Build a Reservation-Call Script and Multilingual Menu for Your Pub with Claude Code

It’s 7 p.m. on a Friday and the place is packed. Then the reservation phone rings. The person who grabbed it is a part-timer who started last week, and his voice freezes up.

“Uh, four people, right? And, um, the set menu… sorry, let me get the owner.”

By the time I sprint over from the floor and take the receiver, the caller already sounds annoyed. Behind me someone is calling out, “Excuse me, can we order?” Orders and food running both stop. One phone call, and the whole front of house seized up for thirty seconds.

On top of that, a couple of foreign tourists had walked in earlier that night, and we had no English menu. We got through it with pointing and gestures, but when they asked about allergies I had no answer. The only reason it didn’t turn into a “there was egg in that” complaint later was plain luck.

The reservation calls, the multilingual menu, I kept saying “I’ll sort that out someday” for a year. Print it on paper and it never gets updated, and none of the part-timers ever look at it. This article is about turning both of those into something real in a single day, using generative AI, specifically Claude Code.

Key takeaways

• Turn your phone-reservation handling into templates organized by the usual conversation flows, so a new hire can read straight off the page.
• Generate first-draft English, Chinese, and Korean menus from your Japanese (or English) menu in one pass.
• Draw a clear line for allergies and personal data, the parts a human must always confirm rather than handing to AI.
• Use a copy-paste prompt and a check script that catches missing allergen labels in the translated menu.
• Setup takes half a day to a day; updates after that run about ten minutes each. You escape the reprint-the-laminated-card grind for good.

What’s actually happening on the floor

Let me make the reader clear first. This article helps if you run a small bar or pub, maybe 20 to 50 seats, owner-operated or a small chain. More than half your staff are part-timers and turnover is high. Foreign customers are slowly increasing, and you get an English-menu question a few times a month. That kind of place.

Write out the reservation workflow and it looks roughly like this.

1. Phone rings, somebody stops what they’re doing and answers.
2. Get the date, time, party size, whether they want the set menu, seating preference.
3. Write it in the reservation book (paper or app).
4. Ask about allergies or ingredients they avoid.
5. Mention the day’s flow in one line (cover charge, last call time).

Those five steps take a veteran 30 seconds. But a new hire stumbles on step 2, forgets step 4, and skips step 5 entirely. It all lives in people’s heads, so training takes forever too.

The menu side gets neglected even worse. The dishes change with the season, but there’s no time to make an English version. You patch it with a translation app each time, so “oyako-don” comes out as a different English phrase every single visit. Foreign customers keep growing, but this one corner of the shop is frozen in the past.

The usual rework and headaches

• A new hire mishears a reservation (wrong time, wrong party size). You end up calling back to apologize.
• Someone forgets the allergy check, and it surfaces after the food is served. Worst case, it’s an ambulance call.
• An English menu from an agency runs you a few hundred dollars, and you redo it every time the dishes change.
• The translation app’s literal output is bizarre and the foreign guest doesn’t understand it. “Cover charge” comes out as “Pass.”

Use case 1: Template your reservation-call handling

This is the first thing to tackle. Phone handling follows a fixed pattern, so it’s the best possible fit for AI.

The job is to map out the common conversation patterns and turn them into a script a new hire can succeed with just by reading aloud. Hand Claude Code “the facts about my shop” and it writes you a gap-free manual in one shot.

You can use the prompt below as-is. Only rewrite what’s inside the brackets to match your own place.

You are the front-of-house trainer at a small pub. Write a reservation-handling
script that a new part-timer can read aloud on the phone.

[Shop facts]
- Name: (The Corner Tavern)
- Seats: (40), private rooms: (2 rooms, up to 6 people each)
- Hours: (5:00 PM - midnight, last call 11:00 PM)
- Set menus: (two options, $35 and $45, require one day's notice)
- Cover charge: (a $4 small dish per person, always mention it)
- Parking: (none, point them to the nearby paid lot)

[What to produce]
1. An opening-line template for when you pick up the phone
2. A checklist of items to capture (date / time / party size / set menu yes-no /
   private room yes-no / contact number)
3. Sample answers to common questions (parking, kids allowed, allergy handling,
   same-day cancellation)
4. A closing line for reading the booking back to confirm it
Output in plain spoken English a new hire can read aloud. No jargon.

Take the manual it returns, fit it on one page, and stick it next to the phone. That alone makes “let me get the owner” almost disappear.

For the items you capture, keep a printed checklist and you’ll drop far fewer of them. At minimum, this is it.

Item	Why you ask	The accident if you skip it
Date and time	Obvious	Double booking
Party size	Securing seats	Too few seats / too many
Set menu yes-no	Drives prep volume	Run out of ingredients on the day
Allergies	Directly about safety	Serving accident, complaint
Contact number	For day-of confirmation	Can’t reach a no-show

Use case 2: Build a multilingual menu from your menu

Next is the multilingual menu. This is the part AI makes the easiest.

Hand over your dish list and have it translated into English, Chinese, and Korean. Translate not just the dish names but a short description for each, and foreign customers relax. Writing “Motsu Nikomi” alone means nothing, but add “simmered pork offal in miso broth” and it lands.

The prompt looks like this.

Translate the following pub menu into English, Simplified Chinese, and Korean.

[Rules]
- Keep the romanized reading of each dish name and add a short English note
  beside it. Example: Motsu Nikomi (simmered pork offal in miso)
- Leave prices as the original numbers (do not change tax-included / tax-excluded
  notation)
- For dishes that need an allergy warning (egg, milk, wheat, buckwheat, peanut,
  shrimp, crab), add a note in each language
- Use natural wording that a real local menu would use, not a literal translation
- Output a table per dish: "Japanese / English / Chinese / Korean"

[Menu]
- Edamame  $4.80
- Dashimaki Tamago  $5.80 (egg)
- Motsu Nikomi  $6.80
- Karaage fried chicken  $6.80 (wheat)
- Ebi Mayo  $7.80 (shrimp, egg)

The thing that matters here: do not leave the allergy labels entirely to the translation. AI sometimes forgets to add “egg.” So a human has to cross-check every time. The check script below helps with that cross-check.

What to delegate to AI and what a human must decide

Get this fuzzy and you’ll have an accident. Settle the line with a table up front.

Task	Safe to delegate to AI	A human must confirm
Drafting the phone script	Yes, up to the draft	Whether it fits your operation
Menu translation and descriptions	Yes, up to a first draft	Accuracy of dish names and prices
Allergy labeling	Only suggesting candidates	Always a final human check
Writing in the reservation book	No	A human writes it
Handling personal data (contact info)	No	A human manages it

The rule of thumb is simple. If a mistake can be fixed with an apology, let AI draft it. If a mistake affects someone’s health or your reputation, a human decides. Allergies and personal data, those two alone, you never hand off completely.

A copy-paste check script

If you eyeball the translated menu only, you will miss something once the dish count climbs. So I prepared a script that mechanically finds “dishes where the Japanese has an allergen note but the English is missing the matching word.” It runs if you have Node.js.

Save it as check-menu.mjs and run it with node check-menu.mjs.

// Map Japanese allergen notes to the English word the translation should contain
const allergenMap = {
  "卵": "egg",
  "乳": "milk",
  "小麦": "wheat",
  "そば": "buckwheat",
  "えび": "shrimp",
  "かに": "crab",
};

// Paste your own menu here (jp = Japanese line, en = English translation)
const menu = [
  { jp: "だし巻き卵 $5.80(卵)", en: "Dashimaki Tamago (rolled egg omelet) $5.80" },
  { jp: "鶏の唐揚げ $6.80(小麦)", en: "Karaage (fried chicken) $6.80" },
  { jp: "海老マヨ $7.80(えび・卵)", en: "Ebi Mayo (shrimp with mayo) $7.80" },
];

let problems = 0;
for (const item of menu) {
  for (const [jp, en] of Object.entries(allergenMap)) {
    const needsAllergen = item.jp.includes(jp);
    const hasAllergen = item.en.toLowerCase().includes(en);
    if (needsAllergen && !hasAllergen) {
      console.log(`Check: "${item.jp}" is missing the ${jp} (${en}) label in the English`);
      problems++;
    }
  }
}

if (problems === 0) {
  console.log("OK: no missing allergy labels found");
} else {
  console.log(`\n${problems} item(s) to confirm by eye`);
}

In the example above, “Karaage” has no “wheat,” so it correctly raises a warning. It isn’t a perfect inspection, but as a gatekeeper that points you at the likely gaps it’s plenty. Use it on the assumption that a human always looks last.

What changes before and after

These are my own rough numbers, but the shift is about this much.

Before, a single reservation call took a veteran 40 seconds on average, and a new hire over a minute counting the hand-off. At 20 calls a day, the new-hire shifts were burning more than 20 minutes on the phone. After we stuck up the manual, even a new hire wraps it in 30 to 40 seconds, and the owner hand-off all but vanished.

For the menu, an agency English version runs $300 to $500 the first time, plus an added fee every seasonal update. An AI first draft plus my own check costs half a day the first time, and about ten minutes per update. Near-zero upfront cost, and being freed from reprinting paper is the biggest win.

A rough ROI looks like this.

Item	Before	After
One call (new hire)	60 sec + hand-off	35 sec
First English menu	Agency $300-$500	Half a day of my own work
Menu update	Re-order from agency	10 minutes
Missed allergy check	Happens now and then	Sharply cut by the check script

If you save even 10 minutes a day of phone time on a staffer at $12 an hour, that’s about $60 a month. Skip four seasonal agency orders a year and that alone is a few hundred dollars. The setup cost is basically just your own time, so the payback comes fast.

Security and personal-data cautions

Don’t paste everything into AI just because it’s convenient. Even a pub handles personal data.

• Don’t paste a reservation guest’s name or phone number into generative AI. Drafting the manual template needs no real customer data. Build the samples with fake values like “John Smith” and “555-0000.”
• The final call on allergy handling always goes to the kitchen and a human. Treat the AI output as a draft.
• Before the translated menu leaves the shop (to a print vendor or social media), have two or more people confirm the prices and dish names.
• Pasting long text into a free translation service can mean the content gets used for training. Don’t put your shop’s original recipes or cost data in.

The whole “what’s okay to hand over and what isn’t” question is easier to settle when you write it up as project rules. The thinking in CLAUDE.md best practices is a useful reference.

FAQ

Q. Can an owner who isn’t tech-savvy use this? A. The phone script and menu translation are just things you ask for in plain words, so no programming is involved. Only the check script needs Node.js, and you can substitute an eyeball check if you don’t have it. The best starting point is the on-ramp for non-engineers.

Q. Can I trust the translation accuracy? A. Dish names and descriptions come out largely natural. But always confirm prices, allergies, and any specific cooking method by hand. AI sometimes makes a “plausible mistake.”

Q. My menu changes by season. Do I rebuild it every time? A. Just add the changed dishes to the prompt and regenerate, and it’s done in ten minutes. There’s no need to rebuild the whole thing.

Q. I can’t judge whether the Chinese or Korean is correct myself. A. For life-critical parts like allergy notes, it’s worth having a translation-savvy friend or a paid service check it once to be safe. For the rest of the descriptions, a little awkwardness is within tolerance.

Q. Any tips for writing good prompts? A. The more concretely you hand over the conditions as a bulleted list, the better the accuracy. I’ve written up the approach in the basics of prompting.

What I confirmed when I actually tried this

I ran this whole flow end to end on a fictional pub menu of 15 dishes.

For the phone script, just handing over the shop facts as bullet points produced, in one shot, a single page covering everything from the opening line through the allergy check to reading the booking back. The only thing I fixed was the wording for my own last-call time. It was usable nearly as-is.

For the menu translation, I deliberately removed “egg” from the English of “Dashimaki Tamago (egg)” and ran the check script, and it correctly raised a “Check” warning. The correctly translated lines passed straight through. The payoff was that the machine reliably caught the one item an eyeball would miss.

On the other hand, the AI had phrased the Korean allergy note for one dish differently, and the script couldn’t catch that. So we come back to the obvious conclusion: the life-critical parts need a final human check.

For both the phone and the menu, just switching from “build it from scratch” to “a human fixes the AI’s draft” turned a half-day job into realistic work. If you want to lighten your daily tasks the same way, productivity tips and habits are a good companion read. If you want to talk through a shop-wide rollout, including staff training, training and consulting is where we can design the concrete steps together.

For reference, when picking tools it helps to check primary sources on how generative AI works and what it costs, such as the Anthropic documentation, so your decisions don’t drift.