Last week we wrote about how an AI agent weighs your website against everything else said about you, drawing on Helena Dillon’s talk for Microsoft Advertising at Spotlight, the London conference on brand visibility in AI search. The same talk showed how the agent decides what to look for in the first place, through something Dillon called query fan-out.
Take a marketing director looking for a new web partner: in traditional search she types something like “website agency London” into Google and gets a page of results ranked for those words, and SEO is the work of being near the top of it. If she asks an AI assistant “Which agency should rebuild our website?” instead, it splits her question into searches of its own, on what a rebuild costs, how sites are hosted, who builds to accessibility standards and who has done similar work before, and writes one answer from the pages those searches turn up. Google says its AI Overviews and AI Mode can work the same way, running related searches across the different parts of a question at once. Your buyer never sees those searches, and until recently neither did you.
What a keyword list misses
Keyword tools show what people type into Google, but fan-out queries are written by the machine and change almost every time it runs: in a test of 1,600 runs, the SEO software company Surfer found that two-thirds of the fan-out queries Gemini wrote appeared only once.
You can see the effect in who gets cited: in July last year Ahrefs found that about three-quarters of the pages Google’s AI Overviews cited also ranked in the top ten for the search itself, but two large studies this year put the figure at between 38% and 41%, with Ahrefs looking at four million cited pages in March and RankRabbit at 2.2 million citations in July. Ahrefs thinks fan-out is the likely reason, and Surfer’s research points the same way, finding that pages ranking for the searches around a query get cited far more often than pages that rank for the query alone.
Chasing fan-out queries one by one doesn’t work, because there are too many and they rarely repeat, but covering the questions behind them does. Marcel Santilli of GrowthX, who also spoke at Spotlight, called the question the atomic unit of intent and each of those needs a page on your site that answers it in full.
Your buyers have already written them down
In July we asked what happens when buyers draft their RFP criteria inside an AI, and we’ve been back through five years of our own deal channels to see what RFPs and procurement teams actually asked us.
The same questions came up again and again: how we secure client data, whether we’ll warrant accessibility to WCAG 2.1 AA in the contract, what each role costs rather than a blended day rate, and where the site is hosted and how fast it runs. This year one RFP went further and asked how any AI features would keep data separate, whichever model provider sat behind them. We’d answered each of those questions from scratch, under deadline, at least twice, yet most of the answers sit within emails, documents and call transcripts held on internal drives.
Buyers want those answers before they talk to anyone, and 6sense’s 2025 survey of more than 4,000 buyers found that most vendor websites don’t clearly explain their AI features, from data sources and security to pricing and performance, and that vendors who can’t give clear answers at that stage are being left off shortlists altogether. The vendor buyers favoured before they spoke to a seller went on to win 80% of deals. If your site doesn’t answer an RFP’s questions, the AI answers them from somewhere else, which is the gap we wrote about last week, and it opens before you know the RFP exists.
What to do now
Here’s how you can start:
- Gather the RFPs and security questionnaires from the last two years, including the deals you lost, and list every question you’ve answered more than once, in the buyer’s own words.
- Find the page on your site that answers each one in full. Where there isn’t one, you’ve found your next page.
- See how the machines phrase the same questions. Bing Webmaster Tools and Microsoft Clarity show a sample of the searches Microsoft’s AI ran to find your pages, while paid tools such as Semrush and Surfer, or simulators such as iPullRank’s Qforia, estimate the fan-outs behind Google and ChatGPT answers. (Semrush is a client.)
- Write each answer once on your site, and link to it in the next RFP.
If you’d like help with any of this, our AI visibility audit goes further than these steps, testing your buyers’ questions across the main AI engines and showing where your competitors get the answer instead.
