Can You Trust a Single AI Search to Measure Visibility?

A study of 14,472 AI citations found business websites led Gemini's local results, but repeated searches rarely returned the same sources.

What Happened

Steady Demand analyzed 14,472 citations from 1,487 Gemini local search queries across 50 U.S. metro areas and 10 service categories. Business websites captured nearly 60% of Gemini citations, beating directories, review sites, and forums combined. Reddit came second at 13.7%. Co-founder Ben Fisher flagged what he called Grounding Drift: repeating the same query surfaced overlapping sources only 40% of the time, and the same top business appeared just 7% of the time.

The Numbers in Context

Google's traditional local pack returns the same top listing about 90% of the time. Gemini returns the same top business 7% of the time. Cross-platform overlap is worse: Gemini and ChatGPT cited the same domains only 8% of the time and recommended the same top business 4.2% of the time. ChatGPT leaned on Reddit and directories, while Gemini favored owned websites.

Why This Matters for B2B Marketing Leaders

If you run demand generation in HR Tech or FinTech, this study reframes how you measure AI search visibility. A single prompt test in Gemini or ChatGPT tells you almost nothing durable. With 60% of Gemini citations flowing to business websites, your owned content strategy still matters more than chasing directory placements. But the 8% cross-platform overlap means winning in Gemini does not translate to ChatGPT, Perplexity, or Copilot. Your team needs a monitoring cadence measured in dozens of queries per platform per week, not one-off spot checks, or you will misread your position in categories where analyst influence and peer forums shape shortlists.

The Starr Conspiracy's Take

Grounding Drift is the story here, not the citation share. B2B buyers in HR Tech and FinTech rarely ask an AI engine the same question twice, so probabilistic answers create probabilistic pipelines. You should treat AI visibility as a distribution, not a ranking, and build your measurement stack accordingly. That means query panels, platform-by-platform tracking, and content engineered for citation across owned domains, Reddit threads, and analyst coverage. Our take on answer engine optimization for B2B walks through how to structure content so it earns citations across engines rather than optimizing for one snapshot in time.

What to Watch Next

Expect Gemini and ChatGPT to tighten grounding as enterprise clients push for reproducibility. Watch for Google to expose citation stability metrics in Search Console, likely within the next 12 months. Until then, assume your AI visibility is more volatile than your rank tracker suggests.

Related Questions

How many AI platforms should B2B marketers monitor?

At minimum, track Gemini, ChatGPT, Perplexity, and Copilot. With only 8% domain overlap between Gemini and ChatGPT, monitoring one engine gives you a false read on category presence.

Do business websites still matter in AI search?

Yes. Gemini pointed nearly 60% of citations to business websites, more than directories, reviews, and forums combined. Owned content remains the highest-leverage asset, though category design determines whether the content gets cited at all.

What is Grounding Drift and how do you counter it?

Grounding Drift is the variance in cited sources when the same AI query is repeated. Counter it by running query panels of 20 to 50 prompt variations per topic, tracking citation frequency across runs, and prioritizing content that appears consistently rather than once.