Can you trust ChatGPT to build a client itinerary?
Not on its own. ChatGPT drafts a convincing itinerary in seconds but gets details wrong in most. Here is what fails, why, and a checklist for using AI safely.
Karthik · Co-founder, Totem
4 min read
On this page
In short
- In a 2024 test of 100 ChatGPT itineraries, 90% contained at least one error, from closed venues to visits outside opening hours.
- General AI writes what an itinerary usually says, not what is true for your dates, client and suppliers.
- AI is reliable when it works from material you trust, such as your own hotel and activity descriptions, and unreliable when it invents.
- Before anything goes to a client, check bookability, opening hours, transfer times and sources, and keep prices out of the prose.
Short answer: not on its own. ChatGPT and similar tools are very good at drafting the shape of a trip, and unreliable on the details a client actually experiences: whether a venue is open, how long a transfer takes, and whether the hotel you named still exists. Use them for a first draft, never as the final source. Every fact that reaches a client should trace back to something your team or a supplier has confirmed.
How often does AI get itineraries wrong?
The best public test we know of is small but telling. In 2024, the agency SEO Travel asked ChatGPT (the 3.5 model) for 100 two-day city itineraries and checked each one by hand. Travolution reported the results:
| Problem found | Share of itineraries |
|---|---|
| At least one error | 90% |
| A visit scheduled outside opening hours | 52% |
| Michelin-starred restaurant included despite the cost | 30% |
| A restaurant, café or attraction that had closed | 24% |
| Backtracking or unnecessary detours | 25% |
Newer models are better than ChatGPT 3.5, and connecting them to web search helps. But the underlying failure hasn't gone away: the model is writing what a good itinerary usually says, not checking what is true for your dates, your client and your suppliers.
Why does a general AI get travel details wrong?
Three reasons, and they matter because each one has a different fix.
- It doesn't know your trip's context. The model doesn't know the client is travelling in Golden Week, that the hotel is closing a wing for renovation, or that your guide in Kyoto doesn't work Tuesdays. Without that, it fills gaps with what is typical.
- It has no idea what you've contracted. A DMC sells what it has rates and relationships for. An AI that recommends a lovely ryokan you have no contract with has produced a sales problem, not an itinerary.
- It is confident whether or not it is right. A closed restaurant is described with the same fluency as an open one. Nothing in the text tells your consultant which sentences to check.
What can AI do well in itinerary work?
Quite a lot, as long as an expert stays in the loop:
- Turning a messy brief into structure. "Family of four, nine nights, cherry blossom, around €32,000, not too rushed" into a day-by-day skeleton in seconds.
- Sequencing and pacing. Spotting a punishing day, suggesting where a rest day belongs, grouping sights by neighbourhood.
- Rewriting your own copy. Adapting a hotel description you've already written for a honeymoon couple instead of a family.
- Translation and tone. Producing a German or Spanish version of a proposal for your team to review.
The pattern is simple: AI is reliable when it works from material you trust and unreliable when it invents material.
A checklist before any AI-assisted itinerary goes to a client
Use this whatever tool you use.
- Every property and supplier is one you can actually book. Cross-check against your contracts, not the AI's suggestions.
- Opening days and hours are checked for the travel dates, including public holidays and seasonal closures.
- Transfer times are realistic. Check them in a routing tool at the right time of day, and add the buffers your team always adds.
- Descriptions come from your library or a verified source, not from the model's memory. Old room names and closed restaurants hide here.
- Prices are not in the copy. Any figure in the itinerary should come from your costing, not the narrative.
- Someone who knows the destination has read it end to end. If nobody on the team would sign it, it doesn't go out.
How should a DMC set up AI so it doesn't invent things?
The fix isn't avoiding AI; it is changing what the AI draws on.
- Give it your knowledge first. Your hotel profiles, activity descriptions, images and local notes are more accurate than anything on the open web. Tools that search your own library before writing anything make far fewer errors.
- Keep the source on every block. Each paragraph should show where it came from: your library, a supplier document or AI-written. Then reviewers know exactly what to check.
- Make "unreviewed" visible. Anything nobody has signed off should be flagged before the proposal can be sent.
- Keep pricing separate from prose. Costs should come from contracted rates in a costing sheet, so a confident sentence can never change a price.
This is the approach we took when building Totem: it reaches for the team's own library first, shows the source of every section, and flags anything unreviewed before a proposal goes out. Whatever tool you choose, those are the questions worth asking in a demo.
Questions people ask
- Is it OK to use ChatGPT for client itineraries?
- Yes, as a drafting tool. Treat its output like a junior colleague's first draft: useful structure, and every fact needs checking before it reaches a client.
- Will AI replace DMC consultants?
- Unlikely. The valuable part of the work is knowing the destination, the suppliers and the client. AI removes the typing; it doesn't remove the judgement. If anything, consultants who use it well can handle more enquiries without cutting corners.
- What is the biggest risk of AI-generated itineraries?
- Confident errors that nobody notices: a closed venue, an impossible transfer, a hotel you can't book. They are cheap to fix before sending and expensive once a client is standing outside a locked door.