Can ChatGPT write
your restaurant website?
(And should it?)

Leer este artículo en español →

Open ChatGPT. Type "write the About page for my Italian restaurant." Paste what comes back into a blank document and read it the way a guest will read it on your live site — cold, on a phone, deciding whether to book. That paste is the artifact this piece tears down: line by line, what the model nailed, where it quietly fell short, and the one fix each gap triggers.

Can ChatGPT write your restaurant website? Yes — it can write parts of one fast and well, and it gets the parts that actually sell a table quietly wrong. The output will be grammatical, competent, and instantly familiar, because it is the exact copy every other Italian restaurant owner has been staring at for two years, with minor word swaps. Nothing is broken. The fit is wrong — it could be any restaurant, and a guest can feel that without being able to name it.

So the useful question isn’t whether AI can write the words. It’s which parts to hand it, how to brief it, and where to stop — because the parts of your site that make a guest choose you instead of the place across the street are exactly the parts AI is worst at writing unprompted. Start with the same prompt run three ways, then read the output against what a working site needs.

All three are grammatical. All three would "pass" an editor's review. Only one of them makes a guest want to eat there — and the distance between stage two and stage three is the whole teardown. Read against a working site, here is where the model earns its place and where it doesn’t.

What it nailed: where AI is genuinely good for a restaurant website

I am not an AI skeptic. I use ChatGPT every day. Read the paste honestly and three sections of it hold up — they are real, durable use cases for an operator doing their own website work at eleven p.m. on a Tuesday. These are the rows where the model earns its keep:

1. Menu item descriptions at scale

If you've got a fifty-item menu and no time, AI is very good at turning a one-line internal prep note ("brick chicken, lemon, rosemary, served over potato") into a two-sentence guest-facing description. That’s genuinely useful. It replaces the "same dish name with no description" default most restaurant sites ship with.

The key is the prompt has to include your voice samples. Give ChatGPT three menu items that are already written the way you want, then ask it to match that register for the rest. Without the voice samples, you get the stage-1 output above, applied fifty times.

Here is the whole teardown in one before-and-after. Same prep note, same dish — drag the divider and watch the AI default on one side become the working copy on the other. The only change to the prompt is the voice samples. That single move is the difference between a line you’d let a guest read and one you wouldn’t:

2. Draft copy for pages the guest won’t actually read carefully

Not every page on your website is load-bearing. Your Privacy Policy is not closing reservations. Your Allergen Information page is not changing a guest's mind. For compliance-flavored pages that just need to exist and be competent, AI is fine. Ship it, move on.

3. Structure and outlining

AI is genuinely better than most humans at outlining. Ask ChatGPT "what sections does a restaurant About page typically have, ranked by how much they affect a guest’s decision to book," and you’ll get a useful framework in thirty seconds. Use the outline. Write the actual words yourself.

What it missed: where AI gets your restaurant quietly wrong

Now the rows that fail the read — and they fail quietly, which is what makes them dangerous. Nothing here trips a grammar check or an editor. The copy looks finished. It is just finished in the shape of a thousand other restaurants instead of yours.

1. The specifics that make guests trust you

Restaurant websites work or don’t on specifics. "We source from a farm in Poolesville" is worth ten sentences about "locally sourced ingredients." "Our pastry chef trained at Del Posto" is worth a paragraph about "our talented team." AI, unprompted, defaults to the general because the general is what’s in its training data. Guests trust the specific because the specific is the opposite of a template.

2. Your cadence

Every restaurant that’s written its own copy has a cadence — how long the sentences run, where the em-dashes fall, what register the voice sits in. AI flattens that within the first paragraph. You can brief it to fix this, but most operators don’t, because briefing the AI properly takes about the same amount of time as just writing the thing.

3. The moments that require judgment

The hardest part of writing restaurant copy isn’t finding words — it’s knowing which three things to say in a four-sentence paragraph and which seven not to. That’s judgment, and judgment is exactly what a language model can’t do. It doesn’t know that the story about opening during the first winter of the pandemic is the one that sells the room, and that the story about the soft opening menu change three years later isn’t. You know that. You’re the one who has to type it.

The parts of a restaurant website that make guests choose you are precisely the parts an AI is worst at writing. Which means the time you save with ChatGPT has to be reinvested somewhere, or the site gets flatter, not faster.

ChatGPT can write your restaurant's website. It can't decide which parts of you make it onto the page. That's still your job.

The one-page briefing that fixes it

If you’re going to use AI for your restaurant copy — and you should — you need to give it something other than "write my About page." Here’s the briefing template I use to handle my own content, and the one I’d hand any operator who wants to do the same. Write it once, save it in a Google Doc, paste it as the first prompt every time you open a new ChatGPT conversation.

1. The restaurant, in one sentence. Neighborhood, cuisine, years open, one distinctive detail. “A twelve-year-old Italian place in Takoma Park known for its Sunday red sauce.” One sentence.
2. Three real specifics. The farm you source from, the chef’s background, the one dish guests ask about most. Specific nouns, specific people, specific places. These are what AI otherwise refuses to invent (rightly).
3. Three voice samples. Three short paragraphs already written the way you want to sound. Your best Instagram caption, your best menu description, your best email to your regulars. AI is a mimic; give it something real to mimic.
4. Three things you never want it to say. “Authentic,” “passionate,” “elevated” — the words every restaurant copy default defaults to. Ban them explicitly.
5. The task, narrow. Don’t say “write my About page.” Say “write a 150-word intro paragraph for the About page that mentions the Sunday red sauce and ends with a line that makes a new guest want to book this weekend.”

With that briefing in place, AI goes from giving you stage-1 output to giving you something closer to stage-2. It won’t reach stage-3 — the native voice, the part that feels like only your restaurant could have written this — but it’ll get you 70% of the way there. Your job is the last 30%: adding the one specific detail the AI didn’t know, cutting the sentence that reads a little too brochure-y, swapping a generic adjective for a concrete noun.

So… should it?

Here is the whole teardown as one table — every page and task on a restaurant site, scored by whether the model earns its place. Read it as the verdict on that pasted draft: the rows marked yes are the hours you get back; the rows marked no are the ones a guest can feel you didn’t write.

Yes, for menu descriptions, outlines, and compliance copy. Hesitantly, for first drafts of pages you’ll heavily edit. No, for the three pages that actually close a guest: your homepage hero, your About story, and the one-line reason someone should pick your restaurant over the identical-looking one four blocks away.

Those three, you write. Even if it takes you a weekend. Even if the first draft is bad. Because a weekend of bad sentences written by the person who owns the restaurant will always beat a polished AI draft written by no-one in particular. That’s not sentimentality — it’s a conversion point. Guests can feel the difference between “someone who cares about this place wrote this” and “something that didn’t care about anything wrote this.” The conversion difference between the two is measurable, and not small.

ChatGPT is a power tool. It’s very good at moving heavy things across the room. It is not a carpenter. Use it for what it’s for, and keep the judgment work — the voice, the specifics, the moments that make a restaurant feel like a place and not a business — for yourself.

And when you want to know which paragraphs on your live site already read as AI-generated, the free Menu Copy Inspector and the restaurant audit will tell you, row by row, before a guest ever does.