The Uncertainty Problem: Nobody Knows What AI Says About Your Business

Ask ChatGPT for the best plumber in your city, and it will name three businesses with confident, polished descriptions. What it read to get there — and whether your reviews made the cut — is something no business owner can currently see.

That's the uncomfortable truth at the center of AI search for local business. Customers increasingly turn to ChatGPT, Perplexity, and Gemini for recommendations, and 67% of consumers don't rigorously fact-check AI answers before choosing a local business. Yet owners have no dashboard showing what these assistants say about them or which sources fed the answer.

The honest state of knowledge looks like this:

• Google's ecosystem is documented. Google's own Places API documentation confirms Gemini-powered review summaries are generated "based solely on user reviews," so for Google Maps and Search, we know the mechanism directly.
• Third-party assistants rely on retrieval. ChatGPT, Perplexity, and Gemini use retrieval-augmented generation — running live searches and fetching current pages before answering. Reviews are almost certainly in that mix, but no independent technical study has documented Google Reviews specifically being ingested.
• The weightings are proprietary. No one outside OpenAI, Google, and Perplexity knows precisely how reviews are weighted, and those systems change every few weeks.

That last point deserves emphasis: anyone quoting an exact percentage — "reviews are 23% of your AI visibility" — is guessing. Directional claims are well-supported; precise figures are not. Even the quantitative research that exists has limits. A study of 22,729 AI Overview queries found review platforms cited in 34.5% of answers, but it focused on commercial and software queries, not Google's native review system or simple local searches.

There's also genuine disagreement about how often AI even answers local questions. AI Overviews appeared for 68% of local searches in one Q2 2025 study, but only 15% of simple local-intent queries — the picture changes dramatically depending on how the question is phrased.

This uncertainty is exactly why monitoring matters more than theorizing. Rather than debating unknown weightings, you can simply check what ChatGPT, Perplexity, and Gemini actually say about your business today — and repeat that check as the models change. That's the logic behind AI-search visibility tracking, and it's why Briefsy treats AI visibility as a monitored signal rather than a one-time audit, alongside your reviews, Google Business Profile, and competitors.

The bottom line: the answers exist whether or not you can see them. The only defensible position is to observe what AI actually says, respond to what you find, and stay skeptical of anyone selling you a formula.

What the Evidence Actually Shows: How AI Assistants Use Reviews

The strongest evidence comes from Google itself. The company's Places API documentation states plainly that Gemini-powered review summaries are generated "based solely on user reviews," extracting structured insights like menu items, pricing, service quality, and even negative sentiment such as "some reviews mention the rooms can be noisy." Since Q4 2024, these AI summaries and "ask about this place" prompts have rolled out across Google Maps, per industry reporting.

For third-party assistants like ChatGPT and Perplexity, the mechanism is retrieval rather than direct reading: they run live searches and fetch current pages before answering, and reviews are among the richest, most decision-relevant text published about any business, as practitioners explain.

The numbers back this up. An SE Ranking study of 22,729 AI Overview queries found that 34.5% of AI Overviews cite at least one review platform — and for queries explicitly mentioning "review" or "rating," that figure climbs to 49%, versus just 17.1% for "best/top" queries.

Reviews influence AI answers at three distinct stages:

• Retrieval — whether the AI finds your business at all
• Selection — which businesses get named in the answer
• Summarization — how your business is described once chosen

Crucially, AI reads review text, not star counts. "An assistant doesn't count stars. It reads what people wrote," as one analysis puts it — a profile with 200 vague "great job!" reviews can be less useful than 40 reviews naming specific services, places, and outcomes.

One important caveat: the data on how often AI Overviews appear for local queries conflicts sharply. Whitespark's Q2 2025 research found AI Overviews on just 15% of simple local-intent queries, but 92% of informational-intent and 97% of hybrid-intent local queries. One practitioner source reports single-digit percentages for clearly local searches. The headline figures can be reconciled by query type, but the discrepancy is a reminder that no one outside OpenAI, Google, and Perplexity knows the precise weighting formulas — and those systems change every few weeks.

For businesses trying to make sense of this shifting landscape, tools like Briefsy that monitor review text and AI-search visibility across ChatGPT, Perplexity, and Gemini offer a practical way to watch what these systems actually say about you — rather than guessing at formulas.

What This Means for Your Reviews: Text Quality, Recency, and Responses

The evidence translates into a surprisingly simple playbook: AI reads review text, not star counts. As one practitioner analysis puts it, "An assistant doesn't count stars. It reads what people wrote, checks whether the story is consistent across the web, and decides if you're a safe name to say." That reframes everything about how you collect and manage reviews.

Specificity beats volume. A profile with 200 vague "great job!" reviews can be less useful to an AI than 40 reviews naming specific services, places, and outcomes. Detailed reviews provide the keyword and outcome language that helps AI match your business to real queries, while generic praise tells a model almost nothing. When requesting reviews, prompt customers to mention what they had done and how it turned out — without scripting them.

Unanswered negative reviews are the biggest avoidable risk. When a complaint sits without a reply, the model has "only one side of the story" — and it may quote the complaint directly into an answer a prospective customer reads first. A thoughtful public reply, by contrast, is itself retrievable AI content that adds context and demonstrates how you operate. Silence next to a bad review is simply "the weaker look."

Recency matters too. Roughly 3 in 4 consumers weight reviews from the last few months more heavily than older ones, and a stale review total can signal a business that has coasted or closed. Consistency across the web compounds this: AI assembles your reputation from multiple sources, so conflicting business information "confuses AI systems and reduces trust signals," per local AI-search guidance.

Skip the GEO hacks, though. Google's official AI optimization guide explicitly debunks the idea that LLMS.txt files, special AI markup, or content "chunking" are needed for generative AI search. The fundamentals count twice:

• Collect specific, first-hand reviews that name services and outcomes
• Respond to every review — negative ones especially
• Keep a steady stream of fresh reviews flowing
• Keep business information consistent across every platform

The practical challenge is watching all of this at once — your reviews, competitors, Google Business Profile, and how ChatGPT, Perplexity, and Gemini actually describe you. That's exactly the monitoring-and-briefing workflow Briefsy automates, so you can fix the one thing that changed this week instead of guessing.

Your Monitoring Playbook: Track What AI Sees, Fix What's Broken

Everything above this point is theory. The formulas governing how AI weighs reviews are proprietary — no one outside OpenAI, Google, and Perplexity knows the precise weighting, and those systems change every few weeks. Anyone quoting an exact percentage is guessing. That's precisely why monitoring beats assuming: you can't optimize what you can't see.

Start with a review-text audit. AI doesn't count stars — it reads what people wrote and checks whether the story is consistent across the web. A profile with 200 vague "great job!" reviews can be less useful to a model than 40 reviews naming specific services, locations, and outcomes. When you request reviews, prompt customers (don't script them) to describe what they bought, where, and how it turned out.

Then close the most avoidable gap: unanswered negative reviews. When a complaint sits with no reply, the model has only one side of the story and may quote it directly into answers customers read first. Owner responses are themselves retrievable content, so a thoughtful public reply becomes part of the reputation signal. Silence next to a bad review is the weaker look.

Keep the pipeline moving. Roughly 3 in 4 consumers weight reviews from the last few months more heavily than older ones, and a stale review profile can signal a business that "may have coasted or closed." Finally, check your NAP (name, address, phone) consistency across platforms — conflicting data confuses AI systems and weakens trust signals.

Your core checklist:

• Audit review text for specificity: service, place, outcome — not just star volume
• Respond to every negative review, publicly and thoughtfully
• Maintain a steady flow of fresh reviews rather than a large stale total
• Verify NAP consistency across Google, Yelp, Facebook, and your website
• Test your actual visibility in ChatGPT, Perplexity, and Gemini rather than assuming it

That last item is the one most businesses skip. Since the systems change every few weeks, a one-time check tells you almost nothing. This is where an AI market analyst earns its keep: Briefsy watches your Google Business Profile, reviews, competitors, and your visibility in ChatGPT, Perplexity, and Gemini as one connected picture, then delivers a prioritized brief of what changed and what to do next. It's the practical version of what practitioners recommend — monitor all platforms as one corpus, not as disconnected silos you check when you remember.

The encouraging part: none of this requires GEO hacks. Google officially states that LLMS.txt files, special AI markup, and inauthentic "mentions" aren't needed for its generative AI features. The same review habits that win traditional local search also win AI search — the payoff counts twice.