Your AI Visibility Score Is Not a Fact. It Is a Moving Measurement

A business checks its visibility in ChatGPT on Monday. It looks healthy.

The same prompts are checked again later in the week and the picture has changed. Different sources appear. Competitors move. A brand that was mentioned disappears from an answer it previously occupied.

It is tempting to treat that as faulty measurement. Increasingly, it is simply the nature of what is being measured.

A September 2026 study by Prefer⁠ asked ChatGPT, Gemini, Claude and Perplexity the same 80 questions three times each. Across the resulting 960 answers, the engines searched the live web for most responses. Yet 72.7% of the 1,329 websites cited appeared in the results of only one of the four engines.

Even repeated runs within the same engine moved substantially. Two runs of the same question shared, on average, only 37% of cited websites in ChatGPT and 35% in Gemini.

For anyone measuring AI discoverability, that creates an important practical problem.

A snapshot can be accurate and still give you the wrong impression.

One prompt, checked once, tells you very little

Traditional rank tracking trained businesses to think in relatively stable positions.

You searched a keyword. Your page ranked fourth. Tomorrow it might be third or fifth, but there was still an understandable position to monitor.

AI answers behave differently.

They are assembled responses, often using live web retrieval, and the sources selected can vary between engines and between repeated answers from the same engine.

The Prefer study found another striking difference: Perplexity cited an average of 19.48 sources per answer, compared with 8.64 for Gemini, 4.58 for Claude and just 3.05 for ChatGPT.

A website could therefore be highly visible in one environment and absent from another without either result being technically wrong.

IndexStream is designed to help businesses understand AI visibility⁠ alongside the wider search, content and technical signals affecting how a website is discovered and understood.

The useful questions are no longer simply:

Did we appear?

They are:

How often do we appear? For which commercially important searches and questions? Which competitors are appearing instead? What is influencing our visibility? And is the underlying health of the website improving over time?

Volatility can hide both progress and problems

Suppose you monitor a set of questions that genuinely resemble what prospective customers ask.

Your brand appears more frequently this week than last week.

That may be progress.

But the more useful analysis goes deeper.

Perhaps the improvement is concentrated around one subject. Perhaps you are increasingly being cited as a source without being recommended as a provider. Perhaps one competitor is losing visibility while another is appearing across an entirely new cluster of questions.

Or perhaps the apparent improvement disappears at the next measurement.

None of those movements can be understood properly from a single check.

Recent citation data reinforces the point. OtterlyAI’s September tracking⁠ found LinkedIn citations rising sharply during the month while YouTube moved in the opposite direction overall, with a particularly large change in Google AI Mode.

Whatever the exact percentages next month, the broader lesson is already useful: the source environment itself moves.

Optimising a website for what an AI engine appeared to favour on one particular day is therefore a poor strategy.

Measure the pattern, not the screenshot

AI visibility becomes more useful when it behaves less like a one-off audit and more like an ongoing measurement process.

The objective is not to pretend that an inherently variable system is fixed. It is to collect enough evidence to distinguish a temporary observation from a meaningful trend.

That starts with the website itself.

IndexStream checks technical, content, search and AI-readiness signals together, identifies what deserves attention and lets you recheck pages and measure progress⁠ after changes are made.

Its growing Visibility Benchmarks⁠ also aggregate anonymised findings across analysed websites, providing another way to see which discoverability problems recur beyond an individual site.

Search rankings gave businesses a language for measuring visibility on the old web.

AI-driven discovery needs a broader measurement discipline.

A single answer is an observation.

A pattern is intelligence.

Check your website with IndexStream⁠ and start identifying what may be limiting its visibility across search and AI discovery.