Being Mentioned by AI Isn't Being Recommended. Here's the Gap.

New research tracked a single brand across 33 prompts on an AI answer engine and found it appeared in responses 42 times, but was actually recommended to the user only 4 times.[1] The same dataset found that 76% of cited URLs showed up for a single week and were never cited again.[1] Most AEO reporting counts appearances. Neither number on its own tells a brand whether it won the moment that matters: the point where the engine names a business as the answer, rather than listing it as one option among several.

What Did the Brainlabs Study Actually Measure?

Brainlabs tracked one brand across 33 prompts on an AI answer engine, reported in Saurce's August 2026 search marketing roundup, and separated two outcomes: whether the brand's content showed up anywhere in the response (an appearance), and whether the engine actually named the brand as the pick for the user (a recommendation).[1] The brand appeared 42 times and was recommended 4 times, a gap of roughly 10x between the two counts.

An appearance is a page or brand name showing up somewhere inside a generated answer, often buried in a list of options or a passing mention. A recommendation is the engine naming that brand as the answer, the thing it's actually telling the user to choose. Reporting that treats these as the same metric is reporting the wrong number.

Why Is "Appeared 42 Times" a Vanity Metric?

A count of appearances treats a brand buried at position nine of a ten-item comparison list the same as a brand named directly as the answer, even though only one of those outcomes moves a buyer. Appearance counts reward volume of mention over quality of mention, and AEO reporting built only around citation counts can look strong while producing almost no actual recommendations.

Buried mentions don't close deals.

What Does the 76% Citation Churn Number Mean, and How Does It Vary by Platform?

76% of the cited URLs in the Brainlabs dataset appeared for a single week and were never cited again.[1] That number describes churn at the citation level: even the appearances that do happen are mostly one-off events, not a stable position an engine keeps returning to.

Citation stability also depends on which engine generates the answer. A separate 2026 study of 34,234 AI responses found that only 11% of domains are cited by both ChatGPT and Perplexity, meaning the two platforms pull from largely different source pools for the same category of query.[2] Optimizing for one platform's citation logic doesn't transfer cleanly to the other.

How Fast Is AI Referral Traffic Shifting Between Platforms?

ChatGPT's share of AI referral traffic fell from 89% to 63% in eight months, while Claude's share grew from 1.4% to 18.5% over the same period.[2] That's a large enough swing to change which platform deserves the first optimization pass this quarter, not just over the next year.

Why Can the Same Content Appear in an Answer Without Becoming the Answer?

Three factors decide whether a citation turns into a recommendation: how the content is structured for extraction, how specific its claims are, and how it frames itself against the alternatives the engine is comparing.

• Structure. Content built as a direct, extractable answer to a specific question is easier for an engine to lift as the primary pick than a page that mentions the topic in passing.

• Specificity. Concrete numbers, named use cases, and stated limitations give an engine something to point to when it's choosing a single recommendation over a list of similar options.

• Framing. Content that positions itself explicitly against named alternatives, rather than describing itself in isolation, gives the engine the comparison logic it needs to pick one answer.

What to Track This Month: Appearance Rate vs. Recommendation Rate

Appearance rate is the share of relevant prompts where a brand shows up anywhere in the response. Recommendation rate is the share of relevant prompts where the engine names that brand as the actual pick, not one option listed among others.

1. Run a fixed set of 15-30 prompts relevant to your category across ChatGPT, Perplexity, and Claude.

2. Code each response two ways: did the brand appear anywhere, and did the engine name it as the recommended choice.

3. Calculate both rates separately. If recommendation rate is far below appearance rate, that gap is the actual optimization target.

4. Re-run the same prompt set weekly for a month to see whether citations that do land are one-off or repeat.

What to Prioritize First

Recommendation rate is the harder number to move, and it's the one that actually correlates with new business. If current tracking only counts appearances, the Brainlabs gap suggests the real number is smaller, in this case by roughly 10x. Start by separating the two metrics in whatever reporting already exists, then test structure, specificity, and framing changes against recommendation rate specifically, not appearance count.

FAQ

What's the difference between an AI citation and an AI recommendation? A citation means a brand's content is referenced or linked somewhere inside a generated answer. A recommendation means the engine names that brand as the answer it's actually telling the user to pick. Brainlabs' research found a brand cited 42 times was recommended only 4 times, a roughly 10x gap between the two.

How is recommendation rate measured? Recommendation rate is typically measured by running a fixed set of prompts relevant to a brand's category, then coding each response for whether the brand was named as the primary pick versus merely mentioned or listed. It requires reviewing the actual answer text, not just counting citations.

Does this vary by platform? Yes. Only 11% of domains are cited by both ChatGPT and Perplexity for the same query set, and referral traffic share is shifting quickly: ChatGPT's share fell from 89% to 63% in eight months while Claude grew from 1.4% to 18.5%, which changes where a recommendation is easiest to earn right now.

See how your brand scores across 4 answer engines.

Sources

1. Brainlabs research, covered in Saurce's August 2026 search marketing roundup — Saurce (August 2026)

2. Goodie AI search traffic report 2026 — higoodie.com (2026)