Use Cases (Updated: 6/7/2026)

Claude Code for Cafe Owners: Cut Review Replies and Seasonal Menu Posts in Half

A practical Claude Code guide for cafe owners: draft review replies and seasonal menu posts with copy-paste prompts and a check script.

Claude Code for Cafe Owners: Cut Review Replies and Seasonal Menu Posts in Half

It’s 11 p.m. You’ve finished closing, the till is counted, and you finally drop into a chair. You open your phone and there it is: a fresh three-star review on Google Maps. “The coffee was great, but the staff felt cold.”

Half of you bristles, half of you feels bad. You have to reply, but what do you even write? Sound defensive and you lose. Paste an obvious template and you lose. You spend ten minutes drafting and end up with “Thank you for your valuable feedback.” You go to bed wondering if that was the right call.

When I talk to cafe owners, almost all of them are worn down by this exact “late-night reply hour.” On top of that, the start of every month brings the seasonal menu announcement. Instagram, X, the chalkboard by the door, the printed counter card. Writing each one a little differently is quietly exhausting.

This is the kind of work where letting generative AI handle the first draft takes a huge weight off. Today I’ll walk through how, with prompts you can copy and paste straight away.

Key takeaways

  • For review replies and seasonal menu posts, the safest split is: let AI write the draft, and you do the final “temperature” adjustment by hand.
  • I’ve included three ready-to-paste prompt templates. Swap in your shop name and details and you can use them today.
  • There’s a pre-post checklist and a verification script that catches AI making up the wrong numbers.
  • Never hand personal data (a visitor’s name, a reservation) to the AI. This one line you draw clearly.
  • The roughly 20 hours a month I see owners spend on writing can drop to under half.

First, who is this for?

I’m picturing an owner like this:

  • A privately run cafe with around 20 seats and 2 to 4 staff.
  • The owner does the serving, the buying, and the social media.
  • Google Maps and Instagram are on your mind, but you can’t keep up.
  • You don’t dislike writing, but you have no spare time to spend on it.

You don’t have a dedicated PR person like a big chain does. So every hour spent writing is an hour taken straight out of prep time or a break. This article is about winning that time back with generative AI.

The “writing” workflow inside a cafe

Lay it out and a cafe’s writing tasks mostly follow this flow.

SituationWhen it happensTime it takes now (per item)
Review reply (positive)Irregular, a few a week5 to 10 min
Review reply (negative / complaint)Occasional, but draining15 to 30 min
Seasonal menu post (social)Once or twice a month30 to 60 min
Seasonal menu post (counter card)Once or twice a month20 to 40 min
Event / temporary closing noticeIrregular10 to 20 min

Stare at this table and the drain points jump out: replying to negative reviews, and announcing the seasonal menu. The first is emotionally hard. The second simply eats time. AI drafts help in both places.

Common rework

Here are the mistakes I’ve watched owners make.

  1. Replying to a bad review while still angry, then going pale rereading it later.
  2. Pasting the same canned reply on every review, until a regular thinks “this one again.”
  3. Never getting in the mood to write the announcement, and missing the season the menu was built around.
  4. Saying slightly different things on social and on the counter card, leaving customers confused.
  5. Chasing the “wow” shot so hard you forget to write the price and the dates it’s available.

None of these are really about being bad at writing. They’re about not being able to hold your focus while you’re slammed. That’s exactly why it’s worth letting a machine produce the first rough draft.

What to delegate to AI vs. what you decide yourself

Get vague here and you’ll have an accident. Here’s the line, in a table.

StepDelegate to AIYou always decide
Review reply3 draft options, polite-tone tuningFact-checking; whether to apologize and how far to own it
Seasonal menu postStructure, headline, hashtag ideasAccuracy of price, dates, allergen labeling
Counter cardShort hooks, layout ideasWhether the tone fits the shop’s vibe
Posting timingSuggested optionsThe final publish button

The principle is one line: anything involving “accuracy” and “the shop’s responsibility” must pass through a human. The AI is your draft assistant, not your manager.

Use case 1: Generate three review-reply options

Replying to negative reviews is the hardest, so start here. All you do is hand the review text to the AI and have it return three options with different tones.

Here’s a prompt you can use as-is. Rewrite only the shop name and the details lines.

You are the owner of an independent cafe. Write three reply options for the review below.

# Shop info
- Name: Dappled Light Coffee
- Character: house-roasted beans, quiet atmosphere, many solo customers
- What we value: helping every guest feel comfortable

# The review we received (3 stars)
The coffee was great, but I went during a busy time and the staff
felt a bit curt.

# Conditions for the reply
- Option A: polite and sincere, acknowledge it first
- Option B: a little personality, warm and friendly wording
- Option C: concise but not cold
- All: no excuses, include one line about how we'll improve, under 60 words
- Don't invent facts. Don't promise discounts or freebies.

From the three options, pick the one that matches your mood and the reviewer’s temperature, tweak a few names or phrases, and send. That’s how the late-night ten minutes becomes two.

The same shape works for positive reviews. Just change the conditions to “all options positive, but don’t sound like a template.”

Use case 2: Write the seasonal menu post per channel

Say you’re announcing a summer special, “Peach Cream Soda.” Have the AI write versions for Instagram, X, and the counter card all at once.

Write announcement copy for a new menu item, tailored to three different places.

# Menu info
- Name: Peach Cream Soda
- Price: $6.50 (tax included)
- Available: July 1 to end of August (until peaches run out)
- Character: white peaches, house-made syrup, strong carbonation
- Allergen: dairy

# Per-channel conditions
1. Instagram: photo assumed, emoji ok, 8 hashtags, under 100 words
2. X: punchy, no link, under 60 words
3. Counter card: handwritten feel, 3 lines max, must include price and dates
- Never change the price, the dates, or the allergen
- No hype words (best ever, guaranteed, world's #1)

The key is nailing down “never change the price, dates, or allergen.” The AI sometimes massages numbers while trying to make the vibe pop. Spell out the facts you want locked.

Use case 3: A pre-post checklist that prevents accidents

Even with a draft in hand, posting it straight up leaves gaps. Run through this every time before you hit publish.

  • Is the price correct and tax-inclusive?
  • Are the availability dates right (including the day of week)?
  • Is the allergen labeling complete?
  • Any typos in the shop name, place names, or people’s names?
  • Did any personal data slip in (a customer’s name, a reservation)?
  • Any hype words or promises you can’t keep?
  • Does the social copy contradict the counter card?

This check is the human’s job. Asking the AI to “check it” can fail, because the AI can’t always spot a number it invented itself.

A copy-paste verification script

One thing a machine can confirm reliably: whether the correct price and availability survived into the announcement. Use it as a final check that the AI didn’t silently change a number. If you have Node.js, it runs. Save it and run node check-promo.mjs.

// check-promo.mjs
// Mechanically confirm the "correct price, dates, allergen" survived in the draft
const draft = `Peach Cream Soda is here.
Whole white peaches, our house-made syrup.
$6.50 (tax incl.) / July 1 to end of August / Allergen: dairy`;

// Hand-write the facts you decided are "definitely correct" before posting
const facts = {
  price: "$6.50",
  startDate: "July 1",
  endDate: "end of August",
  allergen: "dairy",
};

const missing = Object.entries(facts).filter(
  ([, value]) => !draft.includes(value),
);

if (missing.length === 0) {
  console.log("OK: price, dates, and allergen are all present in the draft.");
} else {
  console.log("Needs review: the following are missing from the draft.");
  for (const [key, value] of missing) {
    console.log(`  - ${key}: ${value}`);
  }
  process.exitCode = 1;
}

What it does is simple: it checks whether the numbers you decided were correct still appear, character for character, inside the text. Even so, it reliably stops accidents like the AI rewriting “$6.50” as “$6.80.” It’s humble, but having this gatekeeper means you can delegate with peace of mind.

Once you’re comfortable, you can teach this kind of check-and-draft routine to Claude Code as a saved procedure, so you don’t retype the prompt every time. For where to start building that, Claude Code for non-engineers and the getting started guide are good entry points. If you want sharper prompts, how to think about prompts pairs well. To save procedures cleanly, CLAUDE.md best practices and everyday productivity tips help too.

What changes, before and after

The numbers are rough guides only. Here’s the loose shift I hear from owners.

ItemBeforeAfter
Reply to a negative review20 min each, dreaded5 min each, calm because a draft exists
Seasonal post (3 channels)60 to 90 min20 to 30 min
Monthly writing totalAbout 20 hoursAbout 8 to 10 hours
Missing the seasonHappens occasionallyAlmost never

Say writing drops 10 hours a month, and you value that time at $15 an hour. That’s about $150 a month of breathing room. But it’s less about the money and more the feeling of getting prep and break time back.

Personal data and security notes

The easier this gets, the stricter you draw the line.

  • Don’t paste a visitor’s name, a reservation, a phone number, or card details into the AI. For a review reply, hand over only the public review text.
  • Even if a reviewer wrote their real name or an exact visit date, don’t repeat it back in your reply.
  • Free AI services sometimes use your input to train the model. If you’re using it for work, choose a setting or plan that keeps your input out of model training.
  • If you’re sharing this with staff, put “what’s ok to paste and what’s not” on a single sheet of paper and post it up.

For the official stance, it’s worth reading Google’s prohibited and restricted content policy for reviews once. It’s a good baseline when you’re unsure how a reply should look.

FAQ

Q. Won’t replies feel robotic and get spotted as AI? A. Stop at the “draft” and add one line in your own words at the end; that brings the human warmth right back. The trick is not to paste it whole. For regulars, add a line that touches on their name or your last conversation yourself.

Q. We get reviews in other languages too. Can it handle that? A. Yes. Add to the prompt “reply in the same language as the review” and it will detect the language and respond. As always, do the fact-checking yourself.

Q. Is a free tool enough, or should I pay? A. Starting in the free tier to get a feel for it is plenty. Once you’re using it daily and want to lock in a procedure, that’s the order to consider a paid plan or something like Claude Code. I’ve gathered learning materials over on the products page.

Q. I want my staff to use it too. A. Decide the personal-data line and the pre-post checklist first, then hand it over. If you want to roll it out across the whole shop, we can set up the operating rules together through training and consulting.

What I found when I actually tried it

I ran this three-option prompt on five real reviews from a cafe I know. Even on a harsh two-star review, Option A came back without over-apologizing but also not cold, and the owner laughed: “I might have managed this myself, but staying this calm at midnight? No chance.”

The seasonal menu side gave the verification script a real job once. The AI helpfully “clarified” $6.50 as “$6.50 before tax,” which contradicted the shop’s listing (tax included). The script flagged “price not found” and we caught it before posting.

Don’t hand it the whole job. Use it as a draft writer and a gatekeeper. At that distance, the drain of writing work clearly drops. Start with the one review that came in tonight, and have it give you three options.

#claude-code #small business #cafe #review replies #social media marketing
Free

Free PDF: Claude Code Cheatsheet

Enter your email and download the one-page Claude Code cheatsheet for commands, review habits, and safe workflows.

We handle your data with care and never send spam.

Level up your Claude Code workflow

Start with the free PDF, use Gumroad guides when you need repeatable workflows, and book consultation when rollout or revenue paths need human judgment.

Masa

About the Author

Masa

Engineer focused on practical Claude Code workflows. Runs claudecode-lab.com, a 10-language technical media site.