Strategy and measurement
Measure AI's influence beyond the referral click
Combine AI answer evidence, referral analytics, buyer-reported discovery, and CRM outcomes without inventing a revenue multiplier.
TL;DR
- AI can influence a shortlist without sending a measurable visit to the brand's website.
- Observed research tasks and selected products are not the same as completed purchases.
- Combine referral data, saved answer evidence, buyer-reported discovery, and sales outcomes with clear labels.
- Do not multiply referral revenue by a published no-click percentage to estimate total GEO value.
Referral analytics answers a limited question: which visits reached your site with a recognizable source? A buyer may find your brand in an AI answer. Later, they may search for its name and reach sales through a different channel. So referral visits show only part of AI's role. That does not mean every untracked lead came from AI.
What observed shopping behavior tells us
A July 2026 study commissioned by Profound and conducted with Clickstream Solutions observed 56 people and 221 codable product-research tasks. It recorded screens and choices while users spoke their thoughts. It then compared those choices with brand-presence data. The people in the study were regular chatbot users. The study tracked research and choices. It did not audit completed sales. Original study and methodology.
| Measure | Reported result | What it supports |
|---|---|---|
| Tasks ending without an open-web click | 92.8% | Choices can form without a referral visit |
| Share of voice for chosen brands | 24% | Chosen brands had more observed exposure |
| Share of voice for passed-over brands | 11% | A comparison group within the study |
| Correlation between visibility and choice | 0.57 | An association, not a causal sales lift |
The findings justify measuring what the answer says. They do not establish that the same share of B2B buyers will skip your website, or that an unseen visit would have become revenue.
Independent browsing evidence has a different denominator
Pew Research Center studied browsing data from 900 US adults in March 2025 and matched Google queries to results collected the following month. The analysis covered 68,879 unique searches. People clicked less on pages with AI summaries. The study did not track what they chose or bought. Pew's methods and findings.
| Action | With an AI summary | Without an AI summary |
|---|---|---|
| Click on a traditional result | 8% | 15% |
| Click on a link in the AI summary | 1% | Not applicable |
| End the browsing session | 26% | 16% |
Both studies show why clicks alone can miss part of the picture. Keep their percentages separate. One concerns assigned tasks in ChatGPT. The other concerns visits to Google search pages. Neither tells you how much hidden revenue you earned.
Build an evidence record across the buying path
Use a small set of measures with clear jobs. Keep the original source of each record. If a buyer says “ChatGPT helped me compare vendors,” preserve that as self-reported influence, even if analytics records an organic search visit.
| Layer | Evidence | Claim allowed |
|---|---|---|
| Answer presence | Saved answer, prompt, model, date, and cited page | The brand appeared in this tested answer |
| Referral activity | Recognized source and consented analytics event | An observed visit or conversion came through this source |
| Buyer-reported influence | Optional discovery question or sales note | The buyer reported AI as part of their research |
| Commercial outcome | Qualified inquiry, opportunity, and recorded sale | A documented business outcome occurred |
| Incremental effect | A suitable experiment or credible comparison design | An estimated effect under the design's assumptions |
Do not add these rows together. The same buyer can appear in several of them. Use your normal CRM rules to remove duplicate people and deals. Keep test answers separate from real buyer sessions.
Worked example: a lead with mixed evidence
Illustrative example. A buyer arrives through a Google brand search and submits a project inquiry. On the sales call, the buyer says an AI comparison introduced the supplier. They choose to share the answer. The CRM records an organic-search visit, self-reported AI discovery, and a qualified opportunity.
The review finds that the AI got the service area right. But it left out a minimum-order rule. Semanta's evidence review would link that gap to the right page and propose a clear fix. This creates useful work before the team settles how to assign revenue.
If the opportunity later closes, report the recorded revenue and the documented influence evidence. Do not assign all revenue to AI by default. The buyer may have used referrals, reviews, and other sources too. To claim extra revenue caused by the work, the test must address what would have happened without it.
Common questions
Can we estimate hidden revenue from the no-click rate?
Not responsibly from these studies. Buyers who click and buyers who do not can differ in intent, task, purchase value, and conversion behavior. A reciprocal percentage cannot repair those missing observations.
What should we ask buyers?
Use an optional, neutral question such as “What helped you first find or compare us?” Allow multiple answers and free text. Avoid suggesting that AI must have played a role.
If your current report equates AI referrals with total AI impact, bring the analytics definitions and CRM fields to discuss an evidence-based measurement plan with Semanta.