You run a shop on a busy street. The council builds a covered walkway that lets people see your window display without coming inside. Footfall drops by half. You panic.
Then you look at the till. The people who do come in now have already seen the display, already decided, and already know roughly what they want. Your conversion rate has doubled and your average sale is up.
You have not lost customers. You have lost browsers, and been handed buyers. But only if your window display was any good — and the walkway shows your competitor's window too.
The bad news, measured properly
The most credible measurement here is Pew Research Center's study of real browsing behaviour: 900 US adults who agreed to have their actual URL history collected, covering 68,879 Google searches in March 2025, of which 12,593 produced an AI summary. That is observed behaviour, not a survey. TIER A
| What Pew found | Figure |
|---|---|
| Searches where a user clicked a traditional result, without an AI summary | 15% |
| Searches where a user clicked a traditional result, with an AI summary | 8% |
| Searches where the user clicked a source cited inside the AI summary | 1% |
| Share of all Google searches that produced an AI summary (March 2025) | 18% |
| AI summaries citing three or more sources | 88% |
Read the third row again. One per cent. Being cited in an AI Overview is, in click terms, worth very little on its own. Anyone selling you AEO on the promise of citation traffic is selling you the wrong thing.
The trend since has continued in the same direction. Ahrefs' measurement of click-through-rate loss for the top organic result moved from roughly 34.5% in April 2025 to roughly 58% by February 2026. TIER B SparkToro's clickstream work puts the US zero-click share at 68% in early 2026, up from around 60% in 2024. TIER B Directionally these agree with Pew, which is the main reason I trust them.
The good news, also measured
Two things are true at once. Volume is down. Value per visit is up, substantially.
Semrush's clickstream analysis found AI-search visitors worth roughly 4.4 times the average traditional organic visitor, and around three times as likely to convert. TIER B The mechanism is obvious once stated: someone arriving from an AI answer has already had the basic question resolved. They are not researching what a customs broker does. They are clicking through to check whether you are the one. They arrive later in the journey and closer to a decision.
Meanwhile the absolute volume is still small and you should not be told otherwise. ChatGPT's referral share of global website traffic was measured at around 0.32% in May 2026 TIER B — growing fast, from a very low base. Anyone projecting that AI referrals will replace your organic traffic this year is guessing.
The honest position: AI search will not save your traffic numbers. It decides whether you are on the shortlist. Those are different businesses, and the second one is worth more.
What you should actually be optimising for
This changes the goal, and therefore the scoreboard. Three targets, in order of how much they matter:
Are you named when someone asks for options in your category? This is the whole game. It produces no traffic and no analytics event, and it is still the most valuable thing on this list.
Being named with the wrong services, wrong location or wrong price is worse than not being named. Chapter 19.
Track conversion rate and value of AI-sourced sessions separately. If you judge them by volume you will kill a channel that is outperforming per visit.
For a B2B or considered-purchase business, there is a further finding worth knowing. G2's 2026 buyer research reports that around half of B2B software buyers now begin research with an AI chatbot, that 69% ended up choosing a different vendor than they had originally planned, and that a third bought from a vendor they had never heard of before. TIER B Treat the decimals with suspicion; treat the direction seriously. The shortlist is being formed somewhere you cannot see, before anyone contacts you.
In your analytics, build a segment for referrals from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com. Compare conversion rate and average value against organic — not sessions against organic. If it is a tiny, high-converting trickle, that is the expected, healthy result. Chapter 18 explains why the session count will still understate reality.
Change the question you ask about this work. Not "how much traffic did it bring" but "when a stranger asks the AI for someone like us, are we on the list?" Everything else follows from asking the right one.