AI NewsGo-to-marketReported
Four studies show Google's index now buys only eligibility, says Forrester
Duane Forrester cites four studies, including Pew Research and a Penn and Northeastern field experiment, to argue that an indexed page now buys only the right to be read by the AI that writes the answer.
Image: Search Engine Journal
Why it mattersA team that still spends on content purely to rank is paying for a step that no longer leads to a click, so the next test is whether the page is the shortest honest answer an AI would quote.
A content budget that pays off in a Google-ranked click now has to pay off in a conversation it never sees. Duane Forrester, the former senior product manager at Bing and the founder of CitationIQ, argues that an indexed page today earns one thing: the right to be read by the AI that writes the answer. Whether a visitor arrives from there is a separate event with its own numbers.
Search Engine Journal carried Forrester's piece this week, drawing on his longer post at Duane Forrester Decodes. The piece names four studies, which is what pins it down.
The four numbers
Pew Research Center analysed real browsing. On a result page carrying an AI summary at the top, 8 percent of users clicked through to a site. On the same queries with no summary, 15 percent did. The AI summary cut the click rate from 15 to 8 in that sample.
A Penn and Northeastern preregistered field experiment forced AI Mode on and off for matched users on the same queries. Clicks to external sites fell by 18.8 percentage points when AI Mode was on.
A Microsoft Clarity study of more than 1,200 sites found that AI referrals were under 1 percent of total traffic. Those visitors converted at 1.66 percent against 0.15 percent for search traffic. The AI slice paid more per visit, and it remained a small share of total visits.
A Marketing Science paper by Kaiser and Schulze, over twelve months and 973 e-commerce sites, reports that visitors arriving from a ChatGPT referral converted 13 percent below the organic search baseline. The traffic was thinner on both measures: fewer sessions and worse per-session value than organic search already produced on the same sites.
What Forrester asks a team to change
Three ideas carry the piece. The first is to split a site in two: pages written for the model that writes the AI answer, and pages written for the visitor the referral still sends. Each of those jobs needs its own template.
The second is to redraw the taxonomy around the question a buyer asks. A model answering that question picks the shortest honest page it can find; a site organised by product line gives it a page to combine with four others.
The third is a measurement note. Analytics counts referral sessions. The moment a model quotes a site without sending a visitor is a separate event, which is what the four studies above are describing. Forrester asks a team to keep the two numbers in separate columns, so the second is paid for from its own budget.
A caveat the reader has to carry: the UK Competition and Markets Authority's conduct requirement, which gives sites a per-page opt-out from AI modes, is set to begin by March 2027. Today the opt-out works at the domain, so pulling a page out of the AI layer still pulls it out of the whole index.
Source
Getting Indexed No Longer Means Getting Visited, Duane Forrester, Search Engine Journal, which carried the original post from Forrester's Substack.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
