Vitaliy Levit 50 minute read
What has changed
Your customers still research their options. AI now does much more of that work for them.
Your customers are asking AI which businesses to choose
The person looking for your business may still start in Google. That part has not disappeared.
What changed is the work that happens next.
To keep this report concrete, we are going to follow 1 real example. Morgan Hill Wine Trolley is 1 of the operators in our study, and it appears in roughly 61% of the AI answers we tracked for its business. 1 of the questions came from a group planning a birthday:
Prompt sent to AIMy friends and I want to celebrate a birthday with wine tasting near San Jose. What are the best tours in Morgan Hill?
A few years ago, that question would have created a pile of work for the customer. Search Google. Open a dozen tabs. Read local guides. Check Yelp and Tripadvisor. Visit operator websites. Compare public and private tours, prices, pickup points, winery stops, group policies, and availability.
Now the customer can ask the whole question at once. ChatGPT, Gemini, Perplexity, Google AI Overviews, or Google AI Mode can search, read, and compare the options. The customer gets a short list with reasons and sources, then decides what to check or book.
The research still happens. The customer just does less of it by hand.
1 question can replace a whole afternoon of clicking
Google explains that AI Mode can split 1 question into smaller searches, gather information from different websites, and assemble the findings into 1 response. Google calls this “query fan out.” Google explains how AI search gathers information.
The customer no longer needs to search “wine tours near San Jose,” then “Morgan Hill wine trolley,” then “private wine tours,” and then open 8 websites to compare the details. 1 question can trigger all of that research.
The answer will not always be good. AI can miss businesses, mix up details, and rely on information that is old or wrong. But the basic capability is already here. The customer asks once. The machine does the searching, reading, and sorting.
And this can happen inside the same Google your customers already use. They do not have to become ChatGPT users for their research process to change.
Different questions pull from different parts of the web
AI answers usually show the pages they used as sources. We call those pages citations.
The birthday question above is already close to a booking. That matters. The 3 pages cited most often for that question all belong to Morgan Hill Wine Trolley:
- Homepage: https://mhwinetrolley.com/
- Public tour page: https://mhwinetrolley.com/category/public-tours
- Private group experience: https://mhwinetrolley.com/product/private-wine-tasting-tour
That does not mean operator pages dominate every question. The source mix changes with what the customer asks.
A broad information question
Prompt sent to AIWhat is the Santa Clara Valley wine trail?
The leading sources come from outside the operator's website:
- Visit Gilroy's wine trail article: https://visitgilroy.com/blog/post/explore-the-santa-clara-valley-wine-trail
- County of Santa Clara: https://news.santaclaracounty.gov/news-release/do-you-know-way-santa-clara-valley-wine-trail
- Visit Gilroy's wine trail page: https://visitgilroy.com/trails/wine
A commercial comparison
Prompt sent to AIWhat are the best wine tasting tours in Morgan Hill?
Now the mix changes. The leading sources include:
- Morgan Hill Wine Trolley's homepage: https://mhwinetrolley.com/
- Visit Morgan Hill's winery guide: https://visitmorganhill.org/wine
- The operator's public tour page: https://mhwinetrolley.com/category/public-tours
A specific customer need
Prompt sent to AIMy friends and I want to celebrate a birthday with wine tasting near San Jose. What are the best tours in Morgan Hill?
The operator's own pages take the top 3 spots:
- Homepage: https://mhwinetrolley.com/
- Public tour page: https://mhwinetrolley.com/category/public-tours
- Private group experience: https://mhwinetrolley.com/product/private-wine-tasting-tour
The exact order can change as AI produces new answers.
The larger pattern is what matters. Broad questions lean on destination organizations, government pages, guides, and other outside sources.
As the question becomes more commercial and specific, the operator's own pages become more useful.
Same business. Different question. Different sources.
AI can assemble the old pile of browser tabs before the customer visits a single operator website. Which tabs it chooses depends on the job the customer is asking it to do.
Your customers aren't clicking through to your website like they used to
If direct bookings, website visits, or the number of customers saying “I found you on Google” feel softer than they used to, you're not alone.
Maybe your ranking slipped. Maybe a new competitor entered the market. Maybe demand changed. Maybe the economy, weather, or another factor outside your control changed with it. We cannot explain what happened to every individual business.
However, Gondola hosts and measures hundreds of operator websites... and when we zoom out across that network, the overall direction is clear.
Organic search traffic is down well over 40% in the last 2 years.
The large experience marketplaces move in the same direction. Viator, Klook, Tripadvisor, and Tiqets all sit well below their 2024 organic search peaks.
The biggest drop in clicks comes from top of funnel discovery terms
People who search for your exact business name still click about 12% more year over year.
People who are searching to learn about things to do at the top of funnel and discovery searches fell 48%. These include terms like “best wine tasting tours in Morgan Hill” or “things to do near San Jose.”
Think about those searches through the Morgan Hill operator we are following:
- “Things to do near San Jose” comes from someone who may not know they want a wine tour at all.
- “Best wine tours near San Jose” comes from someone who knows the activity but not the operator.
- “Best private wine tours in Morgan Hill” comes from someone who has narrowed the area and the kind of experience but not the business.
- “Morgan Hill Wine Trolley tours” comes from someone who already knows the business.
- “Morgan Hill Wine Trolley private tours” comes from someone who knows the name and wants a specific experience.
The further that customer sits from knowing your name, the more likely the click has disappeared.
That makes sense when you realize that many AI recommendations do not send the customer directly to an operator website. The customer may hear your name from AI, then search for you by name when they are ready to look closer.
This is the point you need to pay attention to. The customer who already knows you can still get back to you. The harder problem is getting introduced to the customer who does not.
AI is contributing to this shift, but it's not solely responsible
We cannot look at every missing Google click and say AI took it. The data does not support that claim.
Google results now include Maps, ads, booking sites, videos, forums, product carousels, and AI answers. Demand changes. Competitors improve. Rankings move. Algorithms change. Any of those things can affect traffic.
What we can say is simpler. A customer can now complete much more of the research without visiting an operator website. AI did not cause every lost click, but it makes the website visit less necessary during research.
Organic clicks to your website show up in your Google Analytics. AI recommendations do not.
If someone clicks from Google to your website, your analytics records the visit. If an AI system reads your website, uses the information in its answer, and the customer never clicks, your analytics records nothing.
Both events can influence a booking. Only 1 looks like traffic.
This is why direct AI referrals create such a misleading picture. The visible traffic remains tiny. The invisible research can be much larger.
We cannot see every choice a customer makes inside an AI answer. We can measure whether a business appears, which pages AI cites, and which sources repeatedly support the recommendation. That is the evidence behind the rest of this report.
Analysis reference: traffic change
- 2-year cohort: 52 operator websites with uninterrupted analytics in the same comparison months of 2024, 2025, and 2026. Mean Google search visits per site fell about 41%.
- Search Console cohort: 131 sites with comparable data from January through April in both 2025 and 2026. Total clicks fell 25%.
- 12-month site distribution: 126 active sites with at least 100 Google visits in the earlier period, comparing May 2025 through April 2026 with the prior 12 months. Retired sites and sites with no recent data were excluded.
- 3-year sensitivity cohort: 20 sites with complete data from May 2023 through April 2026. Mean search visits per site ended 35% below the starting period.
- Search-type cohort: 94 sites with 16 complete months of Search Console data. The classification uses the top queries Google reported, not every long-tail query, so the contrast between branded and discovery searches is more reliable than either exact decline estimate.
- Interpretation limit: these observational comparisons measure changes in clicks. They do not isolate AI as the cause.
Your website's job description has changed
The website used to handle almost every stage. It helped the customer discover the business, understand the experience, compare options, build trust, and book.
Now AI may handle part of the discovery and comparison before the customer arrives. The website gets the customer later, when they want to check the recommendation, understand the details, see whether the business feels legitimate, check availability, and book.
That does not make the website less important. It gives the website 2 clearer jobs.
First, give AI accurate information it can use. Second, give the customer a trustworthy place to verify the recommendation and act.
But the website cannot do either job if the business never makes the short list.
So that is where we go next: what gets your business found when AI decides which operators belong in the answer?
Now — what do you actually do about it?
Vitaliy walks through the findings live and shows what to tune on your website first. Bring questions.
Save your seat