Sessions referred by AI assistants convert at three to five times the rate of organic sessions, according to client data from growth consultancy Winston Francois — yet one founder saw pipeline drop 18% for two straight quarters while his organic traffic and rankings improved.

The founder's dashboard showed 12 new page-one rankings since spring. But when the consultancy checked his top five buyer prompts across four AI assistants, his company appeared in only 3 of 20 answers. All 12 new rankings were informational terms, not buyer-intent queries.

"We'd gained visibility on questions people ask when they're learning about a category. We were barely showing up when they asked which company to buy from," the consultant wrote.

Winston Francois works with venture- and PE-backed companies. The five metrics below come directly from those engagements and address a reporting gap: rankings and organic traffic no longer capture buyers who compare vendors in an assistant, then navigate directly to a chosen site.

Which five metrics track AI search performance?

• Brand presence on buyer prompts. Run five buyer-intent questions (e.g., "best contract management software for mid-market legal teams") across four assistants weekly. Record whether the company is named, linked, or absent across all 20 answers.
• Answer accuracy. Mark each mention as accurate or inaccurate. One Series A client appeared in 11 of 20 answers, but only 4 were accurate — assistants described an enterprise product while the company sold to businesses under 200 employees.
• Conversion from AI referrals. Group GA4 sessions from ChatGPT.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai by Session source; track demo and trial conversions.
• Branded search. Monitor branded impressions and clicks in Search Console weekly, plus direct traffic to the homepage and pricing page.
• Self-reported attribution. Add "ChatGPT or another AI assistant" to inbound forms and have sales log the answer in the CRM.

How fast does accuracy fix pay off?

After Winston Francois made the misdescribed client's information consistent across readable sources, 10 of 11 mentions became accurate within six weeks. Demo-to-opportunity conversion recovered the following quarter.

The timing pattern across engagements: brand presence improves two to six weeks after cleanup and first content changes; answer accuracy moves in the same window; AI referral conversions appear in weeks four through eight; branded search rises around weeks eight through 12; self-reported attribution can lag by a full sales cycle.

The consultant cautions against crediting AI alone for branded search gains: "Before attributing that increase to AI, check what else ran during the same period, including paid campaigns and PR." He would increase spend only if brand presence, accurate answers and branded search improve together and sales hears buyers mention AI.

What did self-reported attribution reveal?

One client added the AI option to its forms in April. By July, 9 of 54 new opportunities selected it — and those deals closed 30% faster than the rest of the inbound group, data a last-click view would have missed.

The framework carries caveats the consultant states openly. GA4 misses visits without a usable referral source, Google AI Mode traffic can fold into organic, and buyers skip or forget form fields. The identifiable AI referral numbers are a partial view, not a census.

How to start without new tooling

The entire system runs on a spreadsheet updated every Monday by whoever owns AI search. The workload: roughly 40 minutes for 20 answers plus accuracy review, 15 minutes to build the GA4 segment once, and five minutes for the form edit. The recommendation is to wait eight weeks before building a dashboard, and to run the tracking as part of a 90-day growth audit.

The original founder now reports his presence in buyer answers alongside pipeline, and both metrics have started improving — a better basis for the next budget decision than a report that stopped at rankings.

Teams wanting to replicate the approach can start this week: test five buyer prompts across four assistants, then bring the answers and any inaccurate descriptions to the next pipeline review.