A buyer asks an AI assistant which product to use, remembers a name, and searches for it later. The sale appears under organic search. Another buyer clicks the assistant's citation and converts immediately. That sale gets an AI referral label. The reporting system has recorded two different paths, even if the same recommendation started both.
Both journeys belong in an attribution investigation. Neither gives you a multiplier to apply to the AI line in your revenue report.
Where the “10x” claim comes from
Graphite's June 2026 account of n8n compares two measurements: roughly 1% of conversions attributed to AEO through last-touch reporting, versus roughly 9% through a post-conversion discovery survey. Graphite describes the difference as approximately tenfold. That is a finding about one company's reporting and survey, not a correction factor for every website. [1]
The public article does not provide the respondent count, response rate, complete question wording, or enough detail to reconstruct a matched comparison of the two groups. Its prose also says organic and paid search were similar across methods, while a chart description reports a substantial Google Search difference. Those details need clarification before the case can support a general estimate.
The useful question is narrower: how much does your observed click path disagree with what customers report about discovery? You can measure that disagreement without pretending either method captures every influence.
Direct traffic cannot supply the missing answer
Google defines direct/none as traffic without a clear referral source. Its documentation lists several possible causes, including direct URL entry, offline documents, missing campaign information, redirects, and interference with tracking. AI is one possible origin among many. [2]
If direct traffic rises during a GEO campaign, investigate it. Do not automatically move the increase into an AI bucket. An email campaign, a conference appearance, a tracking change, and an AI recommendation can all occur in the same month.
Nor does a low citation-click rate tell you how much revenue went elsewhere. Pew's study of 900 U.S. adults found traditional-result clicks on 8% of Google visits with an AI summary, compared with 15% without one. Summary-source clicks occurred on 1% of visits with a summary. The browsing data came from March 2025; researchers reconstructed search results in April. This observational Google study measures clicking, not AI-caused purchases or ChatGPT attribution. [3]
Build three separate records
For a small team, a useful first version is a weekly report with three views. Keep their definitions visible.
| Record | What to collect | What it answers |
|---|---|---|
| Observed referral | Identifiable AI source, landing page, qualified conversion, revenue | Which recorded visits arrived from AI? |
| Reported discovery | Customer's own answer, response rate, conversion ID, date | Which buyers remember AI as part of discovery? |
| Observed visibility | Brand mentions and citations across a fixed prompt sample | Where does the brand appear in the answers we checked? |
Ask the discovery question close to conversion and leave room for an unprompted answer. Store the original text alongside any channel label. If a customer says “a colleague shared a ChatGPT answer,” preserve both parts. Forcing that response into a single channel throws away the most useful information.
Record how many people answered. Forty mentions of AI among fifty survey respondents says something very different from forty among five thousand. Compare respondents and nonrespondents where your data allows: customer type, product, geography, and order size are useful starting points.
Deduplicate before presenting a bigger number
Consider a hypothetical month with 100 qualified leads. Analytics identifies eight AI-referred leads. A survey receives 40 responses; ten mention AI, including three of those eight recorded referrals.
You have 15 distinct leads with at least one recorded or reported AI touch: eight plus ten minus three. You do not have 18, and you have not established that AI caused all 15. You also cannot assume the 25% survey mention rate applies to the 60 people who did not respond.
Report the eight observed referrals, ten survey mentions, three overlaps, and 40% survey response rate separately. This makes the additional evidence useful without hiding the assumptions inside an “adjusted revenue” figure.
Use visibility to investigate a change
A fixed prompt panel can help explain a movement in discovery reports. Keep product category, geography, platform, account conditions, and sampling schedule consistent. Save the answers so a rising mention rate can be checked against actual language: a prominent recommendation differs from a passing mention in a warning.
When visibility and lead quality rise together, there is a reason to investigate further. There is still no isolation of cause. Product launches, public relations, seasonality, and broader brand demand can move both.
Before a large budget increase, define an intervention and a comparison group. For example, improve a selected content cluster while holding a comparable cluster steady, record other changes, and examine outcomes over a period suited to the buying cycle. Treat a small, noisy comparison as directional evidence. A stronger causal claim requires an experiment designed to support it.
Start with a reconciliation table for your business: observed referrals, reported discovery, overlap, survey response coverage, and qualified outcomes.