Measurement Is AI Search's Biggest Unanswered Question

At BrightonSEO San Diego, our team asked some of the sharpest people in AI Search a simple question: what's YOUR biggest unanswered question about AI search right now?

We expected a spread of answers: agents, Google's next move, the future of the website. What we got was a pattern that was so stark it even felt repetitive by the day's end. Five of the eight leaders who answered said some version of the same thing: we still can't measure this well.

Some framed it as a prompt tracking problem. Nobody knows what people actually type into ChatGPT, Gemini or AI Mode, so nobody fully trusts the visibility scores built on top of those prompts. Others framed it as an attribution problem. Even when you show up in an AI answer, connecting that to pipeline and revenue is murky. 

That matters, because measurement is what earns a budget and internal support. If you can't show what AI search is doing for the business, you'll struggle to keep investing in it. Here's what we heard, who said it, and what we think you should measure instead.

Who we asked

I specifically interviewed the folks that I felt have been leading the AI Search conversation online, as well as the companies that have been diving in and experimenting the most. We do this often with our guests on our podcast, The Campaign. Kevin Indig came on with us and expressed a lot of the same concerns we heard at Brighton. Here was the lay of the land of folks responses to what is still unanswered for them in AI Search:

Chris Donnelly, CEO and Co-Founder of Searchable, mentioned the Prompt Universe, Searchable’s new product offering, as a way to answer the questions of, “What prompts am I tracking? Are they the right prompts? Am I tracking enough prompts?” While Searchable, Profound, and Scrunch are growing tools, trust across the board of their validity doesn’t seem to be strong in the SEO community.

Problem 1: People aren’t trusting prompt tracking just yet, so stick to knowing your audience

In traditional search, Google Search Console tells you which queries you showed up for. AI Search has no equivalent. That gap is what most of our experts are stuck on.

Chris Long, co-founder of Nectiv, put it plainly (watch Chris's clip):

"No one knows the prompts people are putting into AI search systems."

He isn't optimistic it gets solved soon. Every tool has its own method for building prompt sets, and none of the major AI platforms share real query data. Unless a data provider opens it up, he doubts there will ever be a great answer.

Brooke Weller, who leads AI search at LinkedIn, sees the same thing from inside a major brand (watch Brooke's clip):

"I don't think anyone can accurately track prompts."

Even at LinkedIn, she reports an AI search visibility score she isn't sure is accurate. She also points out that AI conversations are personal and long-running. She has threads that go on for months, so no single "first prompt" captures how people actually reach a brand.

Ken Marshall, co-founder of Meet Sona, wants more transparency from the tools themselves (watch Ken's clip). Many visibility platforms don't publish how they choose prompts or model personas.

This is a human element of search that still has to be in place from human marketers in order for a brand to achieve visibility and recognition. Alyssa Felix, one of our senior search marketers, put it plainly, “The best prompt research tool is still talking to your customers, reading their questions, and paying attention to what they actually say.”

Problem 2: Proving AI search drives revenue

The second half of the problem starts after a brand shows up in an AI answer. What did that appearance actually do for the business?

Lucy Hoyle, founding content engineer at Carta, named the pressure many marketers feel (watch Lucy and Nick's clip):

"Attribution is still really difficult. Visibility is one piece and it's very top of funnel, but how does that then track to engagement, revenue?"

Marketing teams are being asked to generate pipeline and prove ROI, and AI visibility on its own doesn't do that.

Nick Lafferty, founding marketing engineer at Profound, shared one low-tech fix that works. Profound's contact form forces every lead to write, in an open text field, how they heard about the company before being able to navigate further. Self-reported attribution catches the AI answers and conversations that tracking can’t uncover. Nick pointed out that this  kind of first-party signal "will become really foundational" for brands if they do want to have somewhat of an answer for attribution as it gets messier and messier. 

This matched what our own Alyssa Felix heard at the BrightonSEO AEO/GEO roundtable. For in-house teams, she said, "the big question right now is reporting and attribution": proving organic still works, and proving the team behind it still matters. The buzz phase of AI Search is over. Executives know it’s important, but HOW important is starting to be questioned and the level of investment will be determined by how much attribution can be shown. 

Rosy Callejas, who works on enterprise SEO and AI search at Microsoft, offered the practical takeaway in approaching executives about AI Search now. (watch Rosy's clip). To win trust from leadership, bring "ANY data to show them," because "we're all data driven at this point." Something is better than nothing! 

The other three answers

The three experts who didn't name measurement still raised questions worth tracking:

• What happens to the website? Martha van Berkel of Schema App wants to see how much brands invest in "the human experience versus the machine experience" as AI agents read and act on sites (watch Martha's clip).
• How will AI platforms monetize going forward? Skyler Rudolfsky of Semrush pointed out that running AI is expensive and few platforms are profitable. Whatever monetization model wins, whether subscriptions, ads or something new, will shape how brands show up (watch Skyler's clip).
• Why do citation sources shift? Thomas Peham of OtterlyAI has watched ChatGPT updates cut Reddit citations and raise YouTube's. The data shows what changed, he says, but "the why question" is still unanswered (watch Thomas's clip). We dug into one of those shifts in ChatGPT's Reddit citations just fell 86%. Here's why you shouldn't panic.

Even these answers come back to measurement. If citation sources can swing overnight, you need reporting that catches the swing and shows what it cost.

What to measure while prompt data stays murky

Perfect prompt data may never arrive. That's no reason to fly blind. Here's the approach we use with clients at 97th Floor:

1. Start with first-party data. Before you buy another tracker, use what the platforms already give you. Google Search Console now has a generative AI performance report that separates AI Overviews and AI Mode impressions from regular search. Pair it with GA4 referral traffic from AI platforms and Bing Webmaster Tools.
2. Track citations, not just mentions. A mention tells you AI named your brand. A citation tells you which of your pages informed the answer, and you can actually optimize that page. Track citation sources, share of voice and your top cited pages. Our guide on how to track brand mentions in AI search walks through it.
3. Treat prompt tracking as directional. Use visibility scores to spot trends and gaps, not as a KPI you promise leadership. As Brooke put it, keep your eye on "the bottom line."
4. Add self-reported attribution. Take Nick's advice and put an open "How did you hear about us?" field on your high-intent forms. It's the cheapest AI attribution tool there is.
5. Connect visibility to pipeline. Report AI search alongside sessions, leads and revenue, not separately. That's how measurement goes from "woo-woo" to budget.
6. Watch what AI says, not just whether it says it. Accuracy and brand sentiment are next. A citation that describes your product wrong can do more harm than no citation at all.
7. Don’t settle thinking you know your customer base. Our GEO/AEO team tracks the KPIs that matter to the business, and centers on getting to know your audience in finite detail, as it’s the best source for prompt research. Talk to us to learn more about how we map your personas in order to capture the right volume in AI Search. 

Build your AI search scoreboard before leadership asks for one

AI search isn't missing strategies. It's missing a scoreboard. The experts we talked to at BrightonSEO know how to earn visibility in ChatGPT, Gemini and AI Mode. What they can't do yet is say, with confidence, which prompts drove it and what it was worth.

Until the platforms open up their data, the teams that win will be the ones that combine imperfect signals: first-party platform data, citation tracking, self-reported attribution and pipeline. Then they bring those numbers to leadership before anyone asks.

For more from the conference, read Mike Witham's 45/40/15: What BrightonSEO San Diego taught me about where SEO is headed.

Ready to prove what AI search is doing for your business? Let's talk.