How ChatGPT chooses which businesses to recommend.
ChatGPT and similar assistants do not keep a private list of good businesses. When a question needs current, local information, they search or retrieve web pages and listings, read what those sources say, and compose an answer that names whichever businesses it can find and verify with agreeing facts, favoring the ones that answer the question most directly. Sources that disagree with themselves, or say nothing plain and specific, tend to get left out. The AI-Ready Audit shows exactly what this process currently produces for a given business, question by question.
ColabContent LLC publishes this page: a Boston AI consulting firm that builds custom AI systems businesses own outright, starting from a $499 audit that shows what Google and AI assistants say about a business today.
Search, read, compose: the three steps behind every answer.
It helps to picture what actually happens between a customer typing "best HVAC company in Nashua" and an assistant answering with two names. First, the assistant recognizes it needs current, local, factual information it cannot reliably know from memory alone, since business details change constantly. Second, it searches or retrieves relevant web pages, business listings, review platforms, and sometimes local news or directory sites, the same open web a person could browse themselves. Third, it reads what those sources say, compares them against each other, and composes a short answer that names the businesses whose information looks most complete, most consistent, and most directly responsive to the question asked.
This is meaningfully different from how a search engine ranks links. A search results page can list ten businesses in a row and let the customer decide. An assistant answering conversationally is making a much narrower, more committed choice: it is naming one or two businesses out loud, in a sentence, as if recommending them. That narrower commitment means the assistant applies a higher bar before naming anyone: it wants sources that agree with each other on the basics, and it wants a source that actually states the answer rather than one the assistant has to infer.
None of this is a black box in the sense of being unknowable. It follows directly from what an assistant needs in order to answer confidently: current information, agreement across independent sources, and content that states the answer plainly. A business that supplies all three tends to get named. A business that supplies none of them, regardless of how good the actual work is, tends not to.
Key Terms
Retrieval: an assistant fetching and reading current web content to answer a question, rather than relying only on what it learned during training. Source agreement: whether multiple independent pages describing a business state the same facts. Citation: a specific instance of an assistant naming a business or linking to a source in its answer. Directness: whether a page states the answer to a likely customer question in plain language, versus requiring the reader to infer it.
Four things that move the decision.
Findability comes first: if the assistant's search step never surfaces a business at all for the phrasing a customer used, nothing else matters, because the business was never a candidate. A thin or generic web presence, or one with almost no content specific to the exact service and town being asked about, tends to lose here before the comparison even begins.
Consistency is second: does the business's name, address, phone number and services agree across its own website and every listing that mentions it. A business with three slightly different phone numbers across different directories reads as an unreliable source, and an assistant naming an unreliable source risks giving a customer wrong information, so it tends to prefer a source that agrees with itself.
Directness is third: does a page actually state the answer in plain language, such as naming the specific service and the specific town, or does it only describe the business in general marketing terms. An assistant can quote a direct statement; it cannot quote an inference it had to make itself.
Corroboration is fourth: do independent sources beyond the business's own website, such as review platforms, directories or local coverage, describe the business the same way. A single self-authored source is the weakest evidence a business can offer; agreement across several independent ones is the strongest.
A more selective decision, not the same list in a new format.
A business that ranks reasonably well in traditional Google search results can still be left out of an AI Overview or a ChatGPT answer for the same question, and the reverse can also happen. Google's own AI Overview is a related but separate system from the ten-blue-links search results below it, and ChatGPT, Claude and Perplexity are separate systems again, each with their own retrieval and their own judgment about which one or two sources to name in a short answer. That is why checking only a business's Google search position is not the same as knowing what customers hear when they ask an AI assistant, and why the two are worth checking separately, side by side, with the same questions asked of both.
The upside is that the underlying fixes overlap heavily. Consistent listings, plain-language service and town pages, and independent corroboration are rewarded by traditional search ranking and by AI-assistant citation alike. A business does not need two separate strategies; it needs one disciplined effort applied to both channels, checked separately so it is clear which one is actually moving.
Seeing exactly where a business currently stands on both, side by side, is what the AI-Ready Audit does: Google rankings and map results from each town served, next to the word-for-word answers of ChatGPT, Claude and Perplexity, with Google's AI Overview read from the search results in each town where Google shows one, to the real questions your customers type, built from your services and towns.
The four factors applied to one unnamed business.
Take a countertop fabricator in Massachusetts, audited in September 2026 and not named here. On findability, ChatGPT and Claude each named it on 2 of 8 questions, and only on the two that contained its own name; for "best countertop installers near" its home town both named other fabricators, because the site never states a town next to a service anywhere, and Google's map results put it in the top three in only 2 of its 12 towns. On consistency, Yelp listed the company in a different town from its own website, which is exactly the kind of disagreement that makes an assistant prefer a cleaner source. On directness, the homepage headline said "Supplying and Installing the Finest Countertops" rather than which stone, which towns and whether it repairs, and the site had no repair page although the search pull showed local repair demand. On corroboration, the 34 Google reviews average five stars, but of the 30 read for the audit, 21 praised one person by name and almost none named the town or the job, which is warm evidence for a person and weak evidence for a service in a place.
None of the four factors alone explains why an assistant names the competitor instead; together they do. A firm that fixed only the consistency problem, correcting the old office listing, would still lose on findability and directness, because the underlying pages never state the plain-language facts an assistant needs to quote. Working through all four in order, cheapest and highest-leverage first, is what actually changes which name gets said out loud. The $499 AI-Ready Audit scores a business on all four from its own pages, its listings and the assistants' recorded answers.
The first week after reading this.
Understanding the mechanism is only useful once it gets applied to one business's actual pages and listings. In the first week, walk through the four factors above against the business's own web presence: search for the business by name and by service to see what comes back for findability; pull every listing that carries the business's name, address and phone number and check them against each other for consistency; read the homepage and main service pages as a stranger would, asking whether they state the plain-language answer to a likely customer question, for directness; and read through recent reviews for corroboration, checking whether they mention anything specific rather than only general praise.
Writing down what fails on each of the four factors, rather than guessing at a single fix, is what turns this page's explanation into an actual plan. The order to work through them is usually consistency first, since it is the cheapest and touches every question at once, then directness, then findability and corroboration together as an ongoing effort.
Related pages.
This page explains the mechanism behind the decision. The links below cover the specific gaps that mechanism produces, how to check your own results, and what the paid audit adds on top of a manual check.
See why ChatGPT skips a business for the specific, common gaps that come from this mechanism.
See how to check what AI currently says about your business for the practical next step.
See the AI visibility audit for what the $499 audit checks and what it costs.
Start with the $499 audit.
The AI-Ready Audit is $499. The report arrives within 3 business days as a private link and a PDF, with a 5-minute video walkthrough and a 20-minute call. If it has no value you get the $499 back, and every quarter your AI answers, rankings and money leak are re-checked free.
FAQ
The questions below cover the mechanics in more detail: whether ChatGPT keeps its own database of businesses, what the assistants actually read when a question comes in, why ChatGPT, Claude, Perplexity and Google's AI Overview can disagree on the same question, whether paying for ads or stuffing keywords influences the answer, and what does move the outcome.
Does ChatGPT have its own database of businesses?
No. When an assistant needs current, local information it searches or retrieves web pages and listings, the same open web everyone else can read, rather than drawing from a private business directory.
Why do different assistants recommend different businesses for the same question?
Each assistant retrieves from somewhat different sources and weighs them differently, so ChatGPT, Claude, Perplexity and Google's AI Overview can each name a different business for an identical question.
Does a bigger advertising budget change who gets named?
Not directly. There is no ad product for being named inside an assistant's answer. Ad spend can indirectly help by driving more mentions and reviews, but it does not buy a citation the way it buys a search ad.
Can old, outdated information about my business hurt me?
Yes. An assistant reading a stale directory listing with an old phone number or a discontinued service can quote it as current, which both misleads the customer and makes the business look unreliable when the details do not match the real website.
Does review volume matter?
It can, as one signal among several, particularly when reviews describe specifics an assistant can use, such as responsiveness or the exact service performed, rather than generic praise.
How do I see which sources an assistant is actually using for my business?
Ask it directly and check whether it cites sources; some assistants show links. For a full picture across ChatGPT, Claude, Perplexity and Google's AI Overview, the AI-Ready Audit records exactly what each one says and, where shown, what it cites.
Next step
Now that you know how the decision gets made, find out what it currently produces for your own business with the $499 AI-Ready Audit.