Most website analytics tools were built to measure clicks from search results. AI tools — ChatGPT, Gemini, Google’s AI Mode, Perplexity, Claude, Copilot — often answer a question directly, with no click at all. That’s created a real measurement gap, and a fair amount of confusion about what’s actually worth tracking as a result.
David Khim’s reporting for Search Engine Journal is the clearest breakdown of this gap so far, built from real internal dashboard analysis rather than theory.
Start with prompts, not tools
The most grounded advice on this: build your tracking around the actual language customers use, pulled from real sales calls and customer research — not the auto-suggested prompts an AI visibility tool generates for you. A generated prompt list optimizes for what’s easy to measure. Real customer language reflects what people actually ask, which is the thing that matters.
Track visibility per AI model, not as one blended number
ChatGPT, Gemini, Perplexity, and the others each pull from different sources and serve different user bases. A brand can show up consistently in one and be invisible in another, and averaging that into a single “AI visibility score” hides exactly the information you’d want to act on.
The attribution problem is bigger than most site owners realize
An estimated 80-90% of leads that actually arrived via an AI platform get mis-tagged in standard analytics as “organic” or “direct” traffic. That’s not a rounding error. It means most businesses are almost certainly getting real value from AI-driven traffic today and have no visibility into it, because their analytics setup was never built to distinguish it. The fix isn’t a better dashboard, it’s a manual one: ask new leads directly, in a form or during onboarding, “how did you hear about us?” and actually capture the answer.
What to watch without over-trusting
A few metrics are worth monitoring as context, not as KPIs to optimize toward directly:
– Citations — how often your content gets cited on pages AI tools frequently pull from. Useful signal, but a citation doesn’t guarantee an actual recommendation in an AI response.
– Sentiment — how AI tools characterize your brand when asked. Treat this as customer-experience feedback, not something a marketing campaign can directly move.
– Referral traffic from AI tools — genuinely too volatile to rely on. A single change to how one AI tool cites sources can swing this number without anything on your end changing.
What actually counts as success
The article’s own framing is the useful part: judge AI-sourced traffic the same way you’d judge paid ad spend, by leads, pipeline, and revenue it actually produces, not by a visibility score that doesn’t map cleanly to business outcomes.
A Concrete Way to Actually Do This as a Small Business
The advice above is right but abstract until it’s turned into something a one-person or small-team operation can actually run without a marketing department. The version that fits a small hosting customer looks like this: add one required field to every contact form, quote request, and signup flow — “How did you hear about us?” — as free text, not a dropdown with pre-written options. A dropdown with “Google,” “Referral,” and “Other” as the only choices trains customers to pick the closest-sounding option rather than tell you they asked ChatGPT, which defeats the entire point of asking. A free-text field occasionally gets a genuinely useful answer like “I asked ChatGPT for cheap PrestaShop hosting and it mentioned you,” which no analytics dashboard would have surfaced on its own.
Once that data starts coming in, the second habit that matters is running the same 15-20 real customer questions through ChatGPT, Gemini, and Perplexity once a week or once every two weeks, logging by hand which platform mentioned the business, whether it was named directly or just cited as a source, and which competitor showed up instead when it wasn’t. This is the same “brand-name-free prompt” discipline that applies to any AI visibility check: a prompt like “is AllCloudHost good” only measures reputation among people who already know the name, not discoverability for someone comparing options cold.
The Citation Number Worth Knowing
One additional data point from the same reporting is worth adding to the citation caveat above: across the sources AI tools actually cite when answering a question, only about 22% are pages the business itself owns and controls. The other roughly 78% are directories, review sites, comparison articles, forum threads, and other third-party pages that happen to mention the business. That reframes what “improving AI visibility” actually means in practice. Optimizing a business’s own site is necessary but nowhere close to sufficient, since most of what an AI system cites when discussing a business isn’t the business’s own website at all. Real, accurate, consistent mentions on the third-party pages an AI system already trusts (directories, genuine reviews, industry roundups) matter as much as anything published on-domain, which is a different, less comfortable lever than the ones most SEO advice focuses on.
Why This Traffic Looks Like Direct Traffic in the First Place
The mechanical reason for the mislabeling is straightforward once it’s spelled out: Google Search passes a referrer when someone clicks through to a site, which is how analytics knows to label that visit organic. ChatGPT, Perplexity, and Claude generally don’t pass the same kind of referrer data, and a visit that arrives with no UTM parameters and no referrer header gets bucketed as direct traffic by default in most analytics setups, the same bucket a visitor gets sorted into when they type a URL straight into their browser. The analytics tool isn’t guessing wrong so much as it’s applying an old rule (no referrer means direct) to a new kind of traffic the rule was never built to handle. That’s also why a manual “how did you hear about us” question outperforms any dashboard fix here: it’s answering a question the tracking pixel structurally cannot answer on its own.
Why This Matters More for Hosting Customers Specifically
A hosting company’s own customers are disproportionately the kind of business where this gap shows up hardest: a WooCommerce store, a PrestaShop shop, a small agency’s client site, all competing in categories (“best hosting for X,” “cheap Y provider”) where an AI assistant increasingly answers the comparison question directly rather than sending a searcher to ten blue links to compare themselves. A business that ranks well organically but has never checked what ChatGPT or Perplexity says when asked the same buying question is optimizing for a channel that’s shrinking relative to one it has no visibility into at all.
None of this requires an enterprise AI-visibility platform to start. It requires one form-field change, a recurring half-hour of manually asking the same real questions to three AI tools, and treating whatever comes back the way an ad channel gets treated: judged on whether it produces real leads and revenue, not on whether a dashboard shows a rising line.
For a small business running its own site, that’s genuinely good news: you don’t need a specialized AI-SEO tool to act on this. You need to ask new customers one extra question, and treat AI mentions as one more channel to measure by results, not as a separate discipline with its own scoreboard.
The flip side of measuring visibility, actually predicting which pages get named when an AI assistant answers a question, is a related but separate problem worth its own read: what actually predicts a ChatGPT mention.