What Should a DMO Ask an AI Visibility Vendor Before Buying?

Before buying, ask an AI visibility vendor five things: whether it asks the same questions every week, whether you can see each engine on its own, whether questions that name you are kept apart from ones that don't, whether you can read every answer, and what your team should do next. The first matters most, because in Drifter AI's data about 1 in 8 discovery answers asked a week apart changed whether the destination was named at all.

That figure comes from 7,482 pairs of completed answers to the same question, from the same engine, asked about eight destinations Drifter AI monitors, with the later answer in each pair recorded between September 20 and November 1, 2026. Every question was a discovery question, the kind a traveler asks before naming a place. In 952 of those pairs the destination was in one week's answer and missing from the other. To tell that kind of movement from a real change, you need many questions asked the same way every week.

How much do AI answers about a destination change from week to week?

AI answers about a destination change enough that a single answer tells you what AI said once, and not much more. Of the 3,412 times one of the eight destinations was named, it was gone from the same engine's answer to the same question a week later 495 times. It was added to 457 other answers that had left it out the week before. So the overall number barely moved while the individual answers churned underneath it. The number to watch comes from a large set of questions asked the same way every week, read as a trend.

The pattern holds over longer stretches. Across 1,384 combinations of question and engine asked in at least four different weeks, about five weeks each on average, the destination was named every week in 390 and never in 600. Of the 784 combinations that named it at least once, about half (394) named it some weeks and not others.

Engine model changes don't explain it. Several engines did switch models during these weeks, but when the model stayed the same, 887 of 6,940 pairs (12.8%) still flipped.

Can you see each engine on its own?

A single headline number is useful only if you can open it and see each engine on its own, because they don't move together. In Drifter AI's data the week-to-week flip rate ran from 9.6% for OpenAI (168 of 1,750 pairs) through 11.6% for Perplexity, 12.1% for Gemini and 14.5% for Claude, to 21.9% for Google's AI Overviews (129 of 588). AI Overviews changed more than twice as often as OpenAI, which a headline number alone won't show you. When AI Overviews named a destination, the next week's Overview for the same question left it out 70 times out of 238. Drifter AI reports Google's AI Overviews beside its other engines and never blends them into Answer Share.

Engines also show very different amounts of evidence. Across 14,865 answers about 12 destinations Drifter AI monitors, recorded between September 15 and October 25, 2026, OpenAI's answers showed 27.0 sources on average, Perplexity's 14.6, AI Overviews 8.4, Claude 5.8 and Gemini 4.7. Ask the vendor to show which sources each engine cited, not only whether you were named.

Google is also changing how it shows sources. On October 15, 2026, it announced more inline links placed higher in AI Mode and AI Overviews, and a test of a section at the top of some answers for cases with "only one or two official places on the web" to complete a task. Ask any vendor whether it records Google's AI Overview for each question and shows which pages it cites.

Does it separate questions that name your destination from ones that don't?

Keep questions that name your destination apart from ones that don't, because the two get very different results. In those same 14,865 answers, official tourism board sites were cited in 39.5% of answers to questions that named the destination and 10.5% of the rest.

A traveler who asks about your place by name already has you in mind, and your site is the obvious source. A traveler who asks for "a walkable small city with good food" is still choosing, and that's where the shortlist gets written. A vendor that mixes the two into one score can make a destination look strong on the questions it was always going to win. Ask to see the question list split into the two groups, with results for each.

Can you see every answer behind the number?

You should be able to open any score and read the answers under it: the question, the engine, the date, the full text and the sources it cited. That's how your team checks the number and explains a change to your board. Ask whether every answer behind each number is there to read, and whether the questions stay the same from week to week so trends compare like with like. In Drifter AI that's Prompt Insights, which keeps every question, every answer word for word and the pages each one cited. We publish how we measure in our methodology, and any vendor should be able to show you theirs.

Does it tell your team what to do next?

Ask what next step comes with the numbers. A useful answer names the question, the page or partner listing that should answer it, and what that page needs to say. In Drifter AI those are the week's Actions, each with a ready draft, and Your published pages later checks whether AI answers cited the page. Ask how a vendor will show what changed after the work went live, measured on the same questions over the following weeks. Be careful with anyone who promises placement in AI answers. No vendor controls what an engine says; what a vendor can show is the same questions measured before and after.

What should you bring to a vendor demo?

This week, pick ten questions your visitors actually ask and ask your current or prospective vendor for the last four weeks of answers to each one, in each engine. Check how many times each question was asked, whether you can read the answers yourself, and how many of them changed from one week to the next. If the vendor can show you that, you have numbers worth building on. If it can only show one score from one run, treat that as a starting point and ask how it will build a baseline your board can track.