The short answer
When you ask an AI assistant to recommend a business, it tends to name the one that the most independent sources vouch for, for that exact job. Size, ad spend and a famous name do not decide it. Evidence from other people does.
That sentence comes from a small test, and we want to be straightforward about both parts. The pattern was clear in what we saw. The sample was small, so we treat it as an observation to keep testing, not a national law. The rest of this page shows what we looked at, what we saw, where the limits are, and what a business owner can do about it.
If the idea is new to you, here is why it matters. A growing number of people ask an assistant "who should I call" or "where should I eat" and take the shortlist. The assistant does not browse like a person. It gathers what it can read about businesses, compares, and writes a few names. If you understand what it reads and what it values, you can improve your position without guessing.
Getting help with that work is a reasonable choice for a busy owner. Our overview of what paid AI search help delivers explains what a service does and what no one can promise.
What we tested
In early October 2026, Matt Goren asked AI assistants for local recommendations in Richmond, Virginia. He used Google AI Mode, ChatGPT logged out, Claude, and Gemini for one category. He asked about five kinds of business: coffee, pizza, AC repair, plumbers and oil change. For each, he recorded who was named and what sources the assistants appeared to lean on.
He then re-ran part of it the same day in DeLand, Florida, a much smaller town, using Google AI Mode and ChatGPT. The idea was to see whether the pattern in a mid-sized city held in a small one.
This is not a laboratory study. There was no control group and no statistical sample. It is a careful look at what real assistants returned to real questions on a given day. That is useful, because it is exactly what a customer would see. It is limited, because one day in two towns cannot speak for everywhere. We list the limits in full later on this page.
We are deliberately not naming any company in this write-up, large or small. The point is the pattern, not any one business, and naming names would invite readers to treat the test as a ranking. It was never meant to be one.
What we saw
The results fell into a few groups.
Coffee, pizza and AC repair
No national chain appeared in any answer. For coffee alone, the assistants gave 27 recommendations across 5 answers, and no national coffee chain was among them.
Plumbers
The answers were led by local companies. The national franchise brands appeared only in Google's Sponsored ads shown under the answer.
Oil change
This was the exception. National chains did lead here. But the assistants chose them location by location, based on reviews, and always next to independent shops. One assistant told the user to skip one specific chain location over one bad review.
Different assistants, different winners
Each assistant picked a different number one coffee shop in Richmond. It is not one ranking anymore. A business can be first in one assistant and absent in another.
The small town
In DeLand, the pattern held and looked stronger. A plumber with 19 reviews was in the answer, while companies with 4,600 and 10,400 reviews, including national franchise brands, sat in the Sponsored ads. In the small town the assistants agreed on the number one coffee shop. In Richmond they did not.
The line we keep coming back to: The big chains are in the ads. The local businesses are in the answer.
The rule it suggests
Looking across those results, one thing explains most of them. Every source the assistants leaned on was third party. They pointed to review sites, Reddit threads, local blogs, niche guides, local polls and press. One assistant described company self-descriptions as "leads, not endorsements."
That suggests a working rule. AI recommends the specific business that the most independent sources vouch for, for that exact job. Size, ad spend and a famous name get you the ad slot next to the answer, not the answer.
Each part of the rule does work.
"Independent sources." A source you control, such as your own site, counts as a claim. A source you do not control counts as evidence. An assistant weighs a claim that is echoed elsewhere more than a claim that stands alone.
"The most." Agreement across several sources beats a single strong one. A business mentioned in a local guide, a forum thread and a set of detailed reviews looks more certain than one with a single glowing profile.
"For that exact job." Assistants answer a specific question. A plumber praised for emergency leaks is a better match for "burst pipe at night" than a plumber praised for remodels. Being known for the exact job beats being broadly known.
The diagram below shows where those ideas sit in the process.
What counts as someone vouching for you
Not every mention is equal, and it helps to know the kinds. These are the source types the assistants leaned on in our test.
Review sites. Detailed reviews that mention a specific job carry more than a star rating alone. The wording gives the assistant something to match against the question. We go deeper on this in reviews and AI recommendations.
Forum threads. Reddit and similar communities, where real people say "I used these folks, they were great." These read as honest because they are unprompted. See Reddit and AI search for how to handle them without causing harm.
Local blogs and niche guides. A writer who covers a town or a hobby and lists their favorites. These often shape local answers more than a business expects.
Local polls and press. "Best of" polls, newspaper features and similar. A claim that a business won an award is only worth citing if it is real, and assistants can repeat unverified claims, so check any award you see quoted.
Notice what is missing from the list: your own marketing copy. It is still useful, because it tells assistants who you are and what you do. It is not what convinces them to recommend you. Our page on what AI engines cite goes through source types in more depth.
The limits, stated plainly
A pattern from a small sample can mislead if it is oversold. These are the limits of this test.
- It covers two towns, one day each, and a handful of categories.
- Some of the assistants ran on a logged-in account, which can change what they say.
- It is an observation to keep testing, not a national law.
A few more points are worth saying out loud. Answers vary from one asking to the next, so another day could produce different names. Categories differ, as the oil change result showed, where chains did lead on review evidence. Award claims quoted by assistants were not verified by us.
None of this makes the observation useless. It makes it a hypothesis you can test cheaply in your own market. If you see the same split between ads and answers in your category, you have confirmation that matters to you, regardless of what happens in the rest of the country.
We plan to keep re-running it in more places. Until we have, the careful way to say it is: in what we saw, in these towns, on these days, independent evidence beat size.
What to do with this
If the rule holds for you, the work is clear, and it is within reach of a small business.
Run the same test on your own market. Write the five questions a customer would ask. Ask them in two or three assistants, in a clean session. Note who is named. Do it again in a month.
Pick one exact job and earn evidence for it. Ask happy customers to mention the specific job in their reviews. Each detailed review is a vote for that job.
Get mentioned where assistants look. Local guides, niche directories, a short local article, honest participation in the forum where your customers ask. The goal is several independent sources that say the same true thing.
Keep your own pages clear. State what you do, where, and for whom, in the first sentence of each section. Your site is the claim that other sources confirm.
Do not try to buy your way into the answer. The ad slot is a separate channel, and we describe it in ads in AI overviews. For the local angle, see why AI recommends local businesses.
For a quick first reading of where you stand, an AI visibility checker is a reasonable start.
How to read an answer for who is vouching
You can turn any assistant answer into a small piece of research about your own market. It takes ten minutes and tells you more than most reports.
Step one: ask a real question. Use the words a customer would use, such as "who fixes a leaking water heater near me." Do it in a clean session, so earlier chats do not shape the reply.
Step two: write down every business named. Note the order, but do not read too much into it. Different assistants order differently.
Step three: look at the sources. Some assistants show links or name where they looked. Others describe their reasoning in a sentence or two. Either way, note which kinds of source come up. Are they review sites, forum threads, local guides, news items?
Step four: ask why. You can simply type, "Why did you recommend these?" The explanation often names the evidence it leaned on. That tells you which kind of vouching matters in your category.
Step five: check yourself against the same sources. If an assistant relies on a particular local guide or forum and you are not in it, you have found a specific gap. If the named businesses all have detailed reviews that mention the job and yours say only "great," you have found another.
Repeat the whole thing next month and compare. The exercise turns a vague worry about AI into a short list of specific things to go and earn.
Three ways to misread this
A pattern that sounds simple gets bent in retelling. Here are three misreadings worth avoiding.
"Reviews are all that matter." Reviews were one of several source types. Forum threads, local blogs, niche guides and press also came up. A business with great reviews and no other mentions has one kind of vouching. The more kinds you have, the more certain you look.
"Ads do nothing." Ads are a separate channel and can pay for themselves. The finding is only that ad spend did not put businesses into the answer. If a campaign brings you customers at a profit, keep it.
"This is a trick." There is no trick here, which is the point. Faking independent sources is hard and tends to backfire, since assistants and readers both notice. The reliable route is slow and honest: do the job well, make it easy for happy customers to say so, and be present in the places where people talk about your field.
If you read the rule as "become the business that real people vouch for, in several places, for one specific job," you have read it the way it was meant.
How RunOctopus handles this
The rule says independent evidence decides. Part of that evidence is your own clear, specific pages, and that part is work RunOctopus does for you.
RunOctopus is a done-for-you service. Its AI, Ollie, researches a business, then writes and publishes pages directly on the business's own website. A team plans, writes, reviews and publishes alongside Ollie. It works with Shopify, WordPress, WooCommerce and Wix. There is nothing to install, nothing to approve, nothing to upload and nothing to schedule, and pages go live on their own.
The first week brings 8 long-form articles, 6 pages and a custom interactive tool built for the business's niche. After that it is 175+ pieces of content a month, written, published and linked. RunOctopus also reports on where ChatGPT, Perplexity, Claude and Gemini name the business. The 90-day promise is more visibility on Google and in AI answers within 90 days, measured against the 90 days before joining, or every dollar back.
For scale, here is what the same approach has done in Google Search Console (28 days to Sep 29, 2026). Sterman, a Los Angeles real estate business, went from 44 to 210 Google clicks a month since April.
You can see the plan before deciding anything. Paste your address into the free site read and Ollie reads the site and shows the plan. No account, no call.