Article Synopsis:
- Track leading indicators now (such as how often AI tools mention you, branded search, and what leads tell you)
- Pair them with the pipeline numbers you already trust -- do not wait for a perfect AI search attribution model. (It does not exist yet, and your CFO is not expecting one!)
The measurement gap is real, and everyone has this problem
AirOps surveyed about 250 CMOs. Close to 90% said AI search visibility is important or a top priority. Only 3% felt confident measuring its impact. Bain's research across roughly 1,400 marketing and finance leaders shows the same shape for AI in marketing overall: 70% expect significant value, but only 6% call it a fundamental driver of performance today.
We heard both numbers at a recent AirOps webinar with Matt Hammel of AirOps and Zak Prauer of Bain. My takeaways are on LinkedIn. The most useful point was an old one. Every new channel goes through this. Search and social handed marketers data first, and the methodology showed up later.
Traffic down, leads up: what it can look like
One healthcare client of ours saw website visitors dip 0.4% quarter over quarter. Online lead flow rose 14% in the same quarter, and traffic was still up 10.5% year over year. Rankings for our key search terms (tracked by Wincher) held in positions one through three for nearly all key terms.
We had shifted the content toward answer-first pages that respond to what patients actually ask. We cannot prove the shift caused the lift. We can say that a dashboard built around sessions alone would have flagged a problem in the one quarter things went well.
A different client, a specialty healthcare provider with fewer than 20 people, told us something no dashboard showed: their last seven leads said they found them through ChatGPT. For a practice where a single referral is worth tens of thousands of dollars, that is the kind of evidence the webinar speakers called circumstantial but worth acting on. It is also exactly what a "how did you hear about us" question catches and a sessions report misses.
What to track while the models catch up
- Mention and recommendation rate. Pick the 10 to 20 questions your buyers really ask. Run them through ChatGPT, Perplexity and Google's AI Overviews on a set schedule. Log whether you show up, and whether you are the recommended answer.
- Rates, not levels. These systems are probabilistic. You will not get the same answer twice. A single check proves nothing. Improvement week over week or month over month does.
- "How did you hear about us." Ask every inbound lead. Compare the answers to what your analytics says. Much AI traffic arrives labeled "direct," so referral numbers undercount it. The size of that gap is a useful number on its own, even if the survey is imperfect.
- Branded search and direct traffic. When AI recommends you, people often search your name next. Watch for that lift.
- Conversion, not clicks. Studies cited in the webinar put AI search visitors at roughly 4 to 10 times the conversion rate of traditional organic traffic. That figure varies by industry and does not always hold. Even so, a smaller audience that converts better is a trade worth checking for.
Why your CFO is probably not the obstacle
Bain asked CFOs what they expect. The answer was thoughtfulness, clear baselines, and directional improvement, not perfection. Most are looking for growth, not a precise ROI figure. If you tell finance what you can measure, what you cannot, and how you will decide anyway, you are in better shape than the many teams waiting for a cleaner dashboard.
Start small
Skip the new attribution platform and the all-channel overhaul. Pick one group of buyer questions, one segment, and one baseline. Run it for a quarter. Marketing mix models are worth planning for, but they typically need a year or more of consistent data, so start collecting the inputs now.
B2B buyers also tend to buy from a shortlist they form early. Bain's earlier work found 80% of B2B buyers have a day-one vendor list and about 90% buy from it. That list is built from what trusted people say about you and what is easy to find. AI answers draw on much the same material. If you need a place to begin, The Citable Standard covers how to get your content cited.
The short version
Perfect measurement is coming. The teams testing now will be the ones who can prove it later. Start with a baseline, a short list of buyer questions, and a monthly check. That is a 30-minute job, and it beats waiting.