My interpretation of a summer spent measuring AI recommendations is that the businesses which keep getting named are simply the ones whose goodness left the most evidence.

This post is different from the six before it, and I want to be straight about how. Everything else in this series reported what our August 2026 study measured: asking "who is the best {service} in {city}" across 216 US markets and recording the ChatGPT web app's own responses, alongside city-targeted Google SERP data. This post is my interpretation of what those measurements add up to. Interpretation, not a measured causal factor. The study says what happened. I am going to say what I think it means for how you get recommended by AI.

The numbers behind every claim here live on the study page at how ChatGPT picks a local business, and the measured findings are in the earlier posts, starting with the churn finding.

The survivors carried more reviews at identical ratings

Here is the single table I keep coming back to. Of the 974 businesses ChatGPT recommended in the study's first wave, 469 survived to the second wave and 505 dropped out. Compare the two groups:

• Median reviews: 226 for survivors versus 164 for dropped
• Median rating: 4.9 for both, identical
• Held the number one card in wave one: 30.7% of survivors versus 13.9% of dropped
• Median local search rank: 6 versus 8
• In-city address: 93.4% versus 89.9%

Both groups were excellent on paper. Everyone in this data is playing at 4.9. What separated the businesses that stayed in the answer from the ones that fell out of it was not being better rated. It was being more evidenced: more reviews, more prominence, more depth behind the same star count.

Review volume records how a business treats customers

Now the interpretive leap, labeled as such. I think review volume at a near-perfect rating is the measurable residue of caring about customers. You cannot accumulate 226 reviews at a 4.9 by gaming anything for long. That number is what a business looks like after years of doing good work, asking for feedback, and fixing what goes wrong, deposited one customer at a time into the public record.

AI systems cannot observe your workmanship. They can only read the residue. So when the study finds that the deep-review businesses kept getting picked while equally-rated thin-review businesses churned out, I read that as AI answers rewarding, imperfectly and probabilistically, the accumulated evidence of a business being genuinely good. The optimization and the operation stop being separate projects.

You cannot hold a rank in AI search

The churn data reframed for me what winning even means on these surfaces. Half the recommended businesses gone within a week (469 of 974 survived, 48.2%). The reading list behind each answer overlapping at just 0.20 between asks. A number one spot that repeated in only about a third of markets.

The line I have landed on, and the one I would put on the wall: you can't hold a rank in AI search. You can only be worth recommending every time it looks, and be everywhere it looks.

Both halves are load-bearing. Worth recommending is the review depth, the 4.9, the in-city presence, the profile completeness that the shown-business profile documented. Everywhere it looks is the coverage story from the retrieval churn post: businesses ChatGPT re-encountered were shown again 72.5% of the time (469 of 647), so the fight is being present on whichever sources get read on any given ask.

The work is ordinary even though the surface is new

If my interpretation is right, the to-do list for getting recommended by AI is almost anticlimactic, and I mean that as encouragement:

• Run a business worth talking about, and ask every happy customer for a review, forever. Depth is the separator at the top; the thin end of the shown panels started around 50 to 95 reviews depending on market size, and survivors ran far deeper.
• Keep both reputation profiles honest and current, since the answers pull from Google and Yelp both.
• Cover your category's directory rotation so you are in the pool no matter which pages get read.
• Say plainly on your own site who you are, what you do, and where you are.

That is the same program I laid out in how to get recommended by ChatGPT before this study existed. What the study added is evidence about which parts carry the weight, and a reason to stop chasing tricks: the one manufactured shortcut we could test, stacking extra GBP category tags, showed no measurable lift.

Consistency compounds where snapshots expire

The last thing the interpretation changes is patience. On a churning surface, no single appearance is the prize and no single absence is a verdict. The prize is an appearance rate that climbs as your evidence deepens, week over week, everywhere AI reads. That favors owners who compound: the review earned today is still testifying for you in every answer built next year.

The businesses that win AI search, I think, will be the ones that were winning customers all along, and simply made sure the record showed it.

The questions owners ask most about getting recommended by AI, answered plainly with the study data underneath.

Ready to build the evidence AI keeps reading?

Every piece of this is buildable: the review engine, the dual-profile hygiene, the directory coverage, the plain-spoken website. Spearleaf builds that whole evidence layer for service businesses as one program, with honest measurement of your appearance rate along the way. If you want a market-specific plan for becoming the business AI keeps finding reasons to recommend, reach out through our contact page and we will build it with you.

This article was written by Joshua Albanese, founder of Spearleaf, a local SEO and marketing agency based in Fort Myers, Florida. Joshua has built six businesses from zero using organic search and content, and he leads Spearleaf's SEO, AI search, and local strategy for service businesses across the U.S.