ChatGPT can't write your menu, but it can write your menu critique.

Every restaurant trying to use AI for menu writing is using it backwards.

The prompt is always something like: "Write me a menu description for our truffle pasta." The output is always something like: "Indulgent house-made tagliatelle tossed in a luxurious black truffle cream sauce, finished with shaved Parmigiano." Beautiful. Generic. Indistinguishable from the next restaurant down the street that ran the same prompt.

You don't need ChatGPT to write your menu. You need it to critique the menu you already have.

That use case is the high-value one. Here's the prompt.

The prompt

Paste your menu into ChatGPT (or Claude, or Gemini — they all handle this) and use this prompt verbatim:

You are a critical reader of restaurant menus with 20 years of experience in hospitality marketing. Review the menu below and identify the following:

1. Any item description that uses three or more food-cliche words (artisanal, house-made, handcrafted, signature, locally-sourced, farm-fresh, etc.). Suggest a more specific alternative for each.

2. Any item where the description doesn't tell me what makes it different from the same item at the restaurant next door. Flag it.

3. Any item where the price is high but the description is short. Customers reading a $42 entree need more description than a $12 starter. Flag undersold items.

4. Any item where the name doesn't match the description's tone. If the item is named "Nan's Pot Pie" but the description sounds like a fine-dining menu, the dissonance hurts both.

5. The three weakest item descriptions on the menu overall, and what specifically would make them stronger.

Be direct. Don't soften the feedback. Treat this as a working session with the chef, not a customer review.

Here's the menu:

[paste menu here]

That's the prompt. Run it on your current menu. Print the output.

What the output looks like

The output will be uncomfortable. That's what makes it useful.

You'll see something like:

Item 3 ("Truffle Tagliatelle - $38") uses "house-made" and "rich" — both food-cliches. The price is high but the description is 12 words. Consider: "Tagliatelle, cut every morning, in butter and Périgord truffle. Six minutes from order to plate, no garnish."

That's the version of menu description that converts. Not because it's AI-written, but because the AI flagged where your existing copy was working against the kitchen's actual story.

Why this works

Three reasons:

The AI doesn't get tired. Your menu has 23 items. You wrote them over six months. By item 18, your editor brain was done. The AI reads every item with the same attention.

The AI has read every restaurant menu on the internet. When it says "this description is generic," what it means is "I've read this exact phrasing on 4,000 other menus." That's a signal.

The AI has no skin in the game. Your chef wrote half these descriptions. Your marketing intern wrote the rest. Neither of them can critique the other's work without it getting weird. The AI can.

What to keep doing yourself

Three parts of menu writing the AI is genuinely bad at:

Naming. Names need to fit your venue's voice, and that voice is the one thing the AI can mimic but never originate. "Nan's Pot Pie" works because Nan is real. The AI doesn't know who Nan is.

Pricing copy. The decision to write "$38" vs "38" vs "thirty-eight" is brand-level. The AI will follow your existing convention but it won't know which one is right for your room.

The story behind a dish. If the rabbit on your menu is from the farm your grandfather started in 1962, the AI will write that line in a way that strips it of meaning. Write that one yourself.

What this means

The frame is simple: AI is a critic, not a copywriter. A good critic makes your writing sharper. A bad copywriter makes every restaurant sound the same.

Run the prompt above on your menu tonight. Print the output. Walk it to your chef before service tomorrow. The conversation will be the most useful menu meeting you've had this year.

— Damon