Twelve Questions DMOs Should Ask Before Buying an AI Trip Planner

Every AI trip planner demo follows the same script. Someone types “three days, two kids, we like hiking and tacos,” and a handsome itinerary blooms on the screen, complete with photos and a map. The room relaxes. Someone starts drafting the purchase order.

The demo answers a question you didn’t need answered: can software write a plausible trip? Of course it can. The questions that decide whether a planner helps or hurts your destination are harder to stage — whether it tells the truth on a Tuesday in March, whose data it learns from, what it does to your members’ businesses, and what remains when the contract ends.

What follows is the list we’d want on the table before signing. It’s written for destination organizations, but most of it applies to any tourism business buying a planner it will put its name on.

Updated August 25, 2026, with figures from a new traveler and DMO survey.

Why the pressure, and the risk, are real

The study, from Mindtrip with Sabre, Sojern and MMGY Travel Intelligence, surveyed 1,006 U.S. adults and 103 DMO leaders in April 2026. Its headline is that travelers are well ahead of the industry:

Travelers	Destination organizations
54% have already used AI to plan travel; 75% plan to	22% offer AI trip planning on their websites
40% have discovered a destination through AI	26% have a documented AI strategy
29% changed plans based on an AI recommendation	72% of leaders say organizations that don’t adopt AI within two years risk becoming irrelevant in destination discovery
63% would use an AI trip planner on a destination website	Only 20% rank itinerary building as a high-value capability

Read the last row twice. Most DMO leaders don’t think itinerary building is where the value is — yet itinerary building is exactly what the demo shows. We think the leaders are right. The value lies in what the itinerary does: sending visitors to quieter towns and quieter weeks, referring them to member businesses, answering accurately when plans change. The twelve questions below are built around that.

The other reading of the table: urgency is high, and only about a quarter of organizations have a written strategy. That is the classic setup for buying fast and regretting slowly.

Truth and grounding

1. Where does each recommendation come from, and can we see it?

Every venue, event and trail the planner mentions should trace to a record — your listings, your partner database, an official notice — with a last-verified date.

Good answer: one click shows the source record behind each suggestion. Red flag: “The model knows the region really well.”

2. When our data and the model disagree, which wins?

Large models carry years-old memories of your destination. Your data has to overrule them, and venues absent from your catalog should be flagged or left out.

Good answer: a documented precedence rule, demonstrated live. Red flag: “We trained it on your website.”

3. How do daily changes reach it?

Closures, fire restrictions, road conditions, events that move. Ask how long it takes from a staff member posting a closure to the planner reflecting it.

Good answer: minutes, through a feed or a simple console your team controls. Red flag: a monthly re-crawl.

Fit with your strategy

4. Does it build days that actually work?

Ask for a day in your shoulder season, for a family without a car. Check travel times, opening hours, seasonal closures and pacing. A planner that schedules five stops in three towns before lunch has no real routing underneath it.

Good answer: feasible days, with travel time shown between stops. Red flag: beautiful prose, impossible logistics.

5. Can we steer it toward our goals, and is the steering disclosed?

Dispersal, midweek visits, shoulder-season demand, smaller communities — these are legitimate goals. Paid placement disguised as advice is not.

Good answer: configurable, documented weights, with sponsored content clearly labeled. Red flag: “premium placement” sold to members inside the recommendations.

6. How are member businesses treated?

Your members will ask why they do or don’t appear. The rules should be explainable, and a small business with a thin website shouldn’t automatically lose to a large one with a marketing team.

Good answer: published inclusion and ranking criteria, and a way for members to correct their own data. Red flag: “The algorithm decides.”

Data and distribution

7. Who owns the conversations and preference signals?

What travelers ask a planner is the best demand research you will ever get. Confirm ownership in the contract and ask for an export sample now, not at renewal.

Good answer: a documented export in a standard format, on request. Red flag: dashboards only.

8. What is collected, what is kept, and for how long?

Trip planning reveals sensitive things: mobility, health, family composition, religion through dietary needs. Ask about consent, minors, retention, and whether conversations train the vendor’s models. Our responsible AI and traveler data governance reviews start with exactly this inventory.

Good answer: minimization by default, and retention limits you set. Red flag: “We anonymize everything,” with no detail.

9. Does our content work only inside your widget?

Sixty-three percent of travelers would use a planner on a destination website, but 40% have already discovered a destination through AI — mostly elsewhere, in assistants you don’t control. The investment should also make your destination legible to those assistants: structured, citable content and, increasingly, an agent-facing interface that you own.

Good answer: structured outputs you can publish and reuse. Red flag: everything lives inside the vendor’s domain.

Measurement and exit

10. What will we measure besides chats?

Conversation counts measure curiosity, not outcomes. Ask for partner referrals, saved plans, a dispersal index (the share of recommended visits outside your ten busiest sites), off-peak share, and accuracy on a fixed test set. If 29% of travelers change plans because of an AI recommendation, measure whether yours changes them in the direction you want.

Good answer: outcome metrics written into the contract and reported monthly. Red flag: “engagement” as the headline number.

11. How do you test quality before launch and after every model change?

Models get upgraded underneath products. Ask for a regression process run against your test questions, with results shared.

Good answer: a test set you co-own and a report after each change. Red flag: “We use the latest model, so it only gets better.”

12. What happens when it’s wrong, and when we leave?

Ask for the incident path (who you call, and how fast a wrong answer is corrected), a kill switch for a topic or for the whole planner, and what you take with you at exit: data, configuration, content and the test set.

Good answer: exit terms in the contract and a practice export in the first quarter. Red flag: “No customer has ever wanted to leave.”

A scorecard you can actually use

Score each question 0 (no answer, or a red flag), 1 (partial) or 2 (strong and demonstrated). Double the weight on questions 1 to 3 and on question 12, because truth and exit are where a bad choice costs most. The maximum is 32.

Here is how a comparison might go:

Area	Vendor A	Vendor B
Truth and grounding (Q1–3, double weight)	2, 0, 0 → 4	2, 2, 1 → 10
Strategy fit (Q4–6)	1, 2, 1 → 4	2, 1, 2 → 5
Data and distribution (Q7–9)	1, 1, 0 → 2	2, 2, 1 → 5
Measurement (Q10–11)	1, 1 → 2	2, 1 → 3
Exit (Q12, double weight)	0 → 0	2 → 4
Total (of 32)	12	27

Vendor A gave the better demo. Vendor B is the one that will still look good after a winter road closure. Set your floor before you see any demos — say, nothing below 20 and no zeros on questions 1 to 3 — so the shiniest presentation doesn’t move the bar.

Run the demo on your terms

Send vendors 30 prompts a week ahead, then run them live in the room:

• 10 ordinary requests — the trips your visitor center hears about every day.
• 10 seasonal and edge cases — a March weekend, a holiday when many businesses close, your biggest event week.
• 5 accessibility requests — step-free, low-exertion, sensory-friendly.
• 5 traps — a venue that closed last year, an event that moved, a road closed in winter, a non-member business, and a request the planner should politely decline.

Count phantom venues, impossible days and missed traps. That number will tell you more than any slide.

Start with the strategy, not the widget

Before the RFP, write one page: what the planner is for, which three metrics will judge it, which data you will own, and who on staff keeps that data true. A short AI strategy and discovery sprint can produce that page and the test set together.

When Nova built an agent-ready planner with a regional destination organization in the Mountain West, the test set turned out to be the most reused artifact of the whole project. It judged vendors first, then models, then every release that followed. It cost a few days to write, and it is the one thing on this list you can start this week. If you’re weighing the wider picture for your organization, our notes for destination organizations are a good next stop.