AI for Tour Guides: Multilingual Enquiry Replies and Review Responses

Write a fact card for each tour (price, meeting point, start time, inclusions, cancellation terms) and let the AI draft enquiry replies only from it, or from your notes when answering a review. Paste each draft into a second translation tool, read it back in your language, and send it when every date, time, price and place matches your source.

The risk is rarely grammar. A modern assistant writes fluent Spanish or Japanese in seconds, and that fluency is exactly what hides a wrong start time, an invented inclusion or a price worked out for the wrong ages. You cannot proofread a language you do not read, so the process has to do the proofreading for you: fixed facts going in, a second opinion coming out, and a short list of details that you check every time.

Follow me on Instagram@sagnikteaches

Build the tour fact card before you write a single prompt

An assistant asked to "reply to this enquiry about my food walk" fills every gap with something plausible. It does not know that your meeting point is under the clock and not at the side door, or that your cancellation window is 24 hours rather than 48. A fact card closes those gaps. It is a plain text note, one per tour, holding every detail a guest might ask about, and it is the only source the AI may use.

Connect on LinkedInSagnik Bhattacharya

Sort the card into three kinds of fact. Hard facts do not vary: price per age band, start time, meeting point, duration, group size. Conditional facts depend on circumstances: the weather rule, the cancellation window, minimum numbers. "Ask me" items are anything you want to answer personally, such as dietary needs, private-group rates or pick-ups. That last group matters most, because it gives the AI a sanctioned way to say "I will confirm this personally" instead of inventing an answer. Below is a filled-in card for an imaginary market-and-harbour food walk.

Subscribe on YouTube@sagnikteaches

TOUR: Market and harbour food walk
LANGUAGES I GUIDE IN: English, Spanish (conversational)
START: 10:00, Tuesday to Saturday
MEETING POINT: Under the clock outside the main market hall entrance (not the side door)
DURATION: 3 hours, about 2.5 km, 12 stops, two flights of steps
GROUP SIZE: maximum 8
PRICE (USD): adult (13 and over) 45; child 6 to 12 30; under 6 free
INCLUDED: 5 tastings, bottled water
NOT INCLUDED: drinks with meals, hotel pick-up, tips
CANCELLATION: free up to 24 hours before the start; inside 24 hours no refund unless I can rebook the places
WEATHER: runs in rain; cancelled only for severe weather, full refund
PAYMENT: card link sent after I confirm; no cash on the day
ACCESS: steps at stop 4 and stop 9; not suitable for wheelchairs
ASK ME: dietary needs, private groups, pick-ups, anything not above
NEVER SAY: guaranteed places, suitable for all ages, vegetarian-friendly

Two habits make the card work. Write times on the 24-hour clock and prices with a currency code, because the dollar sign is shared by several currencies and "10" could be morning or night to a guest used to a different clock. And keep a "never say" line for promises you do not make. Every guide has a few: "guaranteed places", "suitable for all ages", "vegetarian-friendly". The AI will happily add them if the line is missing, because they sound like ordinary tour-guide copy.

One enquiry, start to finish: Spanish in, Spanish out

The routine has five steps and takes about four minutes once you have done it a few times.

1. Paste the card and the message. Use the guest's first name only. Leave out surnames, phone numbers and booking references.
2. Ask for the reply plus a facts list. The reply goes in the guest's language. The facts list, in English, names every date, time, price, place and number the AI used.
3. Back-translate in a second tool. Paste the foreign-language reply into a different assistant or a translation tool and ask for a literal English version. Using a different tool matters, because the same assistant tends to repeat its own slips.
4. Compare with the card. You are checking facts, not style. Recalculate any price yourself.
5. Send and file. If the reply was a good one, save the final version in your reply library.

This is the prompt I would start with. The rules do most of the work: they cap the length (long replies give errors more room), ban idioms, force dates into words and ask for the facts list that makes checking quick.

You are drafting a reply for a solo tour guide. Use ONLY the TOUR CARD below. If the guest asks for something the card does not cover, do not guess: write "I will confirm this personally" in the reply and list the question under OPEN QUESTIONS.

Rules:
- Reply in the language the guest wrote in, in the same level of formality they used.
- Maximum 90 words. Short sentences. No idioms, no jokes, no exclamation marks.
- Write dates as weekday, day and month in words (Saturday 13 March). Times on the 24-hour clock. Prices as "USD 45".
- Copy the meeting point exactly as written on the card, then add one plain description. Do not translate names of places.
- After the reply, add FACTS USED: every date, time, price, place and number you put in the reply, in English, with the sum for any total.

TOUR CARD:
[paste card]

GUEST MESSAGE:
[paste message]

Now a realistic enquiry. A guest writes in Spanish asking about Saturday 13 March for three adults and two children aged 10 and 13, and wants the start time, the meeting point and to know whether cards are accepted.

Hola, somos 3 adultos y 2 niños (de 10 y 13 años). ¿Tienen algún tour el sábado 13 de marzo? ¿A qué hora empieza y dónde nos encontramos? ¿Se puede pagar con tarjeta? Gracias, [first name]

The first draft came back like this, with the facts list underneath.

Hola [first name], gracias por escribirnos. Sí, tenemos un tour el sábado 13 de marzo a las 10:00. Nos encontramos bajo el reloj, frente a la entrada principal del mercado. Para 3 adultos y 2 niños, el precio total es de USD 195. Pueden pagar con tarjeta mediante un enlace de pago que les enviaré tras la confirmación. ¿Desean que les reserve las 5 plazas?

FACTS USED: Saturday 13 March; 10:00; under the clock, main market hall entrance; 3 adults x USD 45 + 2 children x USD 30 = USD 195; card payment by link after confirmation; 5 places.

It reads well, and it is wrong twice. The back-translation, from a second tool, was close to this:

Hello [first name], thank you for writing to us. Yes, we have a tour on Saturday 13 March at 10:00. We meet under the clock, in front of the main entrance of the market. For 3 adults and 2 children, the total price is USD 195. You can pay by card using a payment link that I will send after confirmation. Would you like me to reserve the 5 places?

Setting that beside the card turns up two problems. First, the total treats both children at the child rate (3 x 45 + 2 x 30 = 195), but the card sets the child rate for ages 6 to 12, so the 13-year-old pays the adult rate: 4 x 45 + 1 x 30 = USD 210. The facts list made the error visible in seconds, because it showed the sum the AI had used. Second, the draft quietly dropped "not the side door". On a busy market morning that qualifier is the difference between meeting on time and phoning each other from opposite ends of the building.

A one-line follow-up fixed both ("The 13-year-old pays the adult rate; recalculate; add the side-door warning"), and the reply that went out was:

Hola [first name], gracias por escribirnos. Sí, tenemos un tour el sábado 13 de marzo a las 10:00. Nos encontramos bajo el reloj, frente a la entrada principal del mercado (no la puerta lateral). Los mayores de 12 años pagan tarifa de adulto, así que el total para 4 adultos y 1 niño de 10 años es de USD 210. Pueden pagar con tarjeta mediante un enlace que les enviaré tras la confirmación. ¿Desean que les reserve las 5 plazas?

That took around four minutes, far less than writing it by hand in a language you only half speak, and the guest saw the right price before booking rather than at the meeting point.

The slips that fluent AI replies hide

After a few dozen enquiries the same slips repeat. This table lists the ones worth training your eye on, with the check that catches each.

The back-translation catches a seventh problem that is easy to miss: the AI sometimes answers the question it thinks was asked. If the guest wanted to know whether a pushchair fits and the reply talks about children's prices, the English version reads strangely at once.

Languages you cannot read: a three-tier rule

Back-translation works because you can judge the English. It tells you whether the content is right, not whether the Dutch sounds natural. How much that matters depends on how well you read each language, so I would sort your languages into three tiers.

• Tier one: you read it comfortably. Let the AI draft and edit as you would any text, and still run the facts check.
• Tier two: you recognise words and can follow a sentence. Use the full routine, keep replies under 90 words, and start collecting the replies that went well into a phrase bank.
• Tier three: you cannot read it at all. Do not send free-form AI text. Reply only from an approved phrase bank, checked once by a native speaker, and let the AI swap in names, dates and numbers. For a first message in a brand-new language, reply in plain English with the time, place and price translated, and say a fuller answer follows.

One paid check often earns its keep: ask a native speaker or a freelance translator to review the phrase bank for each tier-three language. A dozen lines is a small job, and you reuse it for every guest afterwards, so get a quote before deciding. Here is the kind of phrase bank I mean, shown in Dutch with the English meaning beside each line.

Store each line as a quick reply in the free WhatsApp Business app. WhatsApp's help pages describe the shortcut (type a slash in a chat to call up your saved replies) and a cap of 50 stored quick replies, so a dozen lines in each of three languages fits with room to spare. Give each line a shortcut such as /meet-nl. The AI's job then shrinks to filling in the date, the number of guests and the price.

If you also write tour descriptions in several languages, the same fact card feeds that job. The tutorial on tour descriptions in several languages covers the page-copy side, and checking AI translations before customers see them goes deeper on the review step used above.

Review replies in the reviewer's language

Reviews are the other place a wrong sentence travels far, because anyone can read your reply. Two platform rules shape what you ask the AI to write. Tripadvisor's guidelines for management responses rule out offers of gifts, services or money, correspondence from guests, personal details such as names and phone numbers, and accusations of review fraud. They also say that unreadable or incomprehensible machine translations are not accepted, which is an argument for the back-translation step rather than against using AI. Google checks owner replies before they appear: its Business Profile help page on managing reviews says that usually takes up to 10 minutes but can take up to 30 days, and that you can edit a reply afterwards.

The routine mirrors the enquiry one, with one important change: the facts come from your own record of what happened, never from the review. Paste the review without the reviewer's name, add two or three lines of your own notes, and ask for four things: the review's language and star rating, a list of each claim it makes, a reply in the reviewer's language of under 70 words that answers each claim using only your notes and offers nothing, and an English version for you.

Below are a guest review and MY NOTES on what happened. Do four things.
1. State the review's language and star rating.
2. List each claim the reviewer makes (praise and complaint).
3. Write a reply in the reviewer's language, maximum 70 words: thank them for specific praise, answer each complaint using ONLY my notes, do not apologise for anything my notes do not confirm, offer no discount, refund or gift, do not mention the reviewer's name or booking details.
4. Give a literal English version of the reply.

REVIEW:
[paste]

MY NOTES:
[what actually happened, with times and numbers]

A worked example. A four-star review in Italian praises the guide but says the walk started about 15 minutes late and the group was large for the narrow streets. My notes say: start delayed 12 minutes because two guests arrived late; group of 8, the stated maximum. The first draft thanked the guest, apologised for the delay and for "the overcrowded group", and offered 10% off a future tour. That is three problems in one reply. The offer breaks Tripadvisor's rule on gifts and money. The apology for the group size admits a fault that is not one, since eight is the advertised cap. And the reply never gave the true reason for the delay. The corrected version states the facts, takes the point about narrow streets seriously, and promises nothing.

REVIEW (Italian): Bel giro, guida simpatica e preparata. Peccato che siamo partiti con 15 minuti di ritardo e il gruppo era un po' grande per le strade strette.

CORRECTED REPLY: Grazie per le belle parole sul giro e sulla guida. Ha ragione: siamo partiti con circa 12 minuti di ritardo perché due ospiti sono arrivati più tardi e abbiamo aspettato per non lasciarli indietro. Il gruppo era di 8 persone, il massimo previsto. Grazie anche per l'osservazione sulle vie strette: ne terrò conto. Spero di rivederla presto.

ENGLISH: Thank you for the kind words about the walk and the guide. You are right: we started about 12 minutes late because two guests arrived later and we waited so as not to leave them behind. The group was 8 people, the stated maximum. Thank you also for the comment on the narrow streets: I will bear it in mind. I hope to see you again soon.

Keep the facts and the courtesy separate. The facts come from your notes; the AI supplies the politeness. For angrier reviews and fuller sample replies, see the tutorial on replying to negative reviews with AI and the review-reply routine for tour companies.

Bespoke itinerary requests: draft fast, verify like a guide

Some enquiries ask for something that is not on your list: a half-day for four adults who like architecture and dislike hills. AI is good at the first draft of this, a time-boxed sequence of stops with walking minutes and rest breaks, and unreliable on the facts that make it workable. It does not know what changed last month: a closed lift, a new ticket price, a market that moved to another day. Treat the draft as a hypothesis, and run a local-fact check before the guest sees any of it.

Ask for the draft as a table with the stop, minutes at the stop, walking minutes to the next stop and a final column called "what must be true" (opening day, ticket, access). That last column becomes your checklist. Verify each item from the venue's own page or with a phone call, and write down when you checked. This is an invented draft and what the check found:

Two rules come out of that table. Never quote a price you have not checked this season. And add buffer minutes for group size, because a group of eight rarely walks at the pace of a map app; I would start with about ten extra minutes per hour of walking for mixed-age groups and adjust with experience. The checking habits are the same ones a travel agent needs, which the tutorial on checking AI itineraries before sending sets out in more detail.

Guest details that never go into the chat

Guest messages carry personal details. Paste first names and the question, and leave out surnames, phone numbers, passport or card details and hotel room numbers; you rarely need any of them to draft a reply. Check which plan you are on. Consumer ChatGPT and Claude accounts have a switch in their privacy settings to stop chats being used for model training, and business plans do not train on your content by default. Translation tools differ as well: DeepL's Pro page, for example, says content on its Pro plans is never used to train its models and does not make the same statement about the free plan, so read the terms for the tier you use. And if you wonder whether guests should be told that AI helps with replies, the tutorial on telling customers you use AI covers when it is needed and how to word it.

A week of replies in numbers

These figures are illustrative. A solo guide receives 16 enquiries in a typical week: six in English, four in Spanish, three in Italian, two in Dutch and one in another language. Written from scratch, an English reply takes about seven minutes and a reply in a half-known language about ten, because every sentence needs a dictionary check. That is 42 plus 100 minutes, roughly two hours and twenty minutes a week.

With the routine, each reply takes about four minutes: one to paste and prompt, two to read the back-translation and facts list, one to send. Sixteen replies come to about an hour, so the saving is a little over an hour a week. Setup is the cost: perhaps two to three hours to write three fact cards and a phrase bank for each of three languages, so it pays back inside a month. The running cost is a paid assistant plan at around $20 a month (ChatGPT Plus and Claude Pro both list at $20; Google AI Pro at $19.99) plus a free translator for the second opinion.

The time saved matters less than a second effect. When the card is complete, guests stop asking "where exactly do we meet?", and that follow-up message, repeated across a season, is where much of the hidden time goes.

What to count after a month of translated replies

• Corrections per ten replies. Count how many AI drafts needed a fix. The number should fall as the card improves; if it does not, the card is missing facts.
• Follow-up questions from guests. A question that appears twice belongs on the card, with an approved answer.
• Slips that reached a guest. Log each one, name the check that should have caught it and add that check to your list.
• Languages you offer versus languages you can check. If enquiries in a new language start arriving, build the phrase bank before offering it.
• Tool changes. Assistants change often, so rerun your ten test messages whenever you switch tool or the vendor changes its model.

A mistake will reach a guest at some point. When it does, correct it in the same thread, quickly and plainly ("I gave you the wrong total; the correct price for your group is..."), then add the missing line to the card or the checklist so the same slip cannot recur.

Questions guides ask before they trust a translated reply

Which AI tool is best for translating guest messages?

The main assistants (ChatGPT, Claude and Gemini) all handle widely written languages well, and quality tends to fall for less common ones. Test two of them on ten real past messages in your busiest languages, then use one to draft and a different tool to back-translate. The routine catches more mistakes than the choice of brand does.

Should I tell guests that AI helped draft a reply?

If you read and send every reply yourself, many guides treat the AI as a drafting aid, much like a dictionary or a spell-checker. If a chatbot answers guests on its own, tell them plainly that it is automated. Either way, never let the AI make promises about prices, times or refunds that you have not checked.

Can I let AI answer WhatsApp enquiries automatically?

For a solo guide I would not. An automatic reply can promise a place on a full tour or quote the wrong price while you are busy leading a group. Draft with AI, check, then send yourself, and keep your approved lines as WhatsApp quick replies so common questions still get an instant, accurate answer.

How do I answer a review in a language I cannot read at all?

First paste it into a translator so you know what it says. Then give the AI your own record of what happened and ask for a short reply in the reviewer's language, with no offers. Back-translate it with a second tool. If the review is angry or accuses you of something serious, reply in simple English and have a native speaker check any translated line.

How many languages should a solo guide offer?

Start with the two or three that already account for most of your enquiries, and add a language only when you have a checked phrase bank for it. Offering ten languages with unchecked replies is riskier than offering three well. For the rest, answer in plain English and translate only the time, place and price.

Further reads

• How to Offer Multilingual Customer Support With AI Translation — The wider picture of multilingual support beyond a solo guide.
• Tour Cancellation Policy: Draft the Wording and Customer Messages With AI — Word your cancellation terms so every translation says the same thing.
• How to Use AI in the Travel Industry: Weather Cancellations for Tours — Message guests quickly when rain or wind cancels a departure.
• Tour Guide Scheduling Software: What to Pick for Freelance Guides — If you hire other guides, schedule them without double-booking.
• Travel Industry AI Chatbot: What a Tour Operator Website Should Answer — What a tour website chatbot should answer, and what it should not.
• Should You Let AI Auto-Post Replies to Google Reviews? — Why auto-posting review replies is a risk even when drafts are good.
• How Small Tour Operators Use AI for Bookings and Itineraries — How small operators use AI across bookings and itineraries.
• Viator and GetYourGuide Listings: Rewrite Titles With AI — Sharpen your marketplace listings once replies are under control.
• AI Receptionist for a Hostel: Night Calls in Several Languages — Price night phone cover for a hostel in several languages, test each language properly and keep late check-in decisions with a person.
• AI Marketing Booklet Creator: Print a Tour Brochure From Itineraries — Turn three tour itineraries into a print-ready booklet: AI-drafted copy, a pricing page, Canva layout, print specs and a proof checklist.
• Walking Tour Software: Ticketing, Guide Rotas and AI Enquiries — How a small walking-tour company sets group caps, guide rotas, weather rules and private-group quotes, and which writing jobs AI can take on.
• 8 Examples of AI in Tourism From Guided Tours to Group Bookings — Eight practical AI uses for tour guides and attractions, each with the tool type, set-up time, a worked example and a check that it worked.
• AI in Travel and Transport: Jobs a Small Coach Hire Firm Can Hand Over — Follow one coach booking from enquiry to return journey to see which admin jobs AI can draft, and which checks stay with people.
• Can You Make an AI Audio Tour Guide? Scripts, Voices and Licensing — A claim-log method for scripts, what four speech services' own terms say, consent wording for cloned voices and the royalty-free music trap.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Tripadvisor Management Responses Guidelines and Google Business Profile Help (reply to reviews), both read in October 2026; WhatsApp Business help pages (quick replies); DeepL Pro page; ChatGPT, Claude and Google AI list prices from the vendors' pricing pages.