Why does an AI recommend one business over another?
It is not a ranking and it cannot be bought. A model assembles the answer from whatever it can read about businesses in that area …
Because the assistant could read your competitor and could not read you. When a customer asks who to call, ChatGPT searches, reads a few pages, and names the business whose claims it could verify. Your competitor's reviews and facts were somewhere it could parse; yours were inside a Google widget and a website it cannot read. It is not judging quality. It is answering from evidence.
The assistant is not ranking you. It is looking for evidence, and finding your competitor's.
We ran the check for a roofing company with 1,118 five-star reviews. When a stranger asked the assistants who to hire, they named a competitor, and the roofer with the thousand reviews was not in the answer at all. That is the pattern in most checks we run, and it is not a judgment about the work. The assistant never saw the 1,118 reviews. They live inside Google's review widget, which a crawler cannot parse, on a listing the assistant reads as a name, a phone number and a star count. The competitor had a page somewhere that said, in plain readable text, what it did and what customers said. The assistant read that, corroborated it against a directory, and answered.
An assistant gives one answer, occasionally three. There is no page two. So the gap between being named and not being named is the whole game, and it is decided by what the assistant could read, not by who is better.
Ranking well on Google is evidence that Google's algorithm likes your site. It is not evidence the assistant can use about your business. HubSpot's measurement puts only about 40% of AI citations on page one of Google; the rest come from review platforms, directories, forums and pages that do not rank. A competitor with a worse ranking and a readable record beats you with a better ranking and none.
The three ways a competitor becomes the answer, none of which involve being better.
When we trace why an assistant named someone else, it is almost always one of these.
They are readable and you are not. Their reviews are published as text on a page; yours are in a widget. Their services and service area are written out; yours are in an image. The assistant quotes what it can parse.
They are consistent and you are not. Their name, phone, hours and area match on Google, Yelp, Bing and their site. Yours has two phone numbers and an old address on a directory. An assistant that finds a contradiction lowers its confidence and moves on.
They are present where it reads and you are not. For your trade the assistant habitually reads certain sources: BBB, a trade directory, Reddit threads, a review platform. Your competitor has a profile there. You do not, so the sources the assistant trusts have never heard of you.
It is not the number of reviews. Twenty honest reviews published as structured data outrank two thousand the assistant cannot parse. It is not your star rating; a real 4.2 with specifics is more citable than a suspicious 5.0 with none, and the models are tuned to notice. And it is not paid placement. No one can buy the answer, including your competitor.
How to find out which one it is, and what closes the gap.
Ask the assistants directly, and read what they cite.
Publish your reviews and business facts as a structured record on a domain you own, so there is a page about you an assistant can read and quote. Make your name, phone, hours and area identical everywhere it looks. Get a profile on the two or three sources it cited when it named your competitor. Then re-ask the same questions a month later. Our free check does the first step for you and shows you exactly which pages the assistants read to name someone else; the subscription does the rest and re-checks every Monday.
Because the assistant could read theirs and not yours. Review count only matters when the reviews are somewhere an assistant can parse. Twenty readable reviews beat two thousand locked in a widget.
No. There is no paid placement inside an assistant's recommendation. If they are named, it is because a page about them was readable and checkable. You can do the same.
There is no directory to add yourself to. What you can do is publish a record the assistant can read, on your own domain, and make sure the sources it already trusts know about you. That is what an AI-readable review site is.
Try it for 14 days. If it isn't what we said, we refund your first month in full and take the site down.