Cited in 98.4% of AI answers, yet weak on evidence in 73.7%: why visibility alone won't win the deal

A company that showed up in 98.4% of answers from five AI platforms, and was named the top recommendation or a finalist in 93.2% of them, had weak or missing evidence behind important claims about itself in 73.7% of those same answers. The finding comes from a recent project that evaluated 190 answers to 38 detailed questions built around purchase criteria.

The case was laid out in an analysis published by Search Engine Journal on November 11. For the past two years, the conversation around AI search has centered on whether a brand gets cited in ChatGPT, Gemini, Claude, Perplexity and Google. The company in question already appeared often on AI-generated shortlists. The concern was what happened next, as the buyer kept the conversation going, and a sales pipeline that had been shrinking.

The questions were deliberately different from broad "best vendor" prompts. They reflected what buyers use to narrow their choice: workflow fit, integrations, regulatory requirements, interoperability, operating conditions, data analytics and platform breadth. In 15.8% of runs, some association, decision criterion or competitive differentiator showed a gap rated significant or critical.

Visibility and decision readiness measure different stages

According to the analysis, the numbers don't contradict each other. The AI models knew the company belonged in the conversation, but they were far less consistent at backing up why it should stay there once the buyer applied specific criteria. Presence gets a brand onto the list; the available information decides whether it stays.

Weak evidence didn't always mean a missing capability. In many cases the company seemed to have the feature, but its published information didn't explain it well. There were promises of an end-to-end process without showing the steps, the connections and how data moved between systems, and claims of efficiency gains without proof from internal testing or customer testimonials.

Fragmentation was another issue. The model had to connect the product page to the integrations page and to a third page on interoperability just to infer whether a requirement was met. Every extra inference opens the door to an incomplete picture, or to a competitor with clearer evidence becoming the safer recommendation.

Your own documentation is enough in some cases

The analysis points out that evidence coming only from the company itself was not a flaw in its own right. Official product documentation can be the best proof that a feature or integration exists. Claims of superiority, effectiveness, adoption, scale, measurable results or differentiation, on the other hand, benefited from independent confirmation by customers, the industry, government or analysts.

The piece also takes aim at content that merely matches what competitors say. If everyone claims to be integrated, scalable and efficient, that helps establish a presence in the category, but it does little when the AI needs to understand what each promise actually covers. And it warns against manufacturing content that only looks like evidence, such as longer pages, shallow diagrams, weak surveys and AI-generated case studies.

What this changes for brand teams

Measuring only the AI's first answer leaves out the stage where the decision gets deeper. The author acknowledges they could not prove the evidence gap caused each lost sale, but argues that a measurement model based on visibility alone would never have surfaced the problem. Closing these gaps takes knowledge from product, engineering, sales and customer-facing teams, which the content team organizes and makes explicit, without inventing anything.