Most AI-referred site visits arrive days after the prompt, not minutes, and they usually land through search or direct traffic, not a labelled AI link. Two independent 2026 datasets point the same way: Gener8's Tracking Prompts to Site Visits report found the prompt-to-visit rate climbs roughly 9x to 10x from one hour to 72 hours after a prompt, and Similarweb found an AI recommendation made people 2.5x more likely to visit a brand's site over the next seven days.
Last reviewed: 10 August 2026. The figures below come from two named sources: Gener8's AI Landscape: Tracking Prompts to Site Visits (July 2026) and Similarweb's The Downstream Impact of AI Visibility (published 21 June 2026). Both are large but single-vendor and self-reported, so treat the direction as firmer than any single percentage, and hedge anything you quote as "as of mid-2026." Gener8's data is mobile-first (Android in-app and mobile web; iOS excluded under Apple's policy), covers the US and UK, spans August 2025 to June 2026, and is weighted to census benchmarks by gender, age (18–64) and region.
How long after an AI prompt do people visit a site?
The delay is the headline. Gener8 measured a prompt-to-visit rate — the share of AI prompts in a category that lead the same person to a relevant site visit within a set time window — at 1 hour, 24 hours and 72 hours after the prompt. In every vertical, the rate multiplied between 9x and 10.3x from the 1-hour mark to 72 hours:
| Vertical (US) | 1 hour | 24 hours | 72 hours | 1h→72h multiple |
|---|---|---|---|---|
| B2B SaaS | 2.7% | 14.1% | 25.6% | ~9.5x |
| Consumer Electronics | 2.2% | 10.2% | 19.9% | ~9x |
| Home Improvement | 1.1% | 5.0% | 10.0% | ~9x |
| Apparel & Fashion | 0.7% | 3.5% | 7.2% | ~10.3x |
Source: Gener8, AI Landscape (July 2026), mobile, US, Aug 2025 – Jun 2026.
The takeaway Gener8 draws is blunt: most of a prompt's eventual visits take days, not minutes to materialise. An AI answer plants intent; the click that acts on it often comes later, once the person is ready to research or buy. If you judge an AI answer's impact by the traffic in the first hour, you are measuring a small slice of what it will eventually send.
Which categories convert prompts to visits fastest?
The rate differs sharply by vertical — about 4x between the top and bottom at 24 hours. Gener8's US 24-hour prompt-to-visit rates:
| Vertical | 24h prompt-to-visit rate |
|---|---|
| B2B SaaS | 14.1% |
| Consumer Electronics | 10.2% |
| Home Improvement | 5.0% |
| Apparel & Fashion | 3.5% |
Source: Gener8, AI Landscape (July 2026), mobile, US.
B2B SaaS leads at every time mark, which Gener8 reads as research-led intent rather than typical shopping behaviour: people investigating software follow the trail to a vendor's site more readily than people browsing apparel. The lesson for measurement is that a single blended "AI traffic" number blurs categories that behave nothing alike — segment by vertical before you benchmark yourself.
Does the AI platform change the prompt-to-visit rate?
Yes, but the ranking is stable even as the size of the gap moves. Gener8 found Google AI Mode had the highest prompt-to-visit rate across all four verticals, with ChatGPT second in every one and Google Gemini third. The gap ranged from about 1.5x to 2.1x over the third-place platform:
| Vertical (US, 24h) | Google AI Mode | ChatGPT | Gemini |
|---|---|---|---|
| B2B SaaS | 20.4% | 16.2% | 11.0% |
| Consumer Electronics | 13.2% | 10.7% | 8.6% |
| Home Improvement | 7.5% | 5.9% | 3.6% |
| Apparel & Fashion | 5.0% | 3.9% | 2.7% |
Source: Gener8, AI Landscape (July 2026), mobile, US. Google AI Mode, ChatGPT and Gemini only.
Gener8 adds an important caveat: Google AI Mode is usually reached inside a Google search rather than opened as a standalone destination, so its traffic likely skews toward more navigational, ready-to-buy intent than a standalone ChatGPT or Gemini session. That framing may explain part of its lead — a reminder that platform comparisons measure the surface and the intent that arrives there.
How do those visits actually reach your site?
Not as an AI referral you can see. Similarweb's Downstream Impact of AI Visibility study traced where AI-influenced visits landed and found most arrive through search, not a citation link:
| Arrival channel | AI-influenced visits | Visits with no AI influence |
|---|---|---|
| Via a search engine | 55.9% | 40.4% |
| Direct (no referrer) | 19.9% | 38.8% |
Source: Similarweb, The Downstream Impact of AI Visibility (June 2026), US desktop panel, Jul–Dec 2025.
The behaviour behind the numbers: someone reads a recommendation in an AI answer, then re-searches the brand or returns to it later — so the visit records as branded search or direct, not as a click from the assistant. This refines, rather than contradicts, the corpus's existing finding that AI traffic hides in the "Direct" bucket: both are true, and the through-line is that an AI-driven visit rarely arrives as a labelled AI referral. Similarweb also reported the impact is real — an AI recommendation made people 2.5x more likely to visit that brand's site over the following seven days — and that those visitors viewed roughly twice as many pages and stayed roughly twice as long, corroborating the independent finding that AI-referred traffic is more engaged.
Is this the same as the citation lag?
No — and keeping the two apart matters for reading your own data. There are two distinct delays on the AI funnel:
The citation lag is the weeks between an AI bot crawling your page and the engine citing it in answers. The prompt-to-visit lag is the hours-to-days between a person reading that answer and clicking through to your site. A page can clear the first and still see its visits arrive slowly across the following week.
The citation lag is a supply-side delay (crawl, then index, then cite). The prompt-to-visit lag is a demand-side delay (see the answer, then act on it). Stacked, they mean a page you publish today may not be cited for weeks, and then may not send its full visit volume for days after each answer — so the honest measurement window for a new page is a quarter, not a fortnight, and shorter than that will read as failure when it is really just latency.
What predicts a higher prompt-to-visit rate?
A geeky finding worth noting for anyone modelling this: Gener8 reported that heavier search intensity predicts a higher prompt-to-visit rate, but heavier GenAI use does not. The top 10% of users by search volume drove a prompt-to-visit rate roughly 2x to 4.3x higher than the lightest decile. Yet across three of the four verticals, prompt-to-visit rate stayed flat or declined as GenAI-use intensity rose. The read: the people most likely to act on an AI answer are heavy searchers, not heavy AI users — consistent with the fact that the visit itself so often arrives through search. It also squares with Gener8's market-level finding that search volume kept climbing (US searches per user up ~15%, UK up ~20% since August 2025) even as AI adoption surged — AI is adding a discovery step, not replacing the search that closes it.
How should you use these figures?
Treat this as a dated reference, not a promise. Every number here is single-vendor and self-reported; cite Gener8 or Similarweb and the date when you reuse one, and prefer your own measured trend over any headline percentage. Because live retrieval favours recently-updated pages, we keep references like this on a refresh cadence and update them substantively as new editions land.
The practical shift is in how you measure: widen your attribution window past the first click, expect the visit in search and direct rather than a labelled referral, and read AI's impact as a curve that builds over days. Doing that — tying which prompts and citations surface your brand to the delayed, indirect visits they eventually send — is exactly what Buffy Intel is built to track. Start with the companion how-to on measuring AI visits that arrive days after the prompt.