Measuring visibility in AI: little evidence, ever less independent data
AI assistants bring in a fraction of Google's traffic, and popular AI visibility metrics fail basic data quality tests. At the same time, platforms are taking control of the data that independent tools depend on.
Dashboards showing “AI visibility” have become an argument in marketing budget discussions this year. They promise to show how often a brand appears in ChatGPT or Copilot answers and what that means for sales. We compared data from several independent analyses. They show that the AI channel exists, but most ways of measuring it do not stand up to confrontation with the data.
For the marketing department, this means two questions at once. First: how much the AI channel really brings in. Second: who will provide the numbers on the basis of which it is assessed. Google is making search results data harder to collect, and OpenAI is preparing its own analytics dashboard.
Traffic from AI assistants is still a fraction of traffic from Google
Marie Haynes, an SEO consultant (search engine optimization), analyzed 12 months of GA4 (Google Analytics 4, the standard website analytics tool) data from 100 websites, totaling more than 145 million sessions. ChatGPT brought traffic equal to 0.62 percent of Google's organic traffic, and all language models together - 0.73 percent. The SEO agency SEO Works, analyzing 95 GA4 accounts from September 2025 to August 2026, came to a similar conclusion: organic traffic was 39 times greater than traffic from AI assistants.
The most frequently cited figure concerns conversion. In Haynes's data, visitors from ChatGPT converted at 13.8 percent, compared with 3.02 percent for Google's organic traffic. At the same time, Google accounted for 96.5 percent of all conversions in her dataset, and organic traffic delivered 27 times more of them than all language models combined. Haynes also noted that the rapid growth in the number of sessions from ChatGPT ended in August 2025, around the launch of GPT-5.
SEO Works shows a much smaller advantage: 7.18 percent conversion from AI versus 6.08 percent from organic traffic. On 23 of 63 websites with sufficient volume, AI traffic converted worse, and in B2B companies (selling to other companies) offering products, it reached only 0.62 of the conversion level of organic traffic. The studies differ in sample and industries, so they cannot be averaged. The only certainty is that higher conversion from AI is not the rule. The most practical comparison from the SEO Works study: a 5 percent increase in organic traffic delivered 10.3 additional conversions per website per month, while doubling AI traffic delivered 6.6.
AI's hidden impact is still waiting for proof
Supporters of visibility measurement respond that the click was never the goal. The user sees the brand in a chat response, then types its address or searches for its name on Google. Kevin Indig, author of the Growth Memo newsletter, checked this on 12 GA4 accounts. Direct traffic (visits with no recognized source) grew by 140.9 percent from January to September, but after filtering out unengaged sessions it changed by -1.0 percent. On one website, GA4 recorded 409,000 direct sessions, of which 198 were engaged. The author of the roundup in Search Engine Journal judges that this is bot traffic, not customers.
Indig also conducted 632 correlation tests across four brands, comparing weekly AI visibility with direct traffic and branded searches with a lag of zero to six weeks. In a control test, he shifted the data to intentionally incorrect dates, and some of the false pairings correlated just as strongly as the real ones. His conclusion: the effect may exist, but it could not be isolated. A similar signal comes from an Ahrefs study from May 2025 on 75,000 brands: visibility in AI Overviews (AI summaries in Google results) correlated with brand mentions on the web at 0.664, and with branded search volume at only 0.392.
Rand Fishkin and Amanda Natividad of SparkToro publicly asked in October whether AI visibility is not a marketing waste of money. Neither of them claims, however, that AI does not matter. Fishkin, following Mike King, proposes treating AI responses as a brand channel, something like a digital billboard, and measuring them like a billboard. Natividad adds that attribution that cannot be proven should not be trusted just because it seems right.
A data quality standard that most metrics do not meet
In May 2026, AMEC, the international association focused on communication measurement and evaluation, announced in Dublin the principles for measuring GEO (Generative Engine Optimization, activities that increase brand visibility in AI responses). An accompanying data quality document, prepared under the direction of Rob Key from Converseon, proposes four questions for every dataset. For each, we note, following Search Engine Journal's analysis, how AI visibility metrics perform on it:
- Source: does the data come exclusively from expected sources - bots inflating direct traffic fail this test.
- Accuracy: is the data description complete - mention, citation, and recommendation are sometimes counted as the same thing.
- Real world: does the data reflect human behavior - visibility metrics do not move together with branded searches.
- Trust: will the data convince key stakeholders - a CFO seeing 0.73 percent of traffic stops listening.
AMEC also reminds us that a one-off assessment put on a shelf is decoration, not protection. Still, one caveat should be noted: some of the most striking figures in the document come from a benchmark test sponsored by Converseon, which sells solutions to this problem.
Platforms are taking over the numbers
According to unofficial reports, OpenAI is preparing a dashboard for companies. Juozas Kaziukėnas, an e-commerce market analyst, described on LinkedIn on November 6 a banner that appeared in the ChatGPT interface. The dashboard is supposed to show daily the number of brand impressions in responses and the number of clicks on links to its site. It is to be part of the Business Hub platform for companies that verify business ownership, for example by connecting an account to a Shopify store. It will not show query content or competitor data, so it will not explain why mentions increased or decreased. OpenAI has not officially confirmed this work. Since February 2026, Microsoft has made a citations report in Copilot available in Bing Webmaster Tools, but without the number of clicks and without data from ChatGPT.
At the other end is SEMSTORM, a Polish SEO analytics platform, which is ending operations on December 31, 2026 after 13 years. The company points to further Google restrictions: anti-bot protections, hiding page URLs behind redirects, and limits that make you wait up to two minutes for the next link. The mechanism is confirmed by independent industry sources: in September 2025, Google removed the parameter that allowed 100 results to be fetched in a single query, and in December 2025 it sued SerpApi, a provider of such data for SEO tools.
We could have spent the next few years giving you data with gaps or raising prices sharply. Neither of those paths is what you chose SEMSTORM for - so we decided to say goodbye.
The company did not say by how much its costs had increased. The editors' conclusion: when visibility data is provided by the same company that decides that visibility, independent verification becomes more expensive, and the platform dashboard will tell you “how much,” but not “why.”
What to do before the next budget review
The collected analyses suggest several practical steps for the marketing department:
The best documented investment is also the least spectacular. Websites gaining AI traffic are the same ones growing on Google (correlation +0.557 in SEO Works data), and brand mentions on the web correlate more strongly with presence in AI responses than searches for the brand name. Correlation does not prove causation, but the data suggests that building brand authority works for both channels at once.