By Yanisa Chubplun, Integrated Communications Manager, Foundeast
A screenshot of ChatGPT mentioning your company is not an AI visibility report.
It records one answer, on one day. It cannot show whether the same answer would appear tomorrow, whether competitors appeared alongside you, whether the description was correct, which sources shaped it, or whether anyone who saw it went on to visit your website. Without a baseline, repetition, or context, one answer is closer to an anecdote than evidence.
A Mention Is a Signal, Not a Measurement System
Mentions are useful signals. The problem is treating one observation as a finding. AI visibility measurement begins when the same questions are asked in the same way over time, and the results are recorded consistently enough to compare.
“A mention is a useful signal. On its own, it is not a measurement.”
Start With Questions, Not Your Brand Name
Asking “Tell me about Company X” tests whether an AI system knows your company exists. It says little about unbranded AI visibility because the question already supplies your company name.
A more useful test relies on a fixed set of prompts drawn from real buyer questions: the category, the problem, the market or city, a specific use case, comparisons between providers, and the criteria behind a decision. A logistics firm might monitor “Which freight forwarders handle cold-chain shipments from Thailand to Vietnam?” rather than ask about itself.
Keep the prompt set stable so changes in AI visibility mean something. Where relevant, run it across more than one platform. Do not expect matching results; differences between platforms are useful information.
Repeat the Test
Generative systems do not return identical answers every time. Ask the same question twice, and the brands, sources, and wording can shift.
A 2026 preprint by Ronald Sielinski tested repeated searches across several generative search systems and found substantial variation in which domains were cited. The research suggests that a single test can make AI visibility look more precise than it really is.
It is early research, but the practical lesson is useful: monitor AI visibility as a pattern rather than treat it as a fixed ranking.
Measure Presence, Then Accuracy
Presence asks whether your company appears in monitored answers. Accuracy asks whether it is described correctly when it appears: the right services, markets, and category; current information; and positioning that reflects how the business actually competes.
A brand described with a discontinued service, an old office location, or the wrong category may have AI visibility, but that visibility is less useful when the description is wrong.
Look at the Sources Behind the Answer
Where platforms show citations, record them. Which domains appear? Are they company-owned or external? Are they current, or is an outdated directory listing or years-old press article still shaping the description?
A citation does not prove that a source caused an answer. Treat it as diagnostic evidence: a way to spot pages that may need correcting or refreshing, rather than a causal map of AI visibility.
Add Competitive Context
A brand appearing in a dozen answers tells you little on its own. Run the same prompts for relevant competitors and, where useful, category leaders. Useful internal measures include presence across monitored prompts, relative share of presence within the comparison set, and how often competitors appear alongside you.
These are measures defined by your AI visibility reporting method, not official platform rankings.
Measure What Platforms Actually Expose
Google says its Search Console generative AI performance report reached all websites worldwide as of 31 August 2026, though a site with too few generative AI impressions may still not see it. It reports impressions from AI Overviews and AI Mode, broken down by page, country, date, and device. It is useful first-party evidence, but it does not show the prompts behind those impressions or their wider influence.
ChatGPT, by contrast, adds utm_source=chatgpt.com to referral links, so analytics tools can identify visitors who clicked through from ChatGPT search results.
The two are not interchangeable. Google reports impressions within its own ecosystem; ChatGPT tagging measures visits that reached your site after a click, not every appearance in an answer. Neither belongs in a single, universal AI visibility score.
What Not to Treat as a KPI
- A one-off screenshot
- One prompt, tested once
- Raw mention counts without context
- Citation counts without checking relevance or source quality
- A proprietary AI visibility score with an unclear methodology
Report Trends, Not Certainty
A useful AI visibility report follows a cycle: monitor, compare, diagnose, improve, and retest. It reports direction and confidence without claiming more precision than the evidence supports.
The useful question is whether your brand is becoming more visible, more accurately represented, and more commercially useful across the questions that matter. This is the question we pose daily for our Foundeast AI discoverability.
If your current AI visibility reporting starts and ends with screenshots and mention counts, build a repeatable baseline first. Get the measurement right before deciding what to optimise.
Sources
- Google Search Console Help, “Generative AI performance report (Search)”: https://support.google.com/webmasters/answer/16984139?hl=en
- Google Search Central Blog, “Introducing Search generative AI performance reports” (June 2026): https://developers.google.com/search/blog/2026/06/gen-ai-performance-reports
- OpenAI Help Center, “Publishers and Developers – FAQ”: https://help.openai.com/en/articles/12627856-publishers-and-developers-faq
- Sielinski, R. (2026), “Quantifying Uncertainty in AI Visibility: A Statistical Framework for Generative Search Measurement”, arXiv preprint 2603.08924: https://arxiv.org/abs/2603.08924