Independent travel agents draft itineraries faster with AI by giving it a structured client brief and a library of building blocks they've already checked (hotels, transfers, day plans, supplier notes). The AI assembles a first draft from those blocks in minutes; the agent fixes gaps and verifies timings. Asking "plan me ten days" from scratch saves far less time.
Most agents who try AI and give up started with the open-ended prompt. The model writes something fluent, full of generic attractions, invented timings and hotels you've never used, and checking and rewriting it takes as long as writing from scratch. The speed comes from restricting the AI to your own knowledge, then letting it do the typing, sequencing and formatting.
Why the blank "plan me a trip" prompt is slow
An open prompt forces the model to guess three things you already know: the client's pace and priorities, the suppliers you trust, and the practical limits of the route. It guesses confidently and plausibly, which is the problem. A day that looks fine on the page might include a transfer the model thinks takes 40 minutes when it takes two hours, a restaurant you'd never send a client to, or a "private sunset cruise" run by a company that doesn't exist.
A single day from an open prompt ("Plan a 10-day honeymoon, relaxed pace, boutique hotels") shows the pattern (illustrative):
Day 5: After breakfast, take the scenic 1-hour coastal drive to the
fishing village. Enjoy a private cooking class at Casa [name], famous
for its family recipes, then relax at the Blue [name] Beach Club.
End the day with a private sunset cruise with [name] Charters.Four proper nouns, none from your supplier list, and a drive time you'd have to look up. Each one is a separate check: does it exist, is it any good, is it vegetarian-friendly, does it run in late May? Four checks for one day is forty for the trip, before you've started on the text.
Every one of those guesses has to be found and replaced. That checking is covered in detail in what a travel agent must check before sending an AI itinerary, and it's unavoidable. What you can avoid is having dozens of guesses to check in the first place.
Stage 1: the client brief the AI drafts from
Capture the enquiry in the same fields every time. It takes five minutes during or after the client call, and it is the single biggest improvement to draft quality. Here is a filled-in example for an illustrative honeymoon enquiry:
CLIENT BRIEF - ref H-0412
Travellers: couple, early 30s, honeymoon. First long-haul trip together.
Dates: 10 nights, flexible by 2 days either way, late May.
Budget: about $9,000 for the land arrangements, flights separate.
Pace: relaxed. Max one hotel change every 3 nights. No 6am starts.
Must have: 3+ nights on a quiet beach; one special dinner; one
cooking class.
Avoid: big resorts, long drives (over 3 hours), group tours over 12.
Style: boutique hotels, sea view where possible, good breakfast.
Dietary/access: she is vegetarian; no mobility needs.
Already booked: nothing.
Notes: loved the small family-run hotel on their last trip; want
"somewhere that feels local".The "avoid" and "pace" lines do more work than any other. They tell the AI which of your blocks to leave out, and they stop it filling empty days with activities.
Stage 2: build a block library you have already checked
A block is a short, reusable description of something you sell and trust: a hotel, a transfer, a half-day tour, a restaurant, a sample day in a town. Each block records the facts you've verified and the date you verified them. Keep them in one document per destination, or in a Project in ChatGPT or Claude so the assistant can read them every time.
A filled-in block might look like this:
BLOCK: Hotel - [boutique hotel name], old town
Type: 18 rooms, family-run. Sea-view rooms: 4 (request specifically).
Best for: couples, relaxed pace. Not for: families with toddlers
(steep stairs, no lift).
Transfer from port: 25 min by car; allow 45 in summer traffic.
Dinner: rooftop restaurant, book 48h ahead in peak season.
Honeymoon extras: fruit and wine on arrival if told in advance.
Our notes: breakfast excellent; rooms at the back can be noisy
on Fridays (bar nearby).
Last verified: March, by phone with the owner.Start with the 20 blocks you use most, not every supplier you've ever booked. Most agents find that a small number of hotels, transfers and experiences appear in the majority of their itineraries. If you want your team to share and reuse these, the same principles as building a shared prompt library apply to blocks.
Stage 3: the drafting prompt, and what comes back
You are drafting a day-by-day itinerary for a travel agent.
Use ONLY the hotels, transfers, experiences and restaurants in the
attached block library. If nothing in the library fits a need in the
brief, write [NEEDS OPTION: what's needed] instead of inventing one.
Respect every "pace" and "avoid" rule in the brief.
For each day give: where they sleep, transfers with time from the
block, one main activity at most, and a dinner suggestion.
Add a "Checks for the agent" list at the end: every timing you
assumed, every place you weren't sure about.
Brief: [paste]
Block library: [attached]An illustrative extract of what comes back for the brief above:
Day 4 - Transfer to the coast
Morning at leisure. 11:00 private transfer to [beach hotel]
(2h 40m per block). Afternoon by the pool.
Dinner: hotel terrace.
Day 5 - Cooking class
[NEEDS OPTION: vegetarian-friendly cooking class near the beach
hotel; library has one class but it is in the old town.]
Checks for the agent:
- Day 4 transfer: block says 2h 40m; brief says avoid drives over
3h. Within limit but close; check in late May traffic.
- Day 7: assumed the ferry runs daily; block doesn't say.This is the draft you want: it used your hotel and transfer, respected the three-hour rule, flagged the missing cooking class instead of inventing one, and told you exactly what to check. What you'd still fix: it placed the cooking class on day 5 just because the brief listed it, when the couple would probably prefer it early in the trip, before the beach days. Sequencing for how a trip feels is still your job.
Stage 4: fill the gaps and verify
Work through the draft in this order, because each step can change the next:
- Fill every [NEEDS OPTION] from your own research or supplier contacts, then add the new supplier to the block library so the gap doesn't recur.
- Check the "Checks for the agent" list item by item: ferry days, seasonal closures, transfer times at that time of year.
- Check every proper noun against a supplier record. Even with a restricted prompt, a model occasionally merges two hotel names or renames a restaurant.
- Read it as the client. Does day 3 feel rushed? Is there a free afternoon after the long flight? This is the step AI does worst and clients notice most.
If the AI's draft cites anything from the web rather than your blocks, verify the source itself; checking sources and citations in AI research covers how.
Stage 5: format for the client
Once the content is right, AI is useful again for presentation. Ask it to rewrite the verified plan in your client-facing style: a short welcome paragraph, day headings with an evocative line each, practical details in a consistent box at the end of each day. Give it one of your best past itineraries (with the client's details removed) as a style example. If you use itinerary software, paste the verified text into it rather than letting the AI format dates and prices, since the software handles those more reliably.
The before-and-after on a single day shows what the style example buys you. Without one, a typical AI day description reads:
Day 6: Set off on an unforgettable journey to the breathtaking coastline, where crystal-clear waters and stunning vistas await. Immerse yourself in the vibrant local culture and indulge in mouth-watering cuisine.
It says nothing a client can use. After giving the assistant a past itinerary as a model and the rule "every sentence must contain a fact from the plan", the same day becomes:
Day 6: A slow morning, then your driver collects you at 11 for the drive to the coast, about two and a half hours with a coffee stop halfway. Your room at the beach hotel has a terrace facing the bay. Dinner tonight is on the hotel's terrace; ask for the table at the far end for the sunset.
Every detail in the second version came from a block, which is why it can be trusted, and it reads like an agent who has been there.
Keeping the block library current without a second job
The library only saves time if it stays accurate, and nobody has time for an annual audit. Three habits keep it current as a side effect of normal work:
- Update on contact. Whenever you speak to a supplier or a client returns from a trip, spend two minutes updating that block and its "last verified" date. Client feedback ("the back rooms were noisy") is gold and belongs in the "our notes" line.
- Let the AI write the first version. After finishing an itinerary for a new supplier, paste your notes and ask the assistant to turn them into a block in your standard format. You check it; it does the typing.
- Review by season, not by supplier. A month before each peak season, ask the assistant for every block in that destination older than six months, and check only those.
For most small agencies that adds up to perhaps half an hour a week, which the drafting savings repay many times over.
The second habit needs one rule, or the assistant fills gaps with guesses. Notes jotted after a supplier call might read: "villa hotel above the beach, 12 rooms, owner [first name], pool small but fine, 15 min walk down to sand, steep back up, dinner only Thu-Sat, good for couples, bring cash for the beach bar." An illustrative first attempt at the block:
BLOCK: Hotel - [villa hotel name], above the beach
Type: 12 rooms, boutique. Sea-view rooms: yes.
Best for: couples. Not for: families with young children.
Transfer from airport: approx. 45 min.
Dinner: on-site restaurant, Thu-Sat.
Our notes: small pool; 15-minute walk to the beach, steep return.
Last verified: [today]Three lines are the assistant's inventions: "sea-view rooms: yes", "not for families with young children" and the 45-minute transfer. None was in the notes, and all three are the kind of thing a client later quotes back at you. Add to the prompt: "Write UNKNOWN for any field the notes don't cover," and the block comes back with three gaps you know to fill on the next call, rather than three plausible errors. The beach-bar cash tip was dropped too; tell it to keep every practical detail in "our notes".
Timings: a ten-night draft, before and after
The table below shows illustrative timings for the honeymoon above, for an agent who already has a block library for the destination. Your figures will differ by destination and complexity.
| Stage | Without AI | With brief, blocks and AI |
|---|---|---|
| Write up the brief | 10 min (informal notes) | 5 to 10 min (fixed fields) |
| Sequence the route and hotels | 30 min | 2 min (AI draft) |
| Write day-by-day text | 50 min | included in draft |
| Fill gaps and verify timings | 20 min | 20 to 25 min |
| Format for the client | 20 min | 5 min |
| Total | about 2 hours 10 min | about 35 to 45 min |
When the client asks for changes
Most itineraries go through at least one round of "could we…". Changes are where AI saves time a second time, and where it can quietly break things. The client replies: "Love it. Could we swap the cooking class for a wine tasting, and add a night at the beach?" Asked simply to "make these changes", an assistant will usually make them and leave the rest untouched, including the knock-on effects. A better revision prompt:
Apply ONLY these changes to the itinerary: [paste the client's request].
Use the block library for any new item. Then list every knock-on
effect: dates that move, transfers that change, hotel nights that
change, anything that now breaks a rule in the brief, and the change
to land cost using the prices I give you below.An illustrative knock-on list for this request: the extra beach night pushes every later day back by one, so the flight home no longer matches unless the old-town stay shrinks by a night; the wine tasting in the library runs Tuesday to Saturday, and the new day 5 is a Monday; the extra beach night at $320 takes the land cost from $8,750 to $9,070, just over the $9,000 budget. That last line is a quick sum you could do yourself, but having all three on one list before you reply to the client is what saves the second phone call. You decide what to offer (drop a night in the old town and stay on budget, or explain the $70 over), and the assistant redrafts only the affected days.
Notice that verification barely shrinks. The saving comes from the writing and sequencing, which is why the block library matters: it is what makes the draft trustworthy enough that verification stays at twenty minutes rather than an hour. The first trip to a new destination, without blocks, won't see much of this saving at all.
A walking specialist and a family-holiday agency, compared
A one-person agency specialising in walking holidays has a small number of routes it sells again and again. Its blocks are stages of each walk (distance, ascent, luggage transfer, the inn at the end), and the AI's main job is assembling a route to the client's fitness and number of days, then writing each day's description. Drafts come out close to final because the blocks cover nearly everything.
The trap for this agency is arithmetic, not invention. A route with eight stages booked as a seven-day holiday forces the assistant to merge two stages, and it will: an illustrative draft combined a 13 km stage and a 14 km stage into one 27 km day with 900 m of ascent, for a client who had described herself as "a steady walker, happy with 15 km". The fix is two rules in the prompt: "never combine stages" and "if the days don't fit, write [NEEDS DECISION: drop a stage, add a day or arrange a transfer]". The agent then offers the client a choice, which is a better conversation than a sore-legged complaint.
An agency selling tailor-made family holidays across many destinations gets less from the library at first, because each trip draws on a wider range of suppliers. Its quicker win is the brief and the formatting stage: consistent briefs mean fewer follow-up calls with the parents, and AI-formatted documents look the same whoever in the team prepared them. The library builds up trip by trip.
A mistake worth planning for
One failure is specific to reusing blocks. An agent's library says a boat transfer between two islands takes 90 minutes. The operator changed its timetable for the season, and the crossing now runs only on alternate days. The AI, faithfully following the block, schedules the crossing on a day it doesn't run. The client discovers this at the quayside.
The fix is the "last verified" line on every block and a habit of re-checking any block older than one season before it goes into a client document. You can even ask the assistant to list every block used in a draft with its verification date, and to flag any older than six months. It turns a vague worry into a two-minute checklist:
Blocks used in draft H-0412:
Hotel, old town ............ verified March ....... OK
Private transfer to coast .. verified April ....... OK
Beach hotel ................ verified January ..... CHECK (8 months)
Island ferry ............... verified last October CHECK (11 months)
Cooking class, old town .... verified June ........ OKTwo phone calls or emails, and the ferry problem above is found in the office rather than at the quay.
Keeping client details out of the wrong place
Itineraries contain personal data: names, dates of travel, sometimes passport numbers and medical notes. Keep passport details and anything medical out of AI prompts entirely; the draft doesn't need them. Refer to clients by booking reference in the brief. If you use a consumer plan, switch off model training in its privacy settings; business plans such as ChatGPT Business and Claude Team don't train on your content by default. For faster replies at the enquiry stage, before any itinerary exists, see how a one-person travel agency can reply to enquiries instantly, and for the tools themselves, the best AI tools for travel agents.
Drafting itineraries with AI: follow-up questions
Should I tell clients that AI helped draft their itinerary?
You don't need to label every document, but don't hide it if asked. What clients pay for is your judgement, supplier knowledge and the checks you make, and those are still yours. If your terms mention how you prepare itineraries, a plain line saying you use software to draft and a person checks every detail is honest and usually reassuring.
Can I paste supplier rates and commission details into a chat assistant?
Treat them as confidential. On a personal plan, switch off the setting that allows your chats to be used for model training, or use a business plan where that is off by default. Better still, keep net rates and commission out of the drafting prompt entirely; the AI doesn't need them to write the day-by-day plan.
Is a dedicated itinerary platform better than ChatGPT or Claude?
Itinerary platforms are better at presentation, client-facing links, pricing and supplier data. General chat assistants are better at turning a messy brief into prose quickly. Many agents use both: the assistant for the first draft and the platform for the final document. Check whether your platform has added its own AI drafting before paying for two tools.
How do I stop the AI inventing hotels or tours?
Tell it to use only the named suppliers and experiences in your block library and to write [NEEDS OPTION] where nothing fits. Then check every proper noun in the draft against a supplier record. Invented names are much less common when the model is choosing from your list rather than from its general knowledge.
Further reads
- How Small Tour Operators Use AI for Bookings and Itineraries — The tour operator's version, with bookings added.
- How to Catch Outdated Information in AI Answers — Spot stale opening times and prices before clients do.
- How to Build a Company Knowledge Base AI Can Answer From — How to structure the block library as a knowledge base.
- Claude vs ChatGPT for Business Writing, Proposals and Reports — Which assistant writes better client-facing prose.
- How to Set Up Human Review for AI Work Without Slowing Down — Keep checks fast without skipping them.
- Reusing Past Client Work With AI Without Leaking Client Data — Reusing past itineraries without exposing client details.
- What AI Costs a Small Travel Agency and What It Saves — The monthly AI bill for a one-to-three adviser travel agency, set against a task-by-task ledger of the hours it can give back.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI and Anthropic help pages and plan pages on Projects and data use; vendor pages for itinerary software.