AI referrals remain small, while visibility metrics face a validation test
Four datasets published between 2025 and 2026 point to a measurement gap in AI visibility: referrals from AI assistants remain modest beside Google organic traffic, and evidence that visibility reliably drives later branded search or direct visits is limited. The findings do not establish that AI has no influence on buyers. They do challenge marketers to distinguish exposure from measurable business outcomes before using visibility scores to justify investment.
The studies cover different samples and methods, so their results should not be treated as one standardized benchmark. Together, however, they give marketing and analytics teams concrete questions to bring to reporting and budget reviews.
Referral volume and conversion results vary
Marie Haynes analyzed 12 months of Google Analytics 4 data from 100 websites, covering more than 145 million sessions. ChatGPT sent traffic equal to 0.62% of Google organic’s volume; all large language model referrals together represented 0.73%. ChatGPT visitors converted at 13.8%, compared with 3.02% for Google organic. Yet Google accounted for 96.5% of conversions in the dataset, and Google organic produced 27 times more conversions than all LLMs combined.
SEO Works’ analysis of 95 GA4 properties, covering September 2025 through August 2026, found organic search generated 39 times more traffic than AI assistants. AI referral traffic had a higher pooled conversion rate than organic—7.18% versus 6.08%—but the advantage was not consistent across sites. Among 63 sites with enough volume for comparison, AI referrals converted at a lower rate on 23. For B2B product businesses, the AI conversion rate was 0.62 times the organic rate.
The same SEO Works analysis compared potential gains: a 5% increase in organic search generated an average 10.3 additional conversions per site per month, while doubling AI referral traffic generated 6.6. These sample-level results are not a forecast for every business, but they underline why traffic share, conversion rate and total conversion contribution should be reported separately.
Correlation does not settle the influence question
One common argument is that people may encounter a brand in an AI answer, then later visit directly or search for the brand. Kevin Indig examined that possibility across 12 GA4 properties and ran 632 correlation tests across four brands. In an initial comparison, Direct sessions rose on eight properties from August to September. After filtering for engaged sessions, however, total Direct traffic in the January-to-September sample was up 140.9%, while engaged Direct traffic changed by minus 1.0%. One property recorded 409,000 Direct sessions, of which only 198 were engaged.
Indig then compared weekly AI visibility with Direct traffic and branded search at delays of zero to six weeks. In placebo tests using deliberately mismatched dates, some correlations were as strong as those found with the original time alignment. His conclusion was that an effect could exist, but he could not isolate one with the analysis.
Ahrefs’ study of 75,000 brands, published in May 2025, provides another caution against treating branded search as a proxy for AI visibility. Branded search volume had a 0.392 correlation with brand visibility in Google AI Overviews. Branded web mentions correlated at 0.664, while branded anchors correlated at 0.527. The figures describe associations, not proof of cause.
Apply data-quality checks before setting targets
The International Association for the Measurement and Evaluation of Communication (AMEC) launched its GEO Principles in May 2026 and later published a companion paper on data quality in the AI era. Its framework offers a useful lens for teams assessing visibility reports: confirm that sources are what they claim to be, that metadata accurately describes the data, that metrics connect to real-world behavior, and that stakeholders can trust the results.
Those checks matter because a mention, citation and recommendation are different kinds of appearance. Results may also vary by AI tool, prompt, time and whether a session is logged in. A dashboard that does not document its collection method or filter bot traffic can create a precise-looking score without demonstrating that it predicts customer behavior.
For marketers, a practical approach is to keep AI referral traffic as a diagnostic rather than treating it as a standalone performance target. Document the tools and prompts used, define what counts as a mention or recommendation, and rerun a governed prompt set across tools and time. Compare visibility trends with engaged Direct sessions, branded search and pipeline, including lagged comparisons and placebo checks. Self-reported attribution and brand research can add context, but should be treated as complementary evidence rather than definitive proof.
The studies support neither dismissing AI visibility nor accepting every score at face value. They suggest marketers should evaluate AI exposure as a potential brand-channel signal while prioritizing measures that connect to customer behavior and business outcomes. Until those links are validated for an individual organization, a visibility score is evidence of an appearance—not evidence of its commercial impact.
