AI Itineraries: What a Travel Agent Must Check Before Sending

Before sending an AI-made itinerary, a travel agent must check that every hotel, restaurant and tour exists and is open on those dates, that timings and transfers are physically possible, that entry requirements come from official sources, that prices and inclusions match supplier quotes, and that the plan suits this client. AI drafts; you verify every line.

The checklist below is grouped into seven parts, each item with why it matters and how to verify it, followed by the red flags of an unchecked draft, a prompt that makes the AI admit its assumptions, and a sign-off record to keep on file. It matters because the itinerary goes out under your name: in many places, package travel rules make whoever puts a trip together responsible for it, and a client stranded by an impossible connection won't care which tool drafted it. Check with your trade body or a solicitor what applies to your business.

Follow me on Instagram@sagnikteaches

The seven-part itinerary checklist

A. Everything named exists, is open, and is what the draft says

B. Time, distance and sequence work in real life

The time-zone row catches more than you'd expect. An illustrative case: a client flies out at 22:40 on the 14th and lands at 06:15 local time on the 15th. The draft books the first hotel from the night of the 14th, because it read "depart 14th" as "arrive 14th", which pays for a room while the client is over the ocean, about $210 wasted, and leaves nowhere to go at 6.15am on the day that matters. The fix is two changes: the hotel starts on the 15th, and the agent requests early check-in or books the room from the 14th deliberately, as a paid early arrival, if the client wants to sleep on landing. Count nights from arrival times, never from departure dates.

Connect on LinkedInSagnik Bhattacharya

C. Documents, entry and health requirements are confirmed officially

Never let an AI answer these questions for the client. It may state last year's rule with total confidence. If your itinerary mentions entry requirements at all, word it as "we've checked the requirements for [nationality] passport holders on [date]; requirements can change, so please recheck before travel", and link to the official source.

Subscribe on YouTube@codingliquids

In practice the difference looks small and matters a lot. An AI draft line for a two-country trip read: "No visa is required for this trip, and your passport just needs to be valid for your stay." Both halves were stated with no source, and the second is wrong for many destinations that ask for months of validity beyond the return date. The replacement, after checking: "We checked entry requirements for both of your passports on 12 September. Country one: no visa needed for stays under 90 days; passport valid for six months beyond your return. Country two: electronic travel authorisation required before you fly; we'll send the link. Requirements can change, so please recheck a week before departure." It's longer, dated and specific to these travellers, and it tells the client what they have to do.

D. Money matches the suppliers

• Every price comes from a supplier quote or your booking system, never from the AI draft. AI-generated prices are guesses. Search the draft for any number with a currency sign and trace each one.
• Inclusions are exact. "Breakfast included" on a room-only rate is a complaint waiting to happen. Check board basis, transfers, taxes and resort or city fees payable locally.
• Deposit, balance date and cancellation terms match each supplier's conditions and your own terms.
• Currency is stated for anything the client pays locally.

A quick sum shows why this section can't be skimmed. Say the draft describes the coastal hotel as "$180 a night including breakfast" for seven nights. The supplier confirmation shows $196 room-only, with breakfast at $22 per person per day. For two people, the client's real cost is $196 plus $44 a night, or $240, against the $180 they read: $420 more over the week, and a client who will reasonably ask why the itinerary said otherwise. The number came from the AI's general sense of what such hotels cost, and it was close enough to look right. That's the danger. Trace every figure to a document with a reference number, and delete any you can't.

E. It fits this client, not a generic traveller

• Mobility and pace. Stairs, walking distances and hills against what the client told you. An 80-year-old and a hiking couple should never get the same day plan.
• Dietary needs and allergies. Restaurants suggested for a coeliac client need to be confirmed with the restaurant, not assumed from a menu.
• Children. Ages against minimum ages for tours and activities; room occupancy against the rooms booked.
• Interests and budget. Did the draft drift into things the client said they didn't want? AI tends to fill every day with the most famous sights.

Client fit fails quietly, because the draft looks like a good trip, just not for these people. An illustrative family booking, two adults with children of 9 and 12, came back from the assistant with a sea-kayaking morning on day 4. The operator's minimum age is 12, so the younger child couldn't go, and the booking form would have stopped them on the day, not in advance. The same draft suggested a wine-cellar tour on day 6 "for the adults", which the parents hadn't asked for and couldn't do without childcare. The agent swapped both: a family paddleboard lesson with a minimum age of 8, and a beach afternoon. Put the client's ages, mobility and stated dislikes at the top of the drafting prompt, then check each activity's own minimum age and fitness notes, which the AI won't know.

F. Safety and timing on the ground

• Official travel advice for each destination, from the government travel-advice service for your client's nationality, checked close to sending and again before departure.
• Season and weather. Rainy or storm seasons, extreme heat, and mountain passes or ferry routes that close in winter.
• Local holidays and events that close attractions, fill hotels or change transport timetables on those dates.

G. The words themselves

• No promises you can't keep: "guaranteed", "exclusive", "private" (if it's shared), "skip the line" (if it isn't).
• Supplier, hotel and place names spelled correctly and consistently.
• Dates, days of the week and night counts agree everywhere in the document.
• No leftover AI phrasing or placeholders ("[insert]", "as an AI", "hidden gem" on every page).
• Your terms, "subject to availability and confirmation" wording, and contact details for emergencies during travel.

A typical wording fix, before and after. The draft: "Enjoy a private, skip-the-line tour of the old town's hidden gems with an expert local guide." What was actually booked: a small-group walking tour, maximum 16 people, with timed-entry cathedral tickets at 10am. The sent version: "Small-group walking tour of the old town (up to 16 people, about three hours), with timed-entry cathedral tickets booked for 10am, so no queue for the ticket office." Every word of it can be checked against the supplier voucher, and nothing in it promises more than the client will get.

Red flags that a draft went unchecked

Experienced agents learn to spot an unedited AI itinerary at a glance. If you see these in your own draft, slow down:

• Every day has the same shape: morning sight, lunch spot, afternoon sight, "evening at leisure".
• Two places far apart on the same half-day, with no transfer listed.
• Restaurants with generic-sounding names and no address, phone number or booking link.
• Museums and galleries scheduled on what, for many of them, is the weekly closing day.
• No mention of luggage between check-out and an evening flight.
• A price, opening time or entry fee quoted to the exact unit with no source.

A prompt that makes the AI list its own assumptions

Before you start checking, ask the assistant that made the draft to tell you where it's least sure. It won't catch everything, but it gives you a head start:

Here is the itinerary you drafted: [paste].
List, as a numbered table:
1. Every named hotel, restaurant, tour, attraction and transport service,
   with the day it appears.
2. Every time, duration, distance and price you stated, and whether it
   came from information I gave you or was your estimate.
3. Every assumption you made about opening days, seasons, entry rules,
   the client's fitness or preferences.
4. The five items you are least confident are correct, and why.
Don't rewrite the itinerary. Don't reassure me.

An illustrative slice of what comes back for a week in one city:

#  Item                              Day  Source          Confidence
4  Transfer airport-hotel, 35 min    1    my estimate     medium
7  Harbour museum, open 10:00-18:00  2    my estimate     low: may close Mondays
9  Dinner, "the fish restaurant      3    my estimate     low: name from memory
   on the square", booking advised
12 Day trip by train, 1h10           5    your notes      high

Two things to notice. Item 9 is the assistant admitting it may have invented a restaurant, which is exactly what you want it to say. And item 12 claims to come from "your notes" when your notes only said "day trip by train, about an hour": the 1h10 was its own figure dressed up as yours. The self-report is a useful map of where to look first, but it can be wrong about its own sources, so it doesn't reduce the list of things you check. The table it returns is your working checklist for sections A, B and D. Anything marked "my estimate" gets verified first. The same approach works for catching invented figures in any AI-drafted document; see catching made-up figures in AI-drafted proposals.

Running the check in about 30 minutes

1. Skeleton first (5 minutes). Dates, nights per hotel, flight times, total days. If the skeleton is wrong, nothing else matters yet.
2. Logistics (10 minutes). Section B, day by day, with the mapping app and timetables open.
3. Existence and opening (8 minutes). Section A, using the assumptions table.
4. Documents and money (5 minutes). Sections C and D against official sources and supplier confirmations.
5. Read it as the client (2 minutes). Sections E to G. Would this person enjoy this day, and does anything promise more than you've booked?

Some travel platforms now help with part of this. Tern, for example, describes an AI that can review an itinerary before it's sent and flag red flags such as a day without transfers or a gap in accommodation. Treat that as a first pass that catches structural gaps, not a replacement for checking facts against suppliers and official sources. For how the drafting side works, see how independent travel agents use AI to draft itineraries faster, and for the wider toolset, the best AI tools for travel agents.

Ten days, two centres: what the check found

Say a client asks for ten days split between a city and a coastal town, with one day trip inland. The agent drafts it with a chat assistant in 15 minutes. The check, using the order above, takes 35 minutes and finds five problems:

• The city-to-coast transfer is given as 90 minutes; the mapping app says just under three hours on the day of travel. The agent moves lunch to the coast and books a later dinner.
• A recommended seafood restaurant closes on Mondays, which is the day it's suggested. Swapped for the agent's own recommendation.
• The museum on day 3 requires timed tickets, which sell out in high season. Booked with the rest of the trip.
• The inland day trip needs a car; the client doesn't drive. Replaced with a guided tour from the supplier list.
• The client's passport expires four months after the return date; the destination requires more. Flagged to the client the same day, before any booking was made.

Fifty minutes in total, against perhaps two hours writing from scratch, and the one error that could have ruined the trip, the passport, was caught because checklist section C doesn't rely on the AI at all.

A sign-off record to keep on file

Keep a short record with each itinerary. If anything goes wrong, it shows what was checked and when; over time it shows which checks catch the most errors. For a wider view of reviewing AI work without slowing down, see setting up human review for AI work.

Filled in for the ten-day trip above, the record takes two minutes and would read something like this (illustrative):

ITINERARY CHECK: 2026-114   Version: 2   Date sent: 19 Sep
Drafted with: chat assistant   Checked by: agent   Time taken: 35 mins
A Existence/opening   [x] done   Notes: seafood restaurant shut Mondays,
                                 swapped; museum timed entry, booked
B Times/transfers     [x] done   Notes: city-coast 2h50, not 90 min
C Documents: passport validity short for client 1, flagged 19 Sep;
             checked via IATA Travel Centre, 19 Sep
D Prices vs supplier confirmations [x] done
E Client fit          [x] done   Notes: no self-drive; guided tour instead
F Travel advice checked on: 19 Sep (recheck 7 days before departure)
G Wording and terms   [x] done
Errors found and fixed: 5

A blank copy to reuse:

ITINERARY CHECK: [client ref]   Version: [x]   Date sent: [date]
Drafted with: [tool]   Checked by: [name]   Time taken: [mins]
A Existence/opening   [ ] done   Notes:
B Times/transfers     [ ] done   Notes:
C Documents (sources + date checked):
D Prices vs supplier confirmations [ ] done
E Client fit          [ ] done
F Travel advice checked on: [date]
G Wording and terms   [ ] done
Errors found and fixed:

Checking AI itineraries: follow-up questions

How long should checking an AI itinerary take?

For a straightforward week in one or two places, allow 20 to 40 minutes once you're practised. Multi-country trips, self-drive routes and anything involving ferries, small airlines or remote lodges take longer. If a check regularly takes longer than writing the itinerary yourself would, the AI draft isn't saving time for that type of trip, and you're better off using AI only for the descriptive text.

Can I tell clients the itinerary was made with AI?

You can, and some agents do, but what clients care about is that it's accurate and that you stand behind it. If you use AI, the checks in this tutorial are what make that promise true. Your terms should say the itinerary is subject to confirmation of each booking, whoever or whatever drafted the text.

Which AI tool makes the fewest itinerary mistakes?

Tools that search the web and cite sources, such as Perplexity or the research modes in ChatGPT and Gemini, tend to invent fewer places than a plain chat answer, and travel platforms that build from supplier data avoid some errors entirely. None removes the need to check. Models and features change often, so judge by testing a trip you know well.

Further reads

• What AI Costs a Small Travel Agency and What It Saves — What AI costs a small agency and where it saves time.
• How a One-Person Travel Agency Can Reply to Enquiries Instantly — Reply to new enquiries fast without sending unchecked detail.
• How to Mention AI Use in Client Contracts and Proposals — How to mention AI use in your terms and proposals.
• AI Hallucinations Explained for Business Owners: Causes and Fixes — Why AI invents confident details, and how to catch them.
• Perplexity vs ChatGPT for Business Research — Which research tool to use for destination checks.
• AI Proposal Mistakes That Cost You the Job — Proposal errors that lose clients, and how to avoid them.
• How Small Tour Operators Use AI for Bookings and Itineraries — How a small tour operator can use AI around its booking system, build private itineraries from a route library, and catch the errors before guests do.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: IATA Travel Centre (Timatic-based passport, visa and health information); Tern's AI product page and Travel Weekly's report on Tern's AI itinerary review, checked September 2026. The worked example is illustrative.