The question boards keep asking about AI visibility has an uncomfortable answer. Brands track how often they surface in ChatGPT and Gemini, then struggle to say what that presence buys them. The problem is not that AI visibility does nothing. It is that the payoff arrives days later, through channels standard attribution files under something else. The predominant journey leaves no click to track and no referral to attribute, so dashboards show almost nothing while real behaviour shifts underneath them.
New research from Similarweb puts numbers on the gap. The firm followed thousands of real user journeys across finance, travel and beauty, tracking people who asked ChatGPT an industry question and got a specific brand in the reply. Over the seven days that followed, those users were 2.5 times more likely to visit the recommended brand than its competitor.
The visit happens, the click does not
The mechanics explain why so much of this stays hidden. A user asks a question, reads the answer, and the session ends. There is no referral to log and no link to follow. When the visit lands days later, analytics record it as search or direct traffic, cut off from the conversation that prompted it.
Similarweb's survey of people who use both search and AI confirms the pattern. AI proved most useful at the start of the journey, with 35% naming it their tool for initial ideas against 13.6% for search engines. Search only caught up at the point of purchase, when users hunt for where to buy.
The effect runs both ways
The traffic swing was symmetrical, which strengthens the case that recommendations cause it. When ChatGPT named American Express in finance, 7.2% of users visited its site against 3.1% for Capital One. Flip the recommendation to Capital One and the numbers reversed, with 14.2% visiting against 3.8%. Skyscanner, Kayak, Sephora and Ulta showed the same pattern.
For marketers, the reading is blunt. AI visibility is a traffic driver with measurable effect on site visits. It runs on a delayed clock that referral tracking was never built to read.
Search is where the demand lands
Once persuaded, users go looking. Over half of AI-influenced visits, 55.9%, arrived through search, against 40.4% for standard visits. These people remembered the brand and sought it out rather than following any AI link. That makes branded search terms, and the paid defence of them, more valuable as AI feeds demand into the same box competitors can bid on.
Higher intent, deeper engagement
The users who arrive this way behave differently. AI-influenced visitors viewed 12 pages on average against 6.5 for standard visits, and stayed 11.8 minutes against 5.6. They had done their research inside the conversation and chosen a brand before reaching the site. Standard visitors were still comparing options.
That reframes what invisibility costs. A brand missing from the answer does not just lose a mention. It loses a high-intent visitor to whichever competitor got named instead. The dynamic is zero-sum, and the visitor rarely knows an algorithm made the choice.
An old measurement problem in new clothes
None of this is unprecedented. Advertisers spent the 20th century measuring billboards, television and radio through lift in store visits and sales, not through a click that never existed. A brand on the side of a highway in 1926 and a brand named in an AI answer in 2026 face the same question. Influence is happening, and the job is to measure it without a direct line from exposure to action.
What this asks of measurement
The fix is not more referral reporting. It is a different frame. Treat visibility as a leading indicator that surfaces later as traffic and revenue. Track real user journeys from recommendation through to visit rather than prompts and rankings alone. Benchmark against competitors, since the question is not only where you appear but where you do not and a rival does instead.
The referral picture may shift. ChatGPT has tested outbound links for recommended brands, generating roughly three times more outbound traffic in early results. The volumes stay small. Both OpenAI and Google now sell paid placement inside AI answers, echoing the old split between organic search and paid search.
None of that changes the immediate finding. The journeys in this study are running now, across millions of users and every sector studied. The influence is already in the traffic. The open question is whether the brands being named, and the ones left out, can yet see it in their own numbers.
