Being mentioned is not being recommended: brands are measuring the wrong AI metric

Counting one thing and calling it visibility

The measurement problem in AI search is no longer whether the channel matters. It is that most organisations are counting one thing, calling it visibility, and drawing conclusions from a sample of one model on one day.

The shortlist is now generated, not browsed. Across four engines the average commercial query surfaces 3.7 named brands.

Semrush's study of 50,000 brands across 1,094 US categories, tracked monthly in ChatGPT from January to June 2026, found that mentions and citations diverge sharply. Only 21% of the most-cited domains in a category were also the most-mentioned brand, and the two signals correlated slightly negatively. Only 15% of categories analysed had a clear brand winner at all.

In plain terms: the site an assistant links to and the company it recommends are frequently different organisations. Optimising to be a source is a different job from optimising to be the answer, and a dashboard that merges the two will report progress that never reaches the sales pipeline.

Every engine has its own memory

A Q1 2026 tracker of 8,400 commercial prompts across ChatGPT, Perplexity, Gemini and Claude found brand density varies widely by engine, and that citation is not a vanity metric: brands cited consistently saw a 23.4% lift in branded search volume over the following 30 days, rising to 41.2% at 90 days.

The divergence between engines can be extreme. An INSEAD study cited in 2026 frameworks found one consumer detergent brand holding close to 24% share of model on Meta's Llama and under 1% on Gemini. Same brand, same category, same week. Measuring a single engine and calling the result "our AI visibility" is a sampling error with a budget attached.

And it does not sit still

EMARKETER principal analyst Nate Elliott has flagged the volatility problem: ask Google the same question ten times and the results are broadly consistent, but between 40% and 60% of cited sources change month to month across Google's AI Mode and ChatGPT. A single check is a snapshot of weather, not climate.

Meanwhile the click that used to justify all of this keeps disappearing. Pew Research found in July 2025 that users click a traditional result only 8% of the time when an AI summary is present (March 2025 data; 900 US adults, 68,879 searches), and Datos put the share of Google searches ending without any external click at 57% in the third quarter of 2025.

• 14% of brands have an AI visibility strategy at all, on the Q1 2026 tracker's own count. The rest are improvising.
• 94% of 250 surveyed enterprise C-level executives plan to increase spending on AI visibility in 2026, per a report cited by Entrepreneur in August 2026.
• 32.5% of marketers say they do not know how to measure it, in a HubSpot study referenced in the same piece. Spend is arriving ahead of method.

Smaller businesses are exposed first

For local and small businesses the shift is sharper than the enterprise numbers suggest. BrightLocal recorded consumer use of AI to find local businesses rising from 6% to 45% during March 2026, and found 63% of active AI users trust AI recommendations for local businesses, though 88% still fact-check them afterwards.

The reporting layer lags badly. One 2026 analysis found around 70.6% of AI referrals misclassified as direct traffic in standard GA4 configurations, with the channel undercounted by an estimated three to four times, partly because paid ChatGPT accounts and Gemini's deep research mode pass no referrer data.

A workable method

The approach analysts converge on is hybrid: automated tracking for broad patterns, paired with monthly manual checks in fresh chats on the prompts that actually drive revenue. Build a library of 20 to 100 prompts weighted toward solution and comparison questions rather than brand-name queries, since asking a model what it thinks of you only proves it has heard of you.

Run them across at least ChatGPT, Gemini or Google's AI Mode, Perplexity and Claude. Log four fields: named or not, position in the list, source cited, and whether anything stated about you is wrong. Repeat monthly, and read the trend rather than the reading.

For the step-by-step version of this method, see the field guide chapters on measuring AI visibility without lying to yourself and how an AI assistant decides who to recommend.

Sources

Semrush (50,000 brands across 1,094 US categories, ChatGPT, January to June 2026); Visionary AI Search Visibility Tracker (8,400 prompts, 14 sectors, Q1 2026); INSEAD share of model study, cited in 2026 measurement frameworks; Nate Elliott, EMARKETER (citation volatility); Pew Research (click rate with AI summaries, July 2025); Datos (zero click share, Q3 2025); Entrepreneur, August 2026 (enterprise spending survey of 250 executives; HubSpot measurement figure); BrightLocal (local AI adoption and trust, 2026); GA4 attribution analysis, 2026.