Which Websites Does AI Cite When Recommending Things to Do in a Destination?

AI systems mostly cite travel media, local businesses, Reddit and review sites, not the destination's own website. When Drifter AI asked five AI systems 32 questions about La Jolla on October 18, 2026, their 158 answers cited 321 different websites, and lajolla.travel was cited in only 30 of them. A local blog, Reddit, Tripadvisor, Wikipedia and the San Diego Tourism Authority each turned up in more.

The questions were the kind a visitor types before a trip: where should I eat in La Jolla, what is happening in La Jolla this weekend, is it better to stay in La Jolla or Coronado, is it safe to swim there. Each one went to OpenAI, Claude, Gemini and Perplexity through their search-enabled models, and to Google's AI Overviews. We kept every link each answer showed as a source. 140 of the 158 answers cited at least one website, 1,816 citations in all.

One disclosure before the numbers. lajolla.travel is the La Jolla site Drifter's trip planner runs on, and we built its pages. It stands in for the destination's own site here, so read anything about what worked on it as our own test, not a neutral sample.

Which kinds of websites does AI cite most?

Travel media and local businesses show up in the most answers, well ahead of booking sites and the destination's own site. Here is how many of the 158 answers cited each kind of site at least once:

• Travel media and local blogs: 114 answers, 82 different sites

• Hotels, restaurants, attractions and tourism boards: 102 answers, 81 sites

• Social sites, almost all Reddit and YouTube: 65 answers, 10 sites

• Booking and review sites such as Tripadvisor, OpenTable and Expedia: 62 answers, 30 sites

• Reference and government sites such as Wikipedia and the City of San Diego: 59 answers, 40 sites

• lajolla.travel: 30 answers

The rest went to sites the analysis couldn't put in a category.

The single site cited in the most answers was La Jolla Mom, a local blog, in 42. Then Reddit in 40, Tripadvisor in 36, and Wikipedia and sandiego.org in 35 each. My read is that engines cite the page that answers the exact question in plain words. A local writer has a page called something like the best Mexican food in La Jolla. A destination site usually has a dining category page. LaJolla.com, another local guide, has a page called "Italian Restaurants in La Jolla: The Best Ones" and changes the month in its title as it updates it. The May, August and September versions all turned up as sources.

Does the answer depend on the kind of question?

It depends on the question a lot, and this is the part an overall number hides. The destination's own site did well on factual planning questions and badly on taste and timing. Here is what came up most for each kind of question, counted in answers:

• When to go (5 answers): lajolla.travel and a Hilton page were cited in 4 each. That was the destination site's best result on any question.

• Where to stay (40 answers): La Jolla Mom in 21, Expedia in 12, and lajolla.travel, La Valencia Hotel's own site and Tripadvisor in 11 each.

• Safety, swimming, getting around, accessibility and families (40 answers): Tripadvisor in 14, Reddit and the local guide lajolla.ca in 13 each, the City of San Diego and sandiego.org in 12 each, lajolla.travel in 9.

• Comparisons with other coastal towns (23 answers): sandiego.org and Visit California in 7 each. The other town's board turned up too, with Visit Santa Barbara's site in 4.

• Where to eat (40 answers): the Michelin Guide, Reddit and Times of San Diego in 10 each, then Eater, OpenTable, Tripadvisor, sandiego.org and the local guide LaJolla.com in 9 each. lajolla.travel was cited in 3.

• What's on this weekend (10 answers): La Jolla Playhouse's own site in 5. lajolla.travel in none.

The pattern makes sense once you think about what the engine is looking for. "When should I visit" has a factual answer, and an official page that states it plainly is a safe source. "Where should I eat" is a matter of taste, so engines go to critics, reviewers and locals arguing on Reddit. "What's on this weekend" needs something current, so they go to the venue that is actually putting on the show.

That also means a destination doesn't need to win every question with its own pages. On lodging, La Valencia's own site was cited as often as the destination's. On events, the Playhouse was the top source. Those are partners telling the destination's story, and a tourism board can help them do it well.

Does it depend on which AI you ask?

It depends on the engine far more than any average shows. OpenAI's answers carried 1,014 of the 1,816 citations, more than the other four systems put together, and every one of its 32 answers cited something. Google's AI Overviews showed a source in only 19 of its 32 answers, 56 citations in total.

Reddit is the clearest example. OpenAI cited it in 31 of 32 answers, 271 times. Claude didn't cite it once. Gemini cited it in 6 answers, AI Overviews in 2, Perplexity in 1.

The same split shows up for the destination's own site. OpenAI cited lajolla.travel in 21 of its 32 answers. The other four systems cited it in 9 of their 126: Perplexity in 5, Gemini in 2, Claude in 1 and AI Overviews in 1. So a single "AI citation share" for a destination blends engines that read the web very differently. It's worth checking them one at a time.

Which pages on a destination's site get cited?

Pages written around one traveler question did best outside OpenAI. lajolla.travel was cited from 17 different pages. Of the 9 answers from Claude, Gemini, Perplexity and AI Overviews that cited it, 6 pointed to a page built around a single question, like the one on the best time to visit La Jolla or the one on whether it is safe to swim there. The best-time page was cited by Claude, Gemini and Perplexity for the question it answers.

List pages did their work with OpenAI and Perplexity. The best-hotels page was cited in 9 answers, all from those two, for questions like where to stay and what the nicest hotel is.

My read is that these engines search the web with words close to the traveler's own, then cite pages that answer that search outright. A page whose title is the question, and whose first lines are the answer, is easy for them to find and safe to quote. A general "Plan your visit" page has to be read and interpreted first. Because we built these pages ourselves, take this as one site's result rather than a rule.

Where does a destination's official site fit?

A destination's own site is one source among many, and for a place inside a bigger region the regional board often stands in for it. The San Diego Tourism Authority's sandiego.org was cited in 35 answers about La Jolla, and Visit California in 15. That's good news for the region and a reasonable outcome for the town, because those are official sources telling the story too.

I think the more useful finding is how wide the field is. 321 sites for 32 questions means no single page decides the answer. The answer gets assembled from hotels' own pages, restaurant listings, local writers, the regional board and Reddit threads. A destination can't write all of those, but it can shape a lot of them.

What should a tourism board do with this?

Make your own list of the sites AI cites for your place, sorted by the kind of question, and work on the ones you can influence first:

• Pick ten questions visitors actually ask about your destination, and cover the different kinds: when to go, where to stay, where to eat, what's on, and how to get around.

• Ask each one in at least two AI systems, for example ChatGPT and Perplexity, since they cite very different sites.

• Write down every website each answer cites, next to the question it answered.

• Sort the list into three piles: sites you run, sites you can influence, and sites you can't.

• Start with the middle pile. That's your partners' own pages, the regional and state tourism boards, local writers and listings. Check that hours, prices and facts are right there, and give them what they're missing.

Then match the fix to the kind of question. For factual planning questions, give each one its own page on your site that answers it in the visitor's words, because that is where an official source wins. For events, make sure the venues' own event pages and your calendar are current and readable without a login or a PDF. For restaurants, put your energy into partners' own pages and listings rather than another roundup of your own, since engines lean on critics and diners there.

Doing this with repeats and counts for each engine is what our guide to measuring AI visibility walks through.

What can't this data tell you?

This is one destination, one week and five AI systems, so treat it as a sample, not a law. Every question named La Jolla, which shows how engines describe a place people already have in mind, not whether they suggest it to someone who hasn't heard of it. A citation is the link an engine showed next to its answer. It tells you which sources the engine chose to show, not which ones shaped what it said. Running the same questions again in a month is how you find out which of these patterns hold for your own destination.