Anonymised Research Case Studies / Description vs. recommendation

Accurate AI descriptions did not translate into recommendations.

In one AI Brand Intelligence Report™ measurement, AI engines described the company specifically and correctly in nearly every deep-dive answer - then, asked which company to trust or choose, did not name it once.

Being described accurately and being recommended are not the same question, but they are often treated as one. This measurement separated them: it tested what AI engines say when asked to describe the company in detail, and, separately, what they say when asked who a buyer in this category should trust or choose.

36 prompts were run against each of four engines - ChatGPT, Gemini, Perplexity and Claude - producing 144 stored responses. Two subsets matter here and were kept separate throughout: six deep-dive prompts asking engines to describe the company in detail (24 responses), and three recommendation and trust prompts asking engines to name who a buyer should choose (12 responses across the four engines). Every response was stored with its prompt, engine and timestamp.

Two supporting facts sit behind that finding. First, the one deep-dive response that did not produce a description was a clarifying question, not an inaccurate or negative answer - so the description-accuracy figure is not weakened by any wrong information, only by one non-answer. Second, the twelve trust-query responses come from three separate questions, each run against all four engines, and the company appears in none of them - the absence is consistent across the question set, not a single engine's quirk.

Observation: when engines are asked to describe the company, they do so accurately. When engines are asked who a buyer should trust or choose in this category, the company is not part of the answer.

Interpretation: accurate description and AI recommendation appear to be two different outcomes, not one continuum - being known does not, on this evidence, mean being surfaced at the point a buyer is deciding whom to trust. This measurement does not show that AI never recommends the company, or that recommendation is impossible; it shows that in this specific, three-question trust subset, it did not happen once. Extending that finding beyond this subset would go past what the evidence supports.

The report's 90-day roadmap placed "contest the category definition, not the position within it" as a high-priority action for the first 15-45 days - on the reasoning that the gap sits upstream of ranking or sentiment, in how the category itself is framed in answers. A second, medium-priority action called for strengthening organisation-level entity binding over the following 30-60 days. This reflects what the report recommended at the time of delivery. SEOWEBI has no record of any of these actions, or a remeasurement, having taken place since.

This report's overall confidence label is Incomplete: two groups of sub-signals - a minority of the total 100-point score - could not be verified from available data and were scored conservatively as zero rather than estimated. Both the deep-dive and the trust-query layers described above were fully and completely measured, with no gaps: all 24 deep-dive and all 12 trust-query prompt-engine pairs returned a stored response. What remains unknown is whether this pattern has since changed: this case study describes AI-generated answers recorded during a single measurement window in 2026, not current or ongoing AI behaviour.

This finding is one expression of the pattern SEOWEBI calls Incumbent Squeeze - a brand crowded out of category-defining answers despite accurate description once it is named. All three case studies in this collection identify the same underlying pattern from a different angle.

More anonymised research: Pricing discrepancies · Comparison framing · All case studies