Search Engine Land urges rigorous prompt tracking for AI visibility checks

Agencies selling AI visibility checks are running into the same problem from every direction: the answers move. Kevin Indig argued on June 10 that the fix is not to abandon prompt tracking, but to treat it like a measurement system that has to be standardized before any client can trust it. The point is bluntly practical for agencies building reports around AI search: one prompt, one screenshot, and one lucky answer do not make a service.

Indig’s case was that AI outputs are probabilistic, not broken. He urged repeated runs, fixed sampling rules, confidence intervals, and journey tracking so teams can separate signal from noise instead of overreacting to a single result. That matters because within-LLM variance can be large, which makes a one-off point estimate unreliable. For agencies, the selling point is not that AI answers can be pinned down perfectly, but that they can be measured well enough to support a repeatable reporting product.

The broader frame is familiar to anyone who watched SEO mature. Indig compared the challenge to weather forecasting and credit scores, two systems that are uncertain but still useful because the methods behind them are disciplined. He also pushed the industry back toward the habits that made rank tracking credible in the first place, including consistent location settings, clean profiles, daily crawls, and other controls that reduce randomness. That same discipline, he argued, now has to extend beyond Google, because prompt tracking is turning into a multi-engine problem across the search and AI landscape.

That shift is already showing up in the platforms themselves. Google Search Central now documents AI features in Search from a site owner’s perspective, including how to appear in AI features and how to measure performance. Microsoft says Copilot Search surfaces concise AI-generated answers with sources, and Microsoft’s own documentation describes GPT models as generating the next words most likely to follow the text that came before. OpenAI’s prompt guidance also stresses that prompt formats need to change with the surface, tools, evals, and UX goals. Together, those cues reinforce Indig’s core argument: AI visibility is becoming a formal measurement surface, but only if the method is tighter than the machine.

The timing helps explain why the message is landing. Search Engine Land has been publishing a cluster of AI-search and AI-visibility pieces, while Search Engine Journal reported that Google is testing dedicated AI Search reports in Search Console for some UK sites. Kevin Indig, who says he sends a weekly Growth Memo to more than 9,000 subscribers and previously led SEO and Growth at Shopify, G2, and Atlassian, is pushing a version of prompt tracking agencies can actually stand behind. The business case is no longer about chasing a mention in an LLM. It is about building a measurement product that survives scrutiny.