Writing good dish descriptions
This page is for whoever writes the long description field on a menu item — chef, owner, or marketing. Following this pattern makes your dish read better on the public menu and makes the AI’s recommendations better, because the AI uses the same structure when it speaks about the dish to a customer.
What it does
Most menu descriptions fail one of two ways. They list every ingredient (“Spaghetti, tomatoes, garlic, olive oil, basil, parmesan”) and bore the reader. Or they pile on adjectives (“Our delicious, mouth-watering, authentic homemade spaghetti…”) and convince nobody.
A good description is built around one hero, one sensory detail, and one anchor — either where the dish comes from or what to pair it with. Three short beats. Anything more is filler.
How to use it
The four-part pattern
Write each description as a small structure. The first part is mandatory; the other three are optional but encouraged.
1. Hero — the one thing that makes this dish what it is
Not all the ingredients. The defining one. For carbonara: guanciale. For risotto alla milanese: saffron. For burrata: the milk and the day it was made.
If you can’t name the hero in two words, the dish description has a structural problem.
2. Sensory anchor — one moment the guest can taste, smell, or feel
- “The crust shatters on the first bite.”
- “The cheese is still warm in the center.”
- “It smells like the orchard.”
Pick something specific or skip this part. Generic intensifiers (“delicious,” “amazing,” “perfect”) are not sensory — cut them.
3. Sourcing or geographic anchor — where it comes from, when it’s interesting
- “The mozzarella is delivered every morning from a small caseificio in Battipaglia.”
- “Our flour is from a 200-year-old mill in Umbria.”
Skip if the honest answer is “the supermarket.” Never invent provenance — the AI is told to drop this section rather than make it up, and you should too.
4. Suggested moment or pairing — when to order it, what to drink with it
- “Order it at the start of the meal — it sets the temperature.”
- “Sangiovese on the by-the-glass list opens it up.”
- “Best for the table to share.”
Worked example
Raw description (data, not prose):
Spaghetti carbonara with guanciale from Amatrice, Pecorino Romano DOP, fresh egg yolks. Al dente.
Following the pattern:
“Guanciale from Amatrice is the heart of this carbonara — pork cheek cured 90 days, rendered until the fat coats the pasta. The black pepper is cracked over the plate at the pass. A glass of Frascati cuts through the richness; we keep one open.”
The four beats are there: hero (guanciale), sensory (fat coats pasta, pepper cracked), provenance (Amatrice, 90 days), pairing (Frascati). Two sentences. No filler.
Words to avoid
- Empty intensifiers — delicious, amazing, mouth-watering, wonderful, truly exceptional. If everything is exceptional, nothing is.
- Critic language — highly regarded, award-winning, world-famous. The guest is sitting in front of the menu; they’re already here.
- Hedging — you might enjoy, some people find, if you like. Just recommend.
- Long ingredient lists — if you find yourself listing six things, you’re writing a recipe, not a description.
Things to write more of
- Concrete pairing reasoning — why this wine cuts the cream, why this beer balances the heat. The AI will quote you.
- Provenance with a specific name — not “imported flour” but Mulino Marino; not “Italian cheese” but Caseificio Stagni di Felino.
- House preference — “this is what the chef opens for himself” — guests love being told what insiders pick.
What happens behind the scenes
The long description field on each menu item is read by two consumers:
- The public menu — diners see this directly when they tap into a dish on the customer-facing site.
- The AI — when it generates a recommendation, it pulls the long description and rewrites it through the same four-part pattern. If your description already follows the pattern, the AI keeps your wording. If it doesn’t, the AI restructures — but with weaker material, so the output is weaker.
This is why writing the description well once saves time forever — the AI inherits your voice.
Examples
- 🍕 Standard case — Pizza Margherita. Hero: the buffalo mozzarella. Sensory: “still molten at the table.” Sourcing: the caseificio in Battipaglia. Pairing: “the house Aglianico.” Done.
- 🍅 Common variation — A dessert. Hero: the citrus, if it’s a lemon tart; the spirit, if it’s a tiramisù. Sensory: temperature and texture. Skip Part 3 unless the ingredient is special. Part 4 becomes a coffee or amaro pairing.
- ⚠️ What to do with a vague dish — “Mixed antipasti platter.” Don’t write generic. Name two or three things specifically: “the salame from Felino, the burrata from this morning, our pickled vegetables from the cellar.” That’s what the guest will remember.
Related features
- How the AI uses your food & beverage knowledge — the three-tier knowledge model
- Recommending wine with food — for Part 4 pairing logic
- Photos and videos on a menu item — what to attach alongside the description