The measurement problem in AEO is not that the data is hard to collect. It is that the most consequential thing happening produces no data at all.
A buyer asks an AI which vendors to consider. Three names come back. Two get shortlisted. Nothing is clicked, nothing is logged, and no analytics platform anywhere records that the decision happened. Weeks later one of those vendors gets a demo request that attributes to branded search.
The Numbers on How Much Is Invisible
Industry analysis puts traditional attribution at roughly 27 percent of the B2B buying journey, with the remaining 73 percent occurring in channels producing no trackable signal. AI conversations, private discussions, peer referrals, and direct navigation.
6sense's 2025 Buyer Experience Report, covering 4,510 B2B buyers, found 80 percent of deals are won by the vendor the buyer already favoured before first contact. Bain's research puts 85 percent of purchases with a vendor on the buyer's Day One list.
Forrester's 2026 survey of roughly 18,000 buyers found 94 percent used AI during their most recent purchase, and 55 percent compared vendors inside AI tools before contacting anyone. Apollo's 2026 data has the average shortlist down to about 2.5 vendors from 3.2.
Read together: the shortlist forms early, in a channel that logs nothing, and it largely determines the outcome. A dashboard built on sessions is measuring what happens after the decision that mattered.
Traffic Will Look Bad Even When AEO Is Working
This is the part that causes internal problems. Forrester reported in February 2026 that B2B companies are seeing traffic declines of 10 to 40 percent as research migrates into AI engines.
A company doing AEO well can see its traffic fall while its win rate improves, because buyers are arriving later in the process, better informed, and already favourable. That pattern is easy to misread as failure if traffic is the headline metric.
Seer Interactive's April 2026 analysis found brands cited inside an AI Overview earn 120 percent more organic clicks per impression than uncited brands on the same page, 2.07 percent against 0.94 percent on informational queries. Citation still drives clicks. It just drives fewer of them than the old model, from a smaller pool, at higher quality.
Three Layers Worth Reporting Separately
Citation presence. Whether you appear in answers to the questions your buyers ask, measured across the engines covered throughout this blog and detailed in the measurement guide. This is the leading indicator and the only one you can measure directly and completely. It has no revenue attached and that is fine, because it precedes everything that does.
Attributed AI referral. Sessions arriving with an identifiable AI referrer, and what they do. Small in volume and useful in quality. Adobe found AI referrals converting 31 percent better than other traffic during the 2025 holiday season, and multiple sources put AI referred B2B traffic converting at several times the rate of standard organic.
Pipeline signals that move without a traceable source. Branded search volume, direct traffic, self reported attribution on forms, and the proportion of deals where the buyer arrived already familiar. These are noisy and they are where the 73 percent shows up if it shows up anywhere.
Reporting these as one blended number destroys the information. Reporting them separately lets you see the leading indicator moving before the lagging one does.
Self Reported Attribution Is the Most Underused Instrument
Adding a how did you hear about us field to demo request forms is unglamorous and it is the only mechanism that directly captures dark funnel influence.
Make it a free text field rather than a dropdown. A dropdown containing your existing channel list will never surface an answer you did not anticipate, and asked an AI is exactly the answer you did not anticipate.
This also surfaces brand description problems early, since a buyer reporting what they heard about you is reporting what an engine said, which connects to the accuracy monitoring in the brand hallucination guide. The data is messy and directionally honest in a way attribution modelling is not. When the same unprompted phrase starts appearing across responses, that is signal, and it arrives months before any attribution model would infer it.
Branded Search as a Proxy
The clearest available proxy for invisible influence is branded search volume moving without a corresponding campaign.
Someone encounters your name in an AI answer, does not click, and searches your brand later. That produces a branded search with no traceable origin. A sustained rise in branded search that does not track with paid spend, PR, or product launches is reasonable evidence of upstream exposure.
It is a proxy rather than proof and it should be reported as such. The share of voice guide covers the citation side of the same question, which is the more direct measure where you can run it.
Timeframes Have to Be Set Honestly Upfront
The staggered engine pattern described in the migration guide applies to any AEO programme. Live retrieval engines reflect changes in weeks. Training dependent engines take considerably longer and you cannot accelerate them.
Content freshness affects this too, since a programme refreshing stale pages will see movement faster than one publishing new content into a cold domain, per the content decay guide. Then the sales cycle sits on top. Reported B2B cycles average around ten months. A citation earned in January influences a shortlist in March and a closed deal in September, by which point nobody is connecting the two.
Meaningful pipeline impact from AEO is generally described as a 60 to 90 day horizon at minimum before demos and qualified leads attributable to AI referral become visible, and longer before revenue. Committing to a quarterly revenue number is setting up a programme to be judged before it can have worked.
What I Would Actually Put in a Board Report
Citation rate across tracked prompts, by engine, trended over months rather than weeks. This is the honest leading indicator.
Share of citations against named competitors on the same prompts, which converts an abstract number into a competitive position leadership can read.
AI referral sessions and their conversion rate against other channels. Small volume, and the quality differential is the point.
Self reported attribution mentions, quoted rather than counted.
And a stated caveat that a meaningful share of influence is structurally unmeasurable. Including that caveat is more credible than omitting it, and it prevents the programme being judged against a completeness no channel can deliver.
The Metric I Would Not Report
A single composite AI visibility score with undisclosed methodology. Google's own guidance on evaluating third party SEO advice, covered in the vendor vetting guide, makes the relevant point: no external tool has access to internal ranking data, so any proprietary score is a model rather than a measurement.
Reporting one to a board invites a question you cannot answer, which is how it was calculated. Reporting observed citations on named prompts invites a question you can answer, which is go and check.
Measuring the Observable Part Properly
The NotionCue Prompt Tracker covers the citation presence layer, run against prompts written the way buyers actually phrase them rather than as keywords, per the conversational search guide. The NotionCue Citation Tracker captures the text of how you are described, which is where accuracy problems surface before they become entrenched.
Neither closes the attribution gap, because nothing does. What they do is measure the layer that precedes it completely and honestly, which is a better foundation for an argument than a modelled number nobody can verify.
Start your free NotionCue trial and build a citation baseline before your next reporting cycle. A trend line needs a starting point and the starting point is only capturable now.
One question worth asking in your next pipeline review: of deals closed last quarter, how many involved a buyer who already knew the product before any tracked touchpoint. If nobody can answer, that is the measurement gap this post is about, and a free text attribution field starts closing it within a quarter.
Common Questions
How do we justify AEO spend without clean attribution?
On the same basis as brand investment, which has never had clean attribution and has never stopped being funded. The leading indicator is citation presence, and the argument is that presence during a research phase which decides 80 percent of outcomes is worth funding even where the click cannot be traced.
Is AI referral traffic worth reporting given the low volume?
Yes, specifically because of the conversion differential. Reporting it as a percentage of sessions understates it. Reporting it as a percentage of conversions or pipeline shows what it is actually contributing.
What if traffic falls while citations rise?
That is a plausible outcome of the shift rather than a failure, given the reported 10 to 40 percent declines across B2B generally. The thing to watch is whether conversion rate and win rate hold or improve on the reduced volume. If both fall together, that is a different problem.