How to rank in ChatGPT

Short answer: you do not rank, you get named

There is no ranking in ChatGPT. No position one, no page two, no list of ten you can climb. Asked for a local recommendation, it returns one to five business names, and you are either in that set or you are not mentioned at all. People search for “how to rank in ChatGPT” because that is the vocabulary search engines taught us, but the thing being asked for is how to get named, and the work is different as a result.

ChatGPT does not have opinions about your business. It retrieves.

When someone asks it to recommend a dentist in Charlotte, it queries an index — mostly Bing’s — pulls back whatever local sources it can find, and composes a short list from them. Getting recommended means being present and well-reviewed in the sources it reaches for. That is an off-site problem, and it is why a new website almost never changes the answer.

What is actually happening when you ask

Three steps, roughly:

1. It decides the question is local. “Best med spa in Austin” gets routed to retrieval rather than answered from memory, because the model knows its training data is stale for this.
2. It retrieves. Bing’s index does most of the work. Business listings, review platforms and directory profiles are the raw material.
3. It composes. It names one to five businesses and usually justifies each in a clause — “highly rated for injectables”, “known for same-day appointments”. That justification is lifted from what it read.

The practical consequence: you are not competing on content, you are competing on presence in a handful of sources. The businesses that get named are the ones those sources describe clearly and rate well.

The eight inputs, ranked by how often they are the problem

1. An unclaimed or incomplete Bing Places listing. The single most common cause, and the cheapest to fix. Why Bing Places decides your ChatGPT visibility covers this in full.
2. Review volume and rating. Observed behaviour puts the practical entry point somewhere near thirty reviews at 4.3 stars, higher in competitive metros. How many reviews before AI recommends you has the detail.
3. Review recency. A business with two hundred reviews and nothing in six months tends to lose to one with sixty updated weekly.
4. Name, address and phone consistency. If four sources disagree about your address, retrieval treats you as four weak businesses rather than one strong one.
5. Category and service specificity. “Med spa” is a category. “Botox”, “CoolSculpting” and “Morpheus8” are what people actually ask for. If your listing does not name them, you cannot match the question.
6. Directory coverage. Foursquare, Yelp and Apple Business Connect feed multiple engines. Category-specific directories matter more than general ones.
7. Third-party mentions. Local editorial, roundups and community discussion give the model something to cite that is not you talking about yourself.
8. Crawler access. Rare, but fatal when it happens: a robots.txt that blocks GPTBot or OAI-SearchBot removes your own site from the pool entirely. Check yours here — it takes a few seconds.

What to do first

If you do nothing else, do these three, in this order:

• Claim Bing Places and complete it fully. About an hour per location. Free. It affects ChatGPT and Copilot together.
• Reconcile your name, address and phone across Google Business Profile, Bing Places, Apple, Foursquare and Yelp. Pick one canonical format and make everything match it exactly.
• Start asking for reviews on a schedule. Not a campaign — a system, built into discharge or checkout, running every week.

Everything else is worth doing and none of it substitutes for those three.

How long it takes

Listing and structured-data work can register within two to four weeks. Review velocity and third-party presence compound over three to six months. Anyone promising a faster result is describing something other than how retrieval indexes update.

What this will not do

It will not put you first every time. Engine output varies between runs even for identical prompts, which is why a single result tells you very little and why measurement has to be repeated. It will not override a genuinely poor rating — if your reviews say the front desk is rude, being retrieved more often surfaces that faster.

And it will not work as a one-off. The sources refresh; so does the answer.

How to check whether it worked

Pick ten prompts a patient would actually type, run each three times in a fresh signed-out session, and record whether you were named, who was named instead, and which sources the answer cited. Repeat monthly. The prompt set builder will generate the list for your market.

That last column — the source cited — is the one that tells you what to fix next.

The caveat worth stating

No AI platform publishes how it selects sources. Everything above is drawn from observed output and correlations documented across the industry over roughly the last eighteen months, not from confirmed ranking factors. The mechanisms change, sometimes quickly. Treat any account of this — including this one — as a working model rather than a specification.