The most common objection to AI search measurement goes like this: LLM answers are random, so tracking them is pointless. Ask ChatGPT the same question twice and you get two different responses; how can anyone report on that? New analysis published in mid-June 2026 dismantled that objection. AI answers are variable; but they are measurable. …
The most common objection to AI search measurement goes like this: LLM answers are random, so tracking them is pointless. Ask ChatGPT the same question twice and you get two different responses; how can anyone report on that?
New analysis published in mid-June 2026 dismantled that objection. AI answers are variable; but they are measurable. The problem isn’t the technology. It’s the methodology.
Single Checks Are Noise; Repeated Runs Are Signal
Checking whether your brand appears in one AI answer on one day tells you almost nothing. LLMs are probabilistic systems, and a single sample of a probabilistic system is noise by definition.
The fix is borrowed from statistics: repeated runs, fixed sampling rules, and confidence intervals. Run the same prompt set on a consistent schedule, record citation frequency across runs, and report a range rather than a snapshot. Variance stops being a reason to quit; it becomes a number you can defend in a client report.
This is how AI visibility tracking matures from guesswork into a discipline. The businesses building this measurement muscle now will have months of defensible baseline data while competitors are still arguing about whether tracking is possible.
Schema Decisions Just Got Easier to Justify
A second development around the same time: Schema.org, working with Google, now publishes monthly usage statistics showing how many domains use each structured data type across the web.
For anyone who has tried to convince a development team to implement schema, this is practical leverage. Adoption data turns “we recommend this markup” into “here is how widely this markup is used and why it matters.” It also helps prioritise: invest in the schema types that machines actually consume, not the ones that merely look thorough.
The Training Data Fight Worth Watching
Around the same time, major publishers including the AP, the New York Times and Bloomberg demanded that Common Crawl stop scraping their content for AI training datasets. How this resolves will shape which sources future AI models learn from; and by extension, which brands those models know and cite.
The strategic takeaway: the sources AI systems trust are being contested and reshaped right now. Brands that build their own authority signals; original content, consistent entities, citations across credible sites; are not dependent on how any single data pipeline shakes out.
Measure What Matters
At Obsidian Pinnacle, we build AI visibility measurement into every GEO engagement: structured sampling, citation tracking across platforms, and reporting that separates real movement from statistical noise.
Want to know your actual AI visibility; not a one-off snapshot? Book a consultation and we’ll set up measurement you can trust.
Related reading: Google Just Gave You a GEO Dashboard · Google’s CEO Just Confirmed What Smart Businesses Already Know
Sources: Search Engine Land (June 10, 2026)
