AI recommends the businesses whose websites it reads

What did we find?

Across 17 companies the correlation between citations to a business's own website and how often it was recommended was Spearman rho +0.92 (95% confidence interval +0.80 to +0.97). Every one of the five businesses that was never recommended in 36 checks had zero citations to its own site. Not one of them.

Google review count, by contrast, correlated at +0.25. The roofer with the most reviews in Georgetown, 705 of them, was recommended once. There is no company in this sample that the engines read and then declined to recommend, and no company they recommended without reading.

Figure 1Own-site citations against times recommended, 17 businesses

Which businesses did AI recommend?

Companies were drawn from Google Places by trade and city and ranked by review count, not taken from the businesses AI already favours.

How often do AI engines cite a business's own website?

Of all 1,306 citations collected, 4.3%pointed at the business being asked about. The rest were directories, trade associations, review platforms and competitors' websites. These were the most cited sources across all three trades:

Three of the top seven are individual contractors' own websites. Those contractors are not incidental to the answers, they are the source material for them. When an engine explains how to choose a roofer in Georgetown, it is frequently explaining it in Ark Roofer's words. Every other business in that market is being described by a listing somebody else wrote.

Do Google reviews help you show up in ChatGPT?

Not in this sample. Review count correlated with visibility at +0.25, and in two of the three trades the most-reviewed business was among the least visible. Grand Endeavor Homes has 141 reviews and was recommended 17 times. Chance Leigh has 875 and was recommended 8.

This does not mean reviews do not matter. Reviews are one of the strongest signals in whether a person chooses you once they have your name, and review content is part of what these engines read when a business is already in the answer. What the data shows is narrower: reviews are not the mechanism that gets your name into the answer in the first place. Those are two different jobs, and most local businesses have only solved one.

Does being a large company help?

Not on its own. Two of the six zero scores belong to companies operating at national or multi-market scale. A D.R. Horton community and Saratoga Homes, trading since 1983 across Texas and New Mexico, were both absent from all 36 checks. Brand recognition built over decades in the physical world does not transfer. These systems read pages, and a page that was never read cannot be recalled.

Is AI visibility the same in every trade?

No, and advice that treats it as one problem is wrong. For remodelers, directories led: Houzz first, then HomeAdvisor, then the Austin chapter of the National Association of the Remodeling Industry. For plumbers, in a separate set of twelve checks in Georgetown covering 898 citations, business websites were 90.5% of citations and directories only 9.5%; Thumbtack, which several guides name as a priority, was 1.2%. For roofers, the Better Business Bureau was the largest non-competitor source.

A remodeler who spends a month getting onto Houzz is doing the right thing. A plumber doing the same is mostly wasting the month.

How did we measure this?

Twelve questions per business, across three engines, giving 36 checks per company. Every answer was classified as recommended, mentioned but not recommended, used as a source but not named, or absent. We recorded every URL cited in every answer, then counted how many pointed at the business being asked about.

This is the complete question set, shown for a roofing contractor in Georgetown. The category and city are substituted per business; everything else is identical:

• best roofing contractor in Georgetown, TX
• affordable roofing contractor in Georgetown, TX
• most reliable roofing contractor in Georgetown, TX
• recommend a good roofing contractor near Georgetown, TX
• who is the top rated roofing contractor in Georgetown, TX
• who should I hire for roofing contractor work in Georgetown, TX
• how do I choose a good roofing contractor in Georgetown, TX
• questions to ask a roofing contractor before hiring
• what should I look for when hiring a roofing contractor
• compare roofing contractor options in Georgetown, TX
• top 5 roofing contractor businesses in Georgetown, TX
• which roofing contractor companies serve Georgetown, TX

These are templates, not observed searches, and that is deliberate. We did not take them from Search Console or a keyword tool, and we make no claim about how often anyone asks them. Two reasons. The first is comparability: 17 businesses were asked the same twelve questions, and bespoke questions per business would make the scores incomparable, which would make the central finding impossible to compute.

The second is that the validation people expect here does not exist. Google Search Console reports what people typed into Google. No AI platform publishes what people type into it. There is no Search Console for ChatGPT. Anyone claiming their AI visibility prompts are volume-validated is either using Google data as a proxy for a different medium, or guessing. We are guessing too, and saying so.

Four of the twelve deliberately carry no city. A business that only appears when the question names its town is visible for a narrower set of questions than one that surfaces on “what should I look for when hiring a roofing contractor”, and we wanted that difference to show.

We supplied business name, category and city explicitly rather than inferring them from page metadata. Deriving a business name from a page title produces questions no engine can answer and a confident zero that means nothing. Checked September 2026; review counts from Google at time of checking.

What are the limitations of this study?

Seventeen businesses, one metropolitan area, three trades. Enough to show a strong relationship between own-site citation and recommendation in this sample. Not enough to quantify how much any single factor contributes, and not a claim about other markets.

Correlation is not causation. A third factor, such as domain authority, may drive both citation and recommendation. What we can say is that in this sample no business was recommended without being read, and every business never recommended was never read.

The questions are ours, not observed demand. A business scoring 23 of 36 is more visible than one scoring zero on the same instrument. Neither number tells you how many people asked.

These engines vary between runs. A business recommended in 17 of 36 is not reliably visible, it is visible to some phrasings and not others. Every figure is one point in time.

We sell software that measures and improves this, which is why we had the tooling to run it. The data is what the engines returned. Anyone can reproduce it by asking the same questions.

What people ask about this