Knowing When ChatGPT and Perplexity Are Actually Citing You
Most AI-visibility tools sell you a sample from probing chatbots with test questions. Our backend design starts with something better: real referral traffic from AI engines, evidence rather than estimate.
AI visibility — whether ChatGPT, Perplexity, and the rest are actually citing your content — is one of the most talked-about content-marketing questions right now, and most of the tooling built for it works the same way: send a batch of test questions to each AI engine, see if your brand comes up, and report the results as a visibility score. That's useful, but it's a sample, not a fact. It tells you what happened to a specific set of questions on a specific day, extrapolated outward.
The ContentOS backend takes a different starting point, one that's honestly the more interesting design decision here: read your actual referrer traffic first.
Real evidence before estimated evidence
When a visitor arrives at your content from a link an AI engine generated — a click that came from chatgpt.com or perplexity.ai showing up in your referrer data — that's not a sample or an inference. That's a real, already-happened citation converting into a real visit. The system classifies referrers by which AI engine they came from and records that as direct, load-bearing data, deliberately built and described internally as the evidence-first layer: free to collect, and true by construction rather than estimated.
Probe-based detection — the question-and-answer sampling approach most of the industry sells — exists in the same system as a second, explicitly-labeled layer. It costs money to run and it's a sample, so every number it produces is required to carry its sample size and its method alongside the result. Nothing from the probe layer is allowed to present itself as more certain than a sample actually is.
Why the ordering matters
Most vendors in this space lead with the probe layer because it's easier to build a product demo around — ask ten questions, show ten answers. Starting instead from real referrer evidence means the foundation of the measurement isn't "here's what we think might be happening" but "here's what already happened, verifiably." The probe layer still has a place for exploring what you can't observe directly, but it's built to sit on top of real evidence rather than stand in for it.
Where this stands today
This is backend design, not a dashboard you can open yet — the analytics logic exists in ContentOS with no public route wired to it. We're surfacing the approach now because the referrer-first methodology is a genuinely better foundation than what's common in the category, and worth explaining honestly even ahead of general availability.