The AI answered before your profile ever loaded.

The best available measurement comes from Whitespark, which ran 540 manual local queries across six industries in a study published May 2025. AI Overviews appeared on 68% of the local searches tested. Traditional local packs appeared on 39%. The AI answer is not an edge case anymore. For many questions it is the default response.

The pattern inside that number matters more than the number. Whitespark found AI Overviews concentrate heavily on research-style questions, the how-do-I-choose and what-does-it-cost queries, while simple near-me searches still mostly return the classic map pack. In plain terms: the AI intercepts your customer during the research phase, before they ever type plumber near me. By the time they search your category directly, many have already read an AI's synthesis of who is good.

Google adjusts the dial constantly. Semrush, tracking more than 10 million keywords through 2025, watched AI Overview presence go from 6.49% of queries in January to a peak of 24.61% in July, then settle to 15.69% by November. Anyone quoting a single permanent percentage is selling you a snapshot. The direction, though, is not in dispute.

You do not need to speculate about whether AI answers read reviews, because Google described its own inputs when it launched Ask Maps on March 12, 2026. The announcement says Maps analyzes information from over 300 million places, including reviews from our community of more than 500 million contributors, and pitches the results as insider tips from real people. That is Google, on the record, saying the review text feeds the answer.

Practitioner testing agrees. Whitespark's guide to AI Mode puts it directly: reviews have emerged as one of the key sources of information Google AI Mode returns for local prompts, surfaced as review-derived summaries alongside ratings and citations. And Glenn Gabe's week of hands-on Ask Maps testing found business listings carrying a know-before-you-go element explicitly sourced from user reviews, though notably he documented no verbatim review quoting inside the conversational answers themselves. Synthesis, not quotation, is the current behavior.

Now the honesty check. A claim circulates in agency content that it's the language of reviews, not the count of stars, that shapes how the AI describes a business. Directionally that matches everything above, and it matches how these systems work. But as a specific demonstrated finding it remains unproven. We are telling you which of our own claims is inference. Notice how rarely the pages selling AI optimization do that.

The old local pack showed three businesses. The AI-generated version often shows one or two. Sterling Sky's State of Local SEO analysis, published June 2026, counted the difference across markets: where traditional three-packs surfaced 18,330 unique businesses across the tracked query set, AI local packs surfaced 5,943. Roughly a third as many businesses get any visibility at all when the AI format takes over a query.

Attention inside the answer is thinner still. The first real user-behavior study of AI Overviews, by Kevin Indig and Eric van Buskirk in 2025, recorded 70 searchers and found 86% skim, with a median scroll depth of 30%, and only a small minority ever clicking a citation. If the AI's two-sentence synthesis of your business is all most searchers read, that synthesis is your first impression, assembled from words you did not write.

Which is exactly the point. The AI has to build its description of you from something. The something is largely what your customers wrote. Thirty reviews reading great service, five stars give it nothing specific to say, and businesses with nothing specific to say do not fill the one or two seats. A profile full of reviews that name the service, the problem, and how it went gives the synthesis actual material. Same business, different raw text, different answer.

The uncomfortable part of AI summaries is that they can be confidently wrong about you. Business Insider documented the failure mode in August 2026: a UK plastics supplier whose feedback the AI summarized as overwhelmingly negative using reviews that belonged to rival companies and to sellers of literal garden sheds, an Idaho property manager merged with a similarly named firm that closed years earlier, and a consultancy labeled a scam on the strength of one old forum thread about an unrelated app. Corrections took weeks. Owners are raising the same problem in Google's own support community.

Whitespark's testing observed the same class of error, noting AI Mode frequently misattributes reviews or pulls from outdated sources. There is no owner dashboard for this, and Glenn Gabe's testing found Ask Maps impressions folded invisibly into regular profile impressions, so you cannot even measure how often the AI is describing you.

You cannot file paperwork to prevent a hallucination. What you can control is the strength of the true signal. A deep, current, specific body of reviews on your own profile is the raw material that describes you correctly, and the businesses that got misdescribed worst in the documented cases were the ones with thin or stale profiles the AI padded out with someone else's story. Think of every detailed review as a correction filed in advance.

The moment AI Overviews arrived, an industry appeared to sell visibility inside them. So it is worth quoting what Google's own documentation says: there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary. No secret schema. No AI text file. Anyone selling a proprietary AI-ranking methodology is contradicting the primary source.

What the documentation does describe is the mechanism: AI features may use a query fan-out technique, issuing multiple related searches across subtopics to build a response. Which means the inputs are the same boring things that always mattered. Content that answers questions plainly. A complete, active profile. And reviews with enough substance to be worth synthesizing.

So the honest to-do list is short. Keep your profile current. Answer real questions on your site in plain language. And fix the one input almost nobody manages deliberately: the specificity of what your customers write. That last one is the only input where the review software you choose changes the outcome.

Put the pieces together. AI answers now front-run a large share of local research queries. They are built partly from review text, by Google's own description. They show fewer businesses than the old results, skimmed by readers who rarely click through. And the one lever an owner truly controls is whether their customers' reviews contain anything worth synthesizing.

That reframes the review ask entirely. The goal was never a number next to a star. It is a body of text that describes what you actually do, in enough detail that a machine assembling an answer about your trade in your town has your customers' sentences to build from. The plumber whose reviews mention the midnight call, the honest quote, and the fixed leak is legible to the synthesis. The one with forty variations of great job is invisible to it.

This is precisely the problem small Talk was built for, before AI made it urgent. Customers want to help but freeze at the blank box, so it asks them a few short questions about the job and drafts the review from their own answers, which they edit and post themselves. The result is honest, specific, and machine-legible for the same reason it is human-legible: it actually says something. Your reviews were always your reputation. Now they are also your training data.