Most claims about AI search killing traffic are extrapolated from a single site's analytics. Pew Research did something more useful: it analysed the real browsing activity of 900 US adults who agreed to share it, covering search behaviour through March 2025.
Two findings matter commercially. First, 58% of participants ran at least one Google search that month producing an AI-generated summary — this is not an edge case. Second, when an AI summary was present, 8% of users clicked through to any website. When it was absent, 15% did. Roughly half the clicks, gone.
The third finding is the one that should change your content strategy: only 1% of users clicked a link inside the AI summary itself. Being cited as a source in an AI answer is not a traffic channel. It is a branding impression at best. Any plan built on 'we'll optimise to be the citation' is planning for a 1% click rate.
The correct response is not to abandon search. It is to stop treating undifferentiated informational content as a growth asset. If a page answers a question an AI can summarise in three sentences, that page's traffic is structurally in decline and no amount of on-page work reverses it.
What survives is content an AI cannot generate: your own data, your own client outcomes, your own priced offer, your own tooling. A page reporting what happened when you ran a specific change for a specific client cannot be synthesised from the open web. Neither can a calculator, a benchmark, or a diagnostic that requires the user's input.
The operational implication for anyone measuring marketing: separate your reporting between informational pages and commercial pages before the trend distorts your averages. Informational traffic will fall. If commercial enquiry volume holds or rises while it does, the system is working. If you only watch sessions, you will panic at the wrong moment or miss a real problem inside a falling average.
Sources
- 1.Google users are less likely to click on links when an AI summary appears in the resultsPew Research Center · 2025