Conclusion and key takeaways

To get booked, your business has to be known and knowable

The platforms will keep changing. Google will add another answer format, ChatGPT will change how it searches, and someone will invent a new acronym and sell a checklist before lunch. You cannot rebuild your strategy every time that happens.

The evidence in this report points to a simpler model: to get booked, your business has to be known, and it has to be knowable.

The searches that used to introduce operators to new customers are producing fewer website visits. Search visits per site in the uninterrupted operator cohort fell roughly 41% in 2 years, while AI can now search, read, and compare sources without sending a visit that appears in ordinary analytics.

The customer did not disappear. More of the research now happens before the customer reaches your website. That makes 2 conditions especially important: AI has to include your business among the options, then both AI and the customer need enough current information to choose it.

Your business has to enter the conversation before it can win the booking

Known does not mean famous. It means the web connects your business with the right place, activity, and customer need often enough for AI to consider it. Visit Morgan Hill's feature about Morgan Hill Wine Trolley is 1 example: an independent local source connecting an operator to wine tasting in a specific destination.

Google rankings still matter because pages near the top have a better chance of earning an AI citation. Google Business Profiles matter too. When Google's AI mentioned an operator in our study, it showed the operator's Google Business Profile 96% of the time.

But the strongest pattern was broader than either one. Businesses mentioned across more independent websites generally appeared in more AI answers. Most of the pages AI cited did not rank in Google's top 20, almost 87% of article citations went to articles published by other people, and most identifiable social citations came from accounts the operator did not own.

The public record is bigger than your website. Your site can support it, but it cannot create the whole record by itself.

Once AI finds you, it still needs a reason to choose you

When a customer asks for general ideas, AI leans heavily on lists and guides. Best-of lists receive 34% of citations during local discovery. Once the customer names an activity, product pages rise from 2.4% to 19.8% of citations and homepages rise from 4.8% to 17.3%. That is a page-type shift across all publishers, not a claim that the tracked operator owns every business page.

When the tracked operator's own site does enter a where-to-book answer, its homepage and product pages capture 80.6% of its citation share. Those are the pages most likely to explain whether a trip fits, what it includes, where it starts, and how to book it.

The AI Start Here experiment was the strongest knowable result in the report. Operators that built the page appeared in 53% of tracked questions, compared with 24% for operators that did nothing. After adjusting for site strength, the estimated lift tied to the page was 7 percentage points. Thirteen of the 14 pages were cited directly, showing that AI used them. That does not prove the pages caused a recommendation or a booking.

Your website now has 3 jobs

Fewer customers may use your website for the entire journey from first idea to booking. That does not make the website less important. AI can read it before the customer visits and use its details while comparing options.

The website now has to do 3 things at once:

1. Give AI a reliable source of current business facts. Publish the products, prices, schedules, meeting points, policies, limitations, and differences that determine fit.
2. Let the customer verify the recommendation. When someone clicks through, the page should confirm what AI said and give them enough context to trust it.
3. Make the booking easy and safe. Show current availability, explain what happens next, and make the direct-booking path clear.

Those jobs belong across the site. The homepage explains the business. Product pages explain whether a specific experience fits this customer. Policies and FAQs handle the questions that can stop a booking. The AI Start Here page is 1 useful reference inside that system, not a substitute for it.

The useful details are usually the ordinary ones operators already explain on the phone: whether a trip is private, how many people can join, what is included, where it starts, what happens in bad weather, and how 2 similar options differ. The work is getting those facts out of phone calls, messages, and people's heads and onto pages that stay current.

Use 2 questions to judge every new tactic

When a new AI tactic appears, ask what job it performs:

1. Does this help the right customers and systems find the business?
2. Does this help them understand, trust, and book it?

An independent article can help people and AI find you. A complete product page can help them understand you. An accurate Google Business Profile can support both, while a current booking page turns that confidence into action. If a tactic does neither job, it deserves a much harder look.

The 2 questions do not guarantee a recommendation or a booking. They give you a way to judge the next tactic without rebuilding your strategy around the latest tool.

Analysis reference: known and knowable

• What changed: Search visits per site fell roughly 41% across the uninterrupted operator cohort, while direct AI referral traffic remained too small to represent the research AI performs before a customer visits.
• Known across the web: Google rankings improve a page's chance of earning a citation. When Google's AI mentioned an operator, the Google Business Profile appeared 96% of the time. Independent website mentions had a much stronger relationship with AI visibility than backlink count, while review count after adjusting for business size and owned-social metrics were essentially flat.
• Knowable on your own terms: Product and homepage citations rose sharply once a customer named an activity. When the tracked operator's site was cited on a where-to-book question, its homepage and product pages captured 80.6% of its citation share. AI Start Here page builders appeared in 53% of tracked questions versus 24% for comparison operators, with a 7-point adjusted estimate.
• What did not provide a shortcut: More blog posts did not improve visibility, schema behaved like useful hygiene rather than a demonstrated growth lever, and the content-depth analysis did not produce a defensible word-count target.
• Important limit: The study measures citations, mentions, rankings, traffic, and observed website changes. It does not prove that completing a fixed checklist causes an AI recommendation or a booking. Known and knowable is Gondola's strategic interpretation of the measured findings.

Your customers still need a reason to choose you

AI can combine what other people say with the current facts you publish. It cannot manufacture independent trust, witness the trip you ran this morning, or fill in an important detail you never put online.

Outside sources help your business enter the conversation. Useful, current pages help AI answer the specific question and give the customer a place they can trust when they are ready to book.

Your customer may ask AI to do the research. They still need a reason to choose you.

---

How we did the research

The short version: we tracked 530 real customer questions about 34 independent tour and activity operators for a little more than 6 months. We recorded which businesses appeared, which pages AI cited, how Google traffic changed, what the wider web said about each business, and what happened when websites changed.

The report measures answers, citations, mentions, rankings, and traffic. It does not measure every AI question in the market or prove which source created a booking. If you want the exact study design, tools, sample sizes, limits, and Gondola disclosure, they are all below.

Study design, tools, data, limits, and Gondola disclosure

The questions, businesses, and AI systems

The main AI answer and citation study covers 34 independent tour and activity operators that use Gondola. They include tours, rentals, attractions, and other bookable experiences across North America and the Caribbean.

From December 1, 2025 through June 9, 2026, Peec.ai monitored 530 distinct questions about those businesses across 5 AI platforms:

• ChatGPT
• Gemini
• Google AI Overviews
• Google AI Mode
• Perplexity

The set includes broad location questions, activity and offer questions, and detailed questions based on what each business actually sells. Prompt text is available for 510 of the 530 prompt IDs. Those 510 prompts cover 99.94% of all citations in the frozen source.

Peec recorded 406,373 AI responses across those questions and platforms. We round that to “more than 400,000 answers” in the report.

The frozen citation source contains 3,388,833 citations across the 530 prompt IDs. These are source citations, not website visits, bookings, or unique pages. The same URL can be cited many times and can appear in more than 1 question.

What we mean by AI visibility

AI visibility is how often a tracked business appears in the recorded answers to its fixed set of questions.

That makes it useful for comparing operators, comparing groups, and watching what happens after a change.

It does not represent every question real customers ask. It does not measure the total volume of AI searches in the market. A different question set could produce different percentages.

The other data behind the report

The AI answer and citation data sits alongside several other sources:

• Google Search Console for search queries and clicks
• First party website analytics from Umami
• Ahrefs for rankings, backlinks, referring domains, and modeled search data
• Google search results collected for specific comparison analyses
• Public websites, articles, platform listings, reviews, and social sources cited in the tracked answers
• Gondola website and server data used to validate pages and website changes

The broader website experiment contains 35 sites. 1 control site has website data but no Peec visibility record, so the AI analysis uses the other 34. Most of the remaining analyses use between 29 and 34 operators after removing businesses with missing data, closures, or a mismatch between the sources. Narrower comparisons use 14 to 18 operators. We show the smaller sample next to the claim when it materially changes how strongly the result should be read.

Study limits

• The question set leans toward finding and comparing experiences. It is not a sample of every real-world question asked across all AI systems.
• A citation shows which page supports an answer. It does not prove that the page caused the recommendation or created a booking.
• Most relationships in the report are observational. When we ran an experiment or examined a business change, we say so and preserve the limits of the comparison.
• Some null findings may reflect the sample size. “We do not detect a relationship” does not mean the factor can never matter for any business.
• The platforms changed during the study and will keep changing after publication.
• Google Search Console and website analytics measure different things, so their traffic totals should not be expected to match exactly.
• Direct AI referral traffic is difficult to isolate. The roughly 0.6% figure in this report should be read as an upper bound, not a precise count of every visit from a chat tool.

A note about Gondola

Gondola funded and conducted this research. Every operator in the main study cohort is a Gondola customer, and several analyses examine pages, websites, or interventions Gondola created.

That relationship is a strength and a source of potential bias.

It is a strength because a standardized network of real small-business websites lets us compare changes that would be nearly impossible to observe across unrelated enterprise brands.

It is a source of potential bias because Gondola sells websites and will use these findings to shape its product. We have tried to handle that plainly: show the results that contradict our assumptions, separate raw comparisons from adjusted estimates, label our interpretations, and avoid claiming that Gondola already solves every problem in the report.

The research points toward a product direction. It does not turn that direction into proof.

Peec.ai provided the AI monitoring platform that made daily tracking across 5 surfaces possible. Gondola performed the analysis and remains responsible for the claims in this report.