How to get your business recommended by ChatGPT
Written by
Odin Lund
Category
Playbook
Date
Share this article
Why ChatGPT recommends one business over another
ChatGPT has no opinion about your business. It assembles an answer from sources it retrieved at the moment of the question, and it repeats the claims those sources agree on.
We traced 4,222 answers in one category back to the sources behind them. Three factors separated the businesses that were named from the ones that were not, and the strongest was consistency: businesses whose facts matched across their site, their listings and their reviews were repeated, while businesses with conflicting facts were hedged or dropped.
The second factor is where those sources live. In that data set the third-party sources carrying the client produced 18,742 retrievals against 4,319 from the client’s own website. Work only on your own pages and you are working on about a fifth of what the model reads.
Start from the questions your customers actually ask
Most businesses guess at this. Write down the questions a customer asks on the way to choosing you, in their words, from the first vague one to the final comparison, and run each on every model you care about.
Stage matters more than volume. In our legal data set, decision-stage prompts named the client in 37.8% of answers while comparison prompts named it in 6.3%. The shortlist forms at the first query: the businesses named in the first answer were, with few exceptions, the businesses named later, and businesses absent from the first answer rarely re-entered.
Record three things for every answer: whether you were named, what was said about you, and the page, listing or article behind each claim. The third is what makes the rest actionable.
Make every source state the same facts
Now fix what you control. Gather the facts a customer needs — what you do, for whom, where, since when, to what standard — and write one version of each.
Then make every source state that version: your own pages first, then your structured data, then every directory listing and review profile that describes you. A conflict anywhere is a claim the model will hedge on.
Publish answers to the questions from the previous step, one page per question, with the answer in the opening paragraph rather than buried under an introduction. Models retrieve passages, not pages.
Earn the sources the model already cites
The rest of the work is off your site. In most categories a small number of domains account for most citations, so the question is narrow: which sources did the models actually retrieve in your category, and are you present on each with the same facts?
Claim and correct what already exists. Pursue placement where you are missing. This is the slowest part and the one that compounds.
Then keep checking, because answers move as sources move and competitors publish. In our average engagement share of voice moves from single digits to first place in the category within three months, and it holds only because the checking continues.
Share Blog
