SEO Visibility vs. AI Visibility: Why an Index Can't See ChatGPT

By Muntazir · Published 17 October 2026

On 2 April 2026, Sistrix told its own customers something unusual for a company that built its name on one number: that number wasn’t enough anymore. A changelog entry described the company’s focus shifting away from classic rankings and toward mentions and citations in AI answers — a quiet admission that the Sichtbarkeitsindex, the score countless SEO reports have opened with since 2008, had stopped covering the thing clients increasingly wanted answered: does an AI assistant mention this brand at all.

That’s a strange thing for a visibility tool to concede. It’s also the right one, because the two things share a name and nothing else.

What does a classic visibility index actually count?

A breakdown of the method lays out three steps. First, the tool crawls Google’s results for roughly a million tracked keywords every day, keeping the top 100 positions for each. Second, it weights each ranking by that keyword’s search volume, the likely intent behind it, and the click-through rate a result at that position typically earns — a first-place ranking on a high-volume term moves the score far more than a first-place ranking on a rarely searched one. Third, it sums the weighted values across every URL on a domain into a single score. Sistrix itself says companies have leaned on that number since 2008. Semrush’s and Ahrefs’ own visibility metrics are built on the same basic idea: rank positions, weighted by volume, aggregated per domain.

It’s a genuinely useful number for the thing it was built to track — how much of Google’s organic search real estate a domain occupies. It was built years before anyone needed it to track anything else.

Why was an AI answer never part of the formula?

Because the formula has no slot for one. Every input into a classic visibility index is a ranking position on a results page — a number between 1 and 100, tied to a specific keyword. A chat answer from ChatGPT or Gemini isn’t a ranked list of ten blue links; it’s one piece of text, assembled from whichever sources the model drew on, with no “position four” to assign a weight to. The method’s own documentation never mentions AI Overviews, ChatGPT or any AI feature, and that’s not an oversight — there’s genuinely nothing in the formula’s three steps that an AI answer could feed into.

That gap matters more now than it used to, since a growing share of search queries resolve into a generated answer before a user ever sees a classic results page.

Does this site’s own data show that gap?

A narrower, honest version of it, yes. Search Console records for this product’s own domain show 126 impressions for the query “ai visibility” at an average position of 1.4 — essentially the top result — over recent months. Run that domain through a classic visibility index and the score would likely sit close to zero, because the index’s basket holds roughly a million generic, competitive keywords built up over nearly two decades of tracking, and a domain’s real ranking on one specific term barely registers against a basket that size, however well that one term is doing.

That’s not proof of anything about ChatGPT specifically — Search Console measures Google’s own organic results, the same ground the classic index covers, just through a narrower lens. What it demonstrates is the mechanism: a real, measurable ranking can sit entirely outside what a fixed-keyword-basket index was built to notice. The gap between a rankings index and an actual AI answer is the same shape, just wider, because an AI answer isn’t drawn from that basket at all.

Did Sistrix build something to see the part the index misses?

It did, and the timing says something: the shift arrived as AI Overviews started appearing in roughly one in five German searches, cutting into the click-through the classic index was built to estimate in the first place. The Q1 2026 changelog describes a separate AI-tracking module covering four platforms — Google AI Overviews, Google AI Mode, ChatGPT and Perplexity — producing its own AI Visibility Index that sits alongside the classic score rather than replacing it. A team that keeps watching only the original Sichtbarkeitsindex still won’t see any of this; the new number lives in a different part of the product.

Is a sampled AI number any more complete?

Not entirely, and that’s worth saying plainly rather than glossing over. Sistrix’s AI tracking works by regularly sending a defined set of prompts to each platform and reading the answers — a sample of plausible questions, not a reading of every question a real customer might type. That’s a limitation shared by every AI-visibility product currently on the market, including the model-opinion step of a deeper technical review, which treats one sampled response as a data point rather than a verdict for exactly this reason. A free scan built on the same engine skips model probes entirely for the same reason the sampling matters: a single answer, asked once, proves less than it looks like it proves.

A rankings index and a prompt sample are both proxies, counting different things through different windows. The question that actually matters — what does an assistant say when it’s asked about this brand, right now — still has no single number that answers it completely, from any vendor, including the ones racing hardest to build one.