Most AEO conversations fixate on mentions. How many times does an AI Overview cite your brand? Does Perplexity pull from your site? Does ChatGPT recommend you when a user asks a category question? These are reasonable things to track. But they are outputs, not inputs. And if the inputs are wrong, more AI processing just makes them more wrong, faster.
Mallory Gray of Skydeo made this point clearly in a recent Search Engine Journal piece: AI models amplify the data you feed them. If that data does not accurately describe who your buyers are and what drives their purchase behaviour, the AI does not correct it. It scales it. That has direct implications for how brands should think about earning and measuring visibility in AI search surfaces.
The Distinction Between Being Mentioned and Being Recommended
There is a meaningful difference between an AI engine surfacing your brand as a passing reference and recommending it to a user who is ready to buy. The former can happen through sheer content volume or link equity. The latter requires the model to have enough signal to understand that your brand is relevant to a specific type of person with a specific type of need.
AI search engines - Google AI Mode, ChatGPT, Perplexity, Gemini - are not simply returning the most-linked result. They are attempting to answer questions in ways that match intent. When a user asks which accounting software is right for a 20-person professional services firm, the model needs contextual signals to reason about which brands are genuinely suited to that profile. Mention count does not provide that. Audience-aligned content and entity signals do.
This is where the gap between traditional SEO thinking and AEO practice becomes most visible. Ranking in organic search rewarded authority signals - links, domain strength, keyword coverage. Earning a recommendation in an AI-generated answer requires the model to understand not just what you do, but who you serve and why that match is credible.
What Bad Audience Data Actually Does to AI Visibility
If your site's content, structured data, and off-site presence describe your audience inaccurately - or not at all - AI models build a representation of your brand that reflects those gaps. A B2B software company whose content is written for a generic technical audience, rather than for the specific buyer persona who actually converts, is giving AI engines the wrong brief. The model then associates the brand with a broad, low-intent signal set.
The amplification problem Gray describes is not hypothetical. When AI agents are tasked with shortlisting, comparing, or recommending products and services, they draw on whatever signals are available. Vague signals produce vague recommendations - or no recommendation at all. Precise signals, grounded in real buyer behaviour, give the model something concrete to reason with.
This is why first-party data quality is not just a paid search concern. The signals you publish about who your customers are - through case studies, testimonials, structured content, and schema - contribute to how AI models understand your relevance. It is essentially the brief you give to every AI engine simultaneously.
How to Structure Content Around Buyer Signals, Not Just Topics
Most AEO content strategies focus on question coverage. Answer the questions users ask. Use headers that mirror natural language queries. Structure content so it is easy to extract. All of that remains valid. But it addresses the retrieval layer, not the relevance layer. For AI engines to recommend your brand to the right person, your content needs to encode audience specificity alongside topic coverage.
In practice, this means creating content that names the specific buyer profiles you serve - industry, company size, role, problem type - rather than writing for the broadest possible audience. A professional services firm that publishes detailed case studies segmented by client type is giving AI engines a richer signal set than one that publishes generic thought leadership. Schema markup on those case studies, including industry and audience type fields, extends that signal to structured data that crawlers and AI bots can read directly.
Digital PR plays a role here too. Citations from sector-specific publications, industry bodies, and credible vertical sources tell AI models something about context - not just that your brand exists, but in what world it operates and who considers it relevant. A citation in a procurement journal signals something different to an AI model than a citation in a general business blog.
Measuring Whether Your Audience Signals Are Working
The measurement challenge in AEO has always been attribution. AI-referred traffic frequently arrives in GA4 without a clear source, particularly from ChatGPT and Perplexity where referral data can be inconsistent or absent. But the question Gray's argument raises is a harder one: not just whether AI engines are sending traffic, but whether they are sending the right traffic.
If you are seeing AI referral sessions but they are converting at a fraction of your organic or paid rates, that is a signal worth investigating. It may indicate that AI engines are surfacing your brand for the wrong queries - or to the wrong audience profiles - because the signals you have published do not accurately represent your buyers. Tracking AI-referred sessions through to CRM outcomes, segmented by source (ChatGPT, Perplexity, AI Overviews), gives you a basis for that analysis.
UTM parameters on the few sources that pass referral data, combined with direct CRM tagging for leads that mention AI search in qualification calls, gives you a practical baseline. The goal is not just volume from AI surfaces - it is qualified volume. Audience signal quality is what determines whether those two numbers converge.
The Practical Implication for AEO Programmes
AEO and GEO programmes that focus purely on content structure and schema are addressing necessary but insufficient conditions. Getting your content into AI extraction pipelines matters. Ensuring crawlers and AI bots can access your site via a properly configured robots.txt and an llms.txt file matters. But those are table stakes. What differentiates brands in AI-generated recommendations is the precision of the audience signal they publish.
The practical audit question is this: if an AI engine read everything your brand has published online, would it understand who your buyers are with enough specificity to recommend you to the right person? Not just what you sell, but to whom, in what context, and why that match is credible. If the answer is no, no amount of schema or content volume will fix it.
Gray's core argument - that AI amplifies bad data rather than correcting it - should reframe how AEO investment is prioritised. Before optimising the structure of your content for AI extraction, it is worth auditing whether the substance of that content accurately reflects your audience. That is the signal AI engines will scale.