Measuring AI referrals first, then building content, trust, and technical foundations against what the data showed.
For one client, the question wasn't just "are we ranking on Google," it was "do AI systems even know we exist, and for what." Traditional search visibility doesn't tell you anything about whether ChatGPT, Copilot, or Perplexity ever mention you, or under which prompts.
Before recommending anything, I needed real data on AI referral behavior, not assumptions. I used a combination of tools, each covering a different gap:
Microsoft Clarity, which surfaces referral traffic from Microsoft Copilot specifically, something most standard analytics setups miss entirely
HubSpot's AEO tooling, for tracking answer engine visibility
GA4, for the traffic and referral side once visitors actually arrived
Manual testing, running the same prompts directly in ChatGPT and Perplexity to see, firsthand, what they said and which sources they cited
Seeing Copilot referral data specifically through Clarity was a genuine turning point. Most AI visibility conversations focus on ChatGPT alone, and this made it clear that different AI platforms needed to be measured separately, not lumped into one generic "AI traffic" number.
The audit surfaced something specific: the site was already ranking for a set of prompts that had nothing to do with its actual focus area. That traffic was real, but it wasn't the traffic that mattered for the business.
Rather than chasing every possible fix, the work was organized around three areas, each answering a different question:
Does the content actually answer the questions people (and AI systems) are asking, structured in a way AI can extract cleanly?
Is there a clear, verifiable reason for an AI system to treat this source as credible, an identified author, references, evidence of expertise?
Can AI systems actually read and parse the site correctly in the first place, structured data, a properly configured llms.txt, and accurate crawling instructions?
The existing prompt-level ranking was left in place rather than disrupted, since it was still driving some traffic, while new content specifically targeted the topics that actually mattered. Six blog posts were published in one month, each structured around a real question and a direct, extractable answer rather than traditional keyword-led copy.
Each piece of content carried a clearly identified author and cited references, rather than being published anonymously. This is a direct EEAT (experience, expertise, authoritativeness, trustworthiness) signal, and it's one AI systems weigh heavily when deciding whether to cite a source.
An llms.txt file was built, the robots.txt was updated to reflect how AI crawlers should be treated, and FAQ content was properly marked up with structured data so the question and answer pairs were machine readable, not just visually formatted.
AI visibility isn't one thing you fix, it's three separate questions that all have to be answered together: is the content actually useful and extractable, is there a real reason to trust the source, and can AI systems technically parse the site at all. Measurement mattered just as much as the fix. Without checking Copilot referrals specifically through Clarity, that channel would have stayed invisible, since it doesn't show up the same way in standard analytics.
This work sits under AI Search & SEO: GEO/AEO strategy, structured data, and content architecture.
See also: Google Ads revenue growth, a different channel from the same evidence-led, measure-first approach.
Standard SEO tools track search engine rankings, but they don't show whether AI systems like ChatGPT, Copilot, or Perplexity reference a site. Measuring this requires a combination of tools: Microsoft Clarity for Copilot referral data specifically, dedicated AEO tracking tools, GA4 for downstream traffic behavior, and manual prompt testing directly in AI chat interfaces.
Content, trust, and technical. Content covers whether material answers real questions in an extractable format. Trust covers whether there's a verifiable reason for an AI system to treat the source as credible, such as identified authorship and references. Technical covers whether AI systems can actually parse the site, through structured data, a correctly configured llms.txt file, and appropriate crawler instructions.
AI systems weigh trust signals, including clear authorship and cited references, when deciding which sources to cite. This maps to the EEAT framework (experience, expertise, authoritativeness, trustworthiness), and anonymous or unattributed content is a weaker citation candidate than content with a clear, verifiable author.
An llms.txt file gives AI crawlers explicit guidance about a site, separate from traditional robots.txt crawl instructions. It's one part of the technical foundation needed for AI systems to correctly read and understand a site, alongside structured data and accurate crawler permissions.
Not necessarily. In this case, traffic from prompts unrelated to the client's actual focus was left in place rather than disrupted, since it was still real, functioning traffic. New, more relevant content was built alongside it instead of replacing it outright.