The AI visibility benchmark: how 10,000 local businesses actually perform in ChatGPT and Perplexity

Most conversations about AI visibility right now are built on opinion. "AI hallucinates a lot." "Reviews matter more than ever." "Technical SEO is dead for AI." All plausible, none of them backed by much more than one agency's anecdote.

We wanted an actual number to point to, so for our AI Visibility Report, we pulled 10,000 US local businesses from the Insites 360-degree audit platform, re-ran them through a clean, standardised audit, and asked ChatGPT and Perplexity about every single one. This is the benchmark that came out of it.

The headline numbers

Four stats do most of the work in this report:

• 93.59% of businesses were known to ChatGPT (93.15% for Perplexity), so outright invisibility is rare
• Only 55.58% of ChatGPT results (48.05% for Perplexity) fully matched the business's actual Google Business Profile details, so being "known" and being described correctly are not the same thing
• 40.88% of businesses were never recommended by ChatGPT in unbranded, intent-driven queries (68.79% for Perplexity), so awareness doesn't guarantee inclusion in the answer
• 133.4 vs. 10.7 average Google reviews for businesses surfaced by both platforms versus neither, the single strongest correlating signal in the entire dataset

Those four numbers map directly onto the three layers we use to measure AI visibility: awareness, reputation and recommendation. Here's what the data shows at each layer.

Layer 1: Awareness, does AI know the business exists?

Awareness asks the most basic question: if you ask ChatGPT or Perplexity directly about a business, does it know who you're talking about, and does it get the details right?

The topline is genuinely reassuring. AI has broad coverage of the local business market, including small, independent businesses with limited digital marketing activity. Where it falls down is accuracy. Checked against Google Business Profile, only around half of results fully matched. The most common error was an incorrect phone number (30.09% for ChatGPT, 38.81% for Perplexity), followed by a wrong website (15.59% / 16.9%) and, in the more concerning cases, a business name that didn't match at all (6.38% / 5.18%). In 1.68% of ChatGPT searches, none of the details matched the reference business, a likely hallucination pointing to a rival or unrelated company entirely.

What correlates with being found accurately: Google Business Profile strength (reviews, completeness, photos, claimed status), local pack presence, directory coverage and consistency, and website content depth. What barely moves the needle: structured data, llms.txt, and Core Web Vitals. We go deeper on the accuracy problem specifically in why ChatGPT gets details wrong for half the businesses it already knows.

Layer 2: Reputation, what does AI think of the business?

Once a business is identified, the next question is how AI talks about it. We asked ChatGPT and Perplexity a fan-out of category-specific questions per business (is it good value, is it fast, is the work good quality) and had the models self-rate sentiment on a five-point scale.

The result is heavily compressed toward positive. ChatGPT rated businesses positively or very positively 79.8% of the time, with "very negative" appearing in just 0.5% of cases. Perplexity showed the same skew, rating 47.5% of businesses "very positive" outright. Genuinely negative AI opinions are rare, which means sentiment alone is a weak differentiator between businesses.

What does separate a positive description from a merely neutral one: website freshness (the strongest single factor), review rating, local listing consistency, and GBP completeness. Crucially, being described well by AI doesn't guarantee being recommended. That's the next, and most commercially important, layer.

Layer 3: Recommendation, does AI actually put the business forward?

This is the layer that matters most for a client's bottom line. When someone asks an open-ended, unbranded question, "recommend a plumber in Denver", does the business make the shortlist?

Here the picture gets harder. We bucketed businesses from "highly recommended" (surfaced across all five test queries, in a strong position) down to "never recommended". 40.88% of businesses fell into the never-recommended bucket for ChatGPT, and 68.79% for Perplexity. Being known to AI, even being described positively by AI, is no guarantee of being chosen when it matters.

What correlates most strongly with recommendation: review volume (304.2 average reviews for the most-recommended tier versus 117.5 for the least), local pack presence (74.6% versus 21.1%), and traditional search visibility (businesses surfaced by both platforms rank for 71.1% of target keywords, versus 13.5% for those never surfaced). What barely correlates at all: technical SEO, site performance, and most AI-specific "optimisation" tactics like structured data and llms.txt. Reviews turned out to be the single strongest lever in the whole dataset, worth a closer look on its own.

The pattern underneath all three layers

Read across all three layers and one theory holds up consistently: AI visibility is built on corroboration, not optimisation. Businesses that are known, well-regarded and recommended tend to be the ones with more evidence backing them up everywhere at once, more reviews, more directory listings, more consistent data, more search visibility, rather than any single technically perfect asset.

That has a direct implication for scale. Businesses with a bigger overall footprint find it structurally easier to accumulate that evidence, which is a real advantage for larger, more established brands. But recency and consistency (an actively updated website, ongoing review generation, accurate listings) correlate strongly too, and those are levers every business, regardless of size, can pull. We unpack that dynamic fully in the scale advantage.

What this means for agencies

Three practical shifts fall out of the data:

• Audit accuracy, not just awareness. A business AI knows about but describes incorrectly is a liability, not a win
• Track recommendation, not sentiment. Positive AI sentiment is common and doesn't predict whether a business actually gets put forward as an answer
• Report on signal accumulation, not single fixes. There's no keyword position to point to. Progress looks like more reviews, more consistent listings, and more corroborating presence over time

It's the same three-layer audit Insites runs on every business in your book, so this isn't a measurement framework agencies need to build from scratch.

This is the summary version. The full report goes deeper on every layer, including the complete correlation tables, the differences between how ChatGPT and Perplexity weigh signals, and what this means for how agencies price and deliver AI visibility services.

Read the complete findings in Insites' AI Visibility Report, based on a clean audit of 10,000 US local businesses across ChatGPT and Perplexity, or run a free AI visibility check on any business in seconds.