback loop on a job that previously had none.

The rule

AI ranks dishes by the words on the page, not by what you meant. The score tells you what you’re giving the AI to work with.

How the score works

The score is built out of six things the AI looks for. Each one contributes a portion; the total adds up to 100.

Length. The first signal is how much you wrote at all. Below twenty words, the description scores zero on length — there’s just not enough substance for the AI to anchor on. Between twenty and eighty words, the length contribution scales up smoothly. At eighty words or more, you’ve filled the bucket. Sixty words is roughly the sweet spot: enough to say something, short enough to read.

Ingredients. The AI looks for named ingredients — guanciale, Pecorino Romano, San Marzano tomato, octopus, truffle. The score counts up to four distinct ingredient mentions and rewards each one. A dish with zero ingredients named is invisible on ingredient queries (“where can I find truffle pasta near Asok?”); a dish with four is everywhere.

Sensory detail. Words that describe texture, flavour, or technique — creamy, crispy, silky, charred, bright, velvety, al dente. These are what the AI quotes back when a diner asks what a dish is like. Four sensory words fills the bucket.

Origin and method. Where the dish comes from and how it’s made — Sicilian, Roman, traditional, hand-rolled, slow-braised, wood-fired, aged. Provenance language is what separates “spaghetti” from “spaghetti with a story”. Three of these fills the bucket.

Dietary. If the dish fits a diet, say so. Vegetarian, vegan, gluten-free, halal, kosher. The AI uses dietary fitness as a hard filter — a vegan diner asking for vegan food will simply not see dishes that aren’t tagged. Two dietary keywords fills the bucket.

Allergens. A Contains: gluten, eggs, dairy line, or named allergens written into the description (nuts, shellfish, egg, soy, sesame). Diners ask the AI about allergies constantly; pages that pre-answer the question rank highest. Two allergen mentions fills the bucket.

Add them all up: a maximum of 100, broken into the six tiles you see below the score circle. Each tile shows your current points against its maximum, a little progress bar, and a one-line nudge telling you what to add to lift it.

The total lands you in one of three colours. Red sits between 0 and 29 — the card labels this Too thin for AI. Yellow sits between 30 and 69 — Could be richer. Green sits at 70 or above — AI-ready. The colour band is advisory, not a gate. Nothing stops you publishing a red dish; the score is a coach, not a guard. But green is where the dish actually gets recommended.

How to use it

Open the menu item you want to work on and scroll to the Long description field. The AI discoverability score card sits immediately below it. The card has three pieces: a large number circle on the left showing your current 0-100, a one-line label and sub-line in the middle telling you where you stand, and a grid of six tiles below covering each of the score’s ingredients.

Write into the long-description field. The score recomputes as you type — there is no save button, no waiting. Each tile beneath shows its current points and a small nudge such as “Mention a couple more ingredients — sauces, cheeses, herbs” or “Disclose major allergens — diners ask, AI ranks pages that answer first.” Follow the nudge that catches your eye; the tile responds.

The fastest path to green is usually: write a full paragraph (length), name three or four ingredients (ingredients), drop in two or three texture or flavour words (sensory), add a provenance cue (origin and method), tag the diet if relevant (dietary), close with a “Contains: …” line (allergens). Sixty to eighty words covers all of it without forcing.

The score reads the English version of your description. If you also write the dish in Thai or Italian, those translations don’t affect the score — they’re for the diner who reads in that language. The English description is what the AI search engines crawl, so that’s the one the score grades.

Writing from scratch with AI

Each description card — Short description and Full description — carries a small ✨ Write with AI button at the top right. Click it and the system generates a draft for that field in a preview modal: the short version is one or two lines, the full version is a paragraph followed by a bulleted ingredient list and a closing line that names the allergens. The draft is editable in the modal — the full version renders with paragraphs, bullets, and bold formatting just as it will on the public page, so what you see in the preview is what lands in the field.

You have three actions in the modal:

Use this → accepts the draft, fills the English slot of the field, closes the modal. The change isn’t saved yet — the editor’s main Save Changes button on the page commits it.
🔄 Regenerate fires a fresh draft. Use this when you don’t like the angle.
The Steer the rewrite field lets you point the regenerate in a direction: “make it more playful”, “emphasize Sicilian origin”, “no nuts mention — we don’t serve any”. The note applies only on the next regenerate, then clears.

The draft is graded in real time by the same six-axis score the card below shows. Most generated drafts land in the green band on first try, so you can accept and move on. If a tile catches your eye — say, the Sensory tile is yellow — you can either edit the draft directly in the modal or close, regenerate with a note like “more texture words”, and watch the score climb.

The AI writes in English only. After accepting the draft and saving the item, the existing ✨ Translate empty button on each description field fills the other languages from the English. The two tools compose cleanly: the AI writes the English, the translator handles the rest.

What happens behind the scenes

The score lives entirely inside the editor in your browser. Nothing is sent anywhere, nothing is saved, nothing waits on the network. The card recomputes the score every time you press a key by running your text against a list of recognised culinary words, sensory adjectives, regional adjectives, dietary terms, and allergen names. A match contributes points; a miss does not. The same text always produces the same score — the number does not fluctuate on save, does not depend on the time of day, does not change if you reload the page. The number you see is the number the AI sees.

Worked example

A chef opens the editor for Spaghetti carbonara on a Tuesday afternoon. The long description currently reads: Spaghetti carbonara. The score circle shows 10, the dot is red, and the label says Too thin for AI. The length tile says 1 word, contributing 0 of 25 points; every other tile is at zero except a single ingredient hit on the word spaghetti.

She expands the description. After a minute it reads:

Hand-rolled spaghetti tossed in a silky carbonara sauce of guanciale, egg yolk, and aged Pecorino Romano. Slow-cured pork jowl from a traditional Roman butcher, finished with cracked black pepper. Contains gluten, eggs, and dairy. Vegetarian variant available on request.

She watches the number climb past 40, past 50, settling at 63. The dot is now yellow, the label says Could be richer, the ingredient tile is full (carbonara, guanciale, egg, pecorino — capped at four hits), origin and method is full (Roman, hand-rolled, slow-cured, traditional), allergens is full, dietary picks up vegetarian. The length tile is sitting at 42 words — partial credit. The sensory tile shows 1 sensory word (silky).

She lengthens the description slightly and adds two more sensory cues — bright, robust — and a line about the egg yolk being velvety. The score climbs to 80, the dot turns green, the label says AI-ready. She saves the dish. The next time a diner in Bangkok asks an AI search engine for handmade carbonara, the restaurant has a chance at the answer that it did not have a minute earlier.

The whole exercise took about ninety seconds.

What the score doesn’t measure

The score is a signal, not a verdict on your prose. It does not grade your writing’s beauty, your menu’s strategy, or your dish’s quality. A description can score 90 and read like an instruction manual; another can score 55 and read like a love letter. The score measures only the signals AI search engines are known to use to decide whether to recommend a dish. The rest — voice, charm, accuracy, honesty — is yours, and the score has no opinion on it.

It is also English-only. Thai and Italian translations are for your diners, not for the score; the AI engines we target rank English content. When the score eventually expands to other languages, this page will note it.

Photos and videos on a menu item — the visual half of discoverability; AI engines that handle images use the photo’s caption and alt text alongside the description.
How the AI thinks about your menu — the broader picture of how the support AI reads your menu, of which the discoverability score is one piece.
Why multiple AI mouths — the four AI engines the score is tuned for, and why none of them on its own is the whole picture.