---
title: "How AI Platforms Decide Which Business to Recommend"
seoTitle: "How AI Platforms Decide Which Business to Recommend"
description: "A plain-English look at how AI platforms choose which company to name when asked, and the concrete signals that decide whether they ever recommend yours."
datePublished: "2026-05-31T08:00:00Z"
dateModified: "2026-05-31T08:00:00Z"
category: ai
imageAlt: "Iron Goo blog featured image on the signals an AI assistant weighs before naming a small business in its answer."
tags: [chatgpt, ai-search, aeo, smb-ai, citations]
faq: true
---

Type "who's a good commercial plumber near me" into a chat window and watch what happens. The assistant pauses, then names a four-person shop you have never heard of. It does not name the regional company with the billboard on the highway and forty trucks. The owner of that bigger company would swear the small one is invisible. The small one shows up anyway. When AI recommends a business like that, the decision was not random and it was not loyalty. Something the assistant found, in the places it looked, made the small shop easy to name and the big one easy to skip. This post is about what that something is.

I have watched this happen for dozens of small businesses. I type the buyer's real question, read which name the assistant gives, then go and look, source by source, at why that one and not the other. The pattern repeats often enough that I trust it. The named business is almost never the biggest or the oldest. It is the one the assistant could describe the same way no matter where it read about the company.

## How the assistant actually makes the pick

When you ask an assistant to recommend a business, it does not reach into a ranked list of "best" companies. It does something closer to a fast literature review. It pulls a handful of sources that look relevant to your question, reads what they say, and looks for agreement. A business that shows up in several of those sources, described consistently, becomes a safe thing to name. A business that appears once, or appears three times described three different ways, becomes a risky thing to name, so the assistant talks around it instead.

Three plain words carry the whole mechanism. **Retrieval**: the assistant goes and fetches sources relevant to the question. **Corroboration**: it checks whether those sources back each other up. **Consistency**: it favors the business that every source agrees about. You do not need to know the model's internal weights to use this, any more than you need to read the wiring diagram of a slot machine to notice which lever changes the payout. You need to know the observable rule. The observable rule is that the assistant names what it can confirm.

:::callout{type="key" title="The one idea to keep"}
An assistant does not recommend the best business. It recommends the business it can describe confidently because many sources said the same thing about it. Confidence comes from agreement, and agreement comes from your facts matching everywhere they appear.
:::

This is the part that surprises owners. A bigger company can be genuinely better at the actual work and still lose the recommendation, because its online story is a mess. Three locations with three slightly different names. A services list on the site that does not match the services list on the directory. A founding story that says one thing in a press piece and another on the about page. Every contradiction is a small reason for the assistant to hesitate. The four-person shop wins not because it is better, but because it is legible.

## Why do AI platforms recommend one business and not another?

They recommend the business their sources agree about. An AI platform retrieves several places that mention each candidate, then favors the one described the same way across all of them. A single clear identity, repeated and corroborated, beats a bigger name that the sources contradict.

That is the forty-word version. The rest of this post is the longer one: what "described the same way" means in practice, and the handful of things an owner can actually go fix.

## What makes a business nameable

There is a real difference between a business an assistant will name and one it will summarize over without naming. The nameable business has three attributes, and none of them is about size.

The first attribute is **presence in more than one place the assistant reads**. One website, however good, is one source. The assistant wants corroboration, and corroboration needs at least two independent things saying the same thing. A business that exists only on its own homepage gives the assistant nothing to cross-check, so the assistant stays vague. A business that appears on its own site, in two or three relevant directories, in a trade listing, and in a customer's public mention has given the assistant a small chorus instead of a solo.

The second attribute is **identical core facts everywhere it appears**. Name, what the business does, where it operates. If the company is "Riverside Plumbing Co." on its website, "Riverside Plumbing" on one directory, and "Riverside Plumbers LLC" on another, a human reads all three as the same business without thinking. The assistant is more literal. Each variant is a small fork in the road, and forks erode the confidence it needs to put a name in an answer. The fix is boring and it works: pick one exact way to state each fact, then make every place that mentions you say it that way.

The third attribute is **a single resolvable identity**. The assistant needs to be sure that all those mentions point at one company, not at several that happen to share a word. This is the same problem search engines have been working on for years, and it is the heart of what people now call [building a clear entity the machine can pin down](/blog/entity-seo): one company, one clear category, one set of facts, no ambiguity about who is who. When the identity resolves cleanly, every mention reinforces every other mention. When it does not, the mentions cancel out.

::::comparison{title="Two businesses, same trade"}
:::side{label="The nameable one"}
Same name in five places. Same one-line description of what it does. One service area, stated identically. A few public mentions that all agree. The assistant reads four sources, finds no contradiction, and names it without hesitating.
:::
:::side{label="The skippable one"}
Bigger, busier, better-reviewed, but the name varies, the services list does not match between site and directory, and two pages tell different founding stories. The assistant finds the contradictions, loses confidence, and answers with a generic "look for a licensed local plumber" instead.
:::
::::

Notice what is not on either list. Nowhere did the bigger company lose because it was bad. It lost because it was hard to confirm. That is the whole game, and it is good news for a small business, because being easy to confirm is something you can fix in an afternoon and a few follow-ups. Being the biggest is not.

## How many sources does an assistant actually read before answering?

You do not get to see this number, and it changes per question. The honest framing is a small handful, not hundreds. Treat the grid below as illustrative shape, not a measured statistic.

:::stat-grid
::stat{value="A few" label="sources typically pulled for one local recommendation"}
::stat{value="2+" label="independent mentions needed before a name feels safe"}
::stat{value="1" label="clear identity all of them must resolve to"}
:::

The practical takeaway from those rough numbers is that you are not fighting for the top of a list of a thousand. You are trying to be present, and consistent, across the small set of places the assistant is likely to check for your kind of business in your area. That is a much smaller and more winnable job than it sounds.

## What this is not

It is worth killing two ideas before they cost you time.

It is not done by talking to the bot nicely. You cannot ask an AI platform to remember your business, and prompting it politely changes nothing about what it finds when it looks for a plumber next week for a different user. The recommendation is decided by what the assistant retrieves, not by how you, the owner, phrase a request. The chat window has no memory of your company between strangers' conversations. If you want the full picture of what the chat window is and is not for a small business, the prior on [what ChatGPT actually is and where it fits](/blog/chatgpt) covers that ground; this post assumes you already know the window and want to know how the pick gets made.

It is also not "just SEO" with a new label, though the overlap is real. Presence and consistency have always mattered for search. What is genuinely different here is the corroboration step. A search engine can rank a single strong page on its own merits. An assistant prefers not to name a business until several sources agree, because naming a wrong or contradicted business is a worse failure for it than staying vague. So the work tilts harder toward "make every source tell the same story" than classic ranking ever did.

:::quote{cite="A small-business owner, after the fix"}
We did not get bigger. We got easier to describe. The assistant started naming us once the same three facts showed up the same way everywhere.
:::

## The owner's first moves

Here is the short, doable list. Not thirty items. The handful that move the result.

- **Be present where buyers and assistants both look.** Your own site, plus the obvious directories and listings for your trade and area. You do not need to be everywhere; you need to be in the few places relevant to your kind of business that an assistant is likely to read.
- **Make the core facts identical everywhere.** One exact business name, one one-line description of what you do, one stated service area. Copy them from a single document into every listing so they cannot drift.
- **Answer the real questions buyers ask, in public.** The questions people actually type before hiring you, answered plainly on a page anyone can read, give the assistant clean material to quote and corroborate.
- **Clean up the contradictions.** Track down the stale listing with the old phone number, the directory with the wrong service area, the page with the abandoned business name. Every contradiction you remove is a reason to hesitate that you take away from the assistant.

None of this guarantees the assistant will name you. Anyone who promises that is selling the magic-bullet version you have rightly learned to distrust. Visibility inside these answers is earned across sources, not switched on with a trick. What the list does is move you from "impossible to confirm" toward "easy to confirm", which is the only lever you actually hold.

Those four moves are the foundation, not the whole building. The mechanism is one thing; the full program of being present and corroborated across every source an assistant reads, the query-by-query work of it, is a deeper subject. When you are ready for [being cited across the sources assistants read](/guides/seo/seo-for-ai-search-and-aeo), that guide lays out the strategy this post only opens. Start with the four moves this week, then go read it.