AI-driven discovery is creating a two-stage customer journey: an assistant first narrows the choices, then a referred visitor decides whether a website confirms the recommendation. The available reporting suggests that these stages are closely connected, but they should not be measured as one channel.
A product’s inclusion in an AI answer can change when web search is enabled, while the people who click through may behave differently from conventional visitors. Understanding both effects helps brands distinguish recommendation visibility from referral performance.
Key takeaways
- AI recommendation visibility can be highly variable: one reported ChatGPT study found that enabling search changed the products appearing in 80.2% of responses.
- AI referrals can bring unusually engaged visitors without guaranteeing stronger conversion. Adobe’s reported travel data showed more time on site and lower bounce rates, but a remaining conversion deficit.
- Category context matters. The same Adobe reporting found that AI-referred retail visitors converted substantially better than non-AI traffic, in contrast with travel.
- Readable, well-structured content may support discovery, but the cited evidence does not prove that improving AI readability directly causes more recommendations or sales.
Recommendation visibility depends on how the AI gathers evidence
An AI assistant does not necessarily produce a stable shortlist from a fixed body of knowledge. A study by Visibility Labs founder and CEO Jeff Oxford, summarized in the second source, ran 1,000 product-recommendation prompts ten times with search enabled and ten times without it, producing 20,000 interactions. Only 19.8% of products suggested without search reappeared when search was active. In other words, the retrieval method altered much more than the wording of the answer; it changed the choice set presented to users.
The most frequently suggested products were not insulated from that change. Of the products consistently recommended in search-disabled responses, the source reported that only 15.8% appeared after search was enabled. Search-enabled answers were also somewhat narrower, averaging 5.2 products per response compared with 6.2 without search. Across ten runs of each prompt, search produced an average of 19 unique products, versus 21.8 without it.
This volatility complicates the idea of a single, permanent AI ranking. A brand can be prominent in an assistant’s model-based answer and absent when the assistant consults the web, or vice versa. Visibility therefore needs to be evaluated across repeated prompts and different answer modes rather than inferred from one favorable result.
The study also found a reported Pearson correlation of 0.4 between how often products appeared in cited sources and how frequently they were recommended. That is useful directional evidence, but the observational design did not establish that source mentions caused inclusion. Citations may reflect broader web prominence, product suitability, accessible information or several factors operating together.
Referral quality reveals intent after the recommendation
The first source, reporting Adobe data, examines what happens after an AI user reaches a website. It said AI-driven traffic to U.S. travel sites increased 194% year over year in May 2026 and 2,215% from the beginning of Adobe’s monitoring in October 2024. The research drew on more than 8 million visits to U.S. travel sites and a March survey of more than 5,000 U.S. consumers.
These visitors displayed stronger engagement than non-AI visitors: Adobe reportedly measured 70% more time per visit, a 41% lower bounce rate and 21% higher engagement. The source interpreted the pattern as consistent with more deliberate, higher-intent browsing. That interpretation is plausible because an assistant can help a traveler compare destinations, hotel features, itineraries and promotions before the click, leaving the destination site to validate details or support a booking.
Engagement did not translate into an immediate travel conversion advantage. AI-referred visitors converted 28% less often than non-AI visitors, although the source said that gap had narrowed by nearly 70% since October 2024. Travel decisions can involve additional comparison and coordination, so time on site should not be treated as a substitute for completed transactions.
Retail produced a different outcome in the same Adobe reporting. AI-driven visits to U.S. retail sites rose 138% year over year in May and 1,324% from October 2024. AI-referred retail visitors converted 54% better than non-AI visitors, reversing the earlier pattern described by the source, when their conversion rate had been nearly half as high. Adobe’s retail analysis covered more than 1 trillion visits and over 100 million SKUs.
The contrast is important: AI referral traffic is not inherently high- or low-converting. Its commercial value depends on the category, the decision cycle and what remains unresolved when the visitor arrives. The recommendation stage may substantially reduce uncertainty for a specifications-led retail purchase while leaving a traveler with dates, availability, policies and other booking details still to settle.
Readable content links discovery with the landing experience
The two reports meet at content accessibility. The product study indicates that activating web search can substantially reshape recommendations and that cited-source mentions have a modest association with product visibility. Adobe’s travel analysis, meanwhile, suggests that a meaningful share of website content cannot be processed effectively by AI systems. Together, they point to an operational dependency: useful information must be available to the system before it can help form or substantiate a recommendation.
Using its AI Content Visibility Checker, Adobe reportedly found that hotel homepages had 63% AI readability and car-rental homepages 59%. Product pages scored higher, at 73% for hotels and 71% for car rentals. Even so, the source said more than one-third of the content on leading travel pages remained unreadable to AI systems.
Performance also varied by page type and sector. Hotels led in areas including destination guides, activities, search results, customer service and promotions. Car-rental companies performed best on FAQ pages, while cruise companies led in blog and news content. Airlines trailed the other major travel segments across the page types Adobe assessed. In retail, cosmetics and electronics benefited from detailed material such as ingredients, tutorials, specifications and how-to information, whereas grocery and furniture lagged.
These findings do not justify writing pages solely for machines. They support a more durable principle: important facts should be explicit, consistently named and placed in accessible page content. Detailed descriptions, amenities, specifications, policies and practical guidance can serve an assistant’s evidence gathering while also helping the referred visitor verify the recommendation.
Measurement must connect exposure, visits and outcomes
A useful measurement model separates three questions. First, how often does the brand or product appear across repeated recommendation prompts, with and without search? Second, which cited pages and on-site facts are associated with those appearances? Third, what do referred visitors do after arrival, including engagement, progression and conversion?
Each layer prevents a misleading conclusion. A single recommendation screenshot cannot establish durable visibility. A citation does not prove that the cited mention caused a recommendation. Strong engagement does not necessarily mean strong conversion, as the travel results demonstrate. Conversely, a lower volume of AI referrals may still be commercially meaningful when visitors arrive with a well-defined need, as the retail results suggest.
The next competitive advantage is likely to come from joining these measurements rather than optimizing them independently. Brands that monitor recommendation variability, expose decision-critical information and evaluate post-click behavior by category will be better positioned to learn whether AI is merely mentioning them or delivering customers who can act.
References
- Search Engine Land — AI Referrals Dramatically Boost Travel Site Engagement
- Search Engine Land — Discover How Search Transforms ChatGPT’s Recommendations
Leave a Reply