How do I rank number 1 in ChatGPT?

There is no ranked list, so there is no position to reach

ChatGPT does not hold a ranked list of businesses you can move up. Every answer is written fresh for that one conversation, one piece of a word at a time, drawn from a probability distribution. Nothing in the reply is stored as position one, and nothing is stored as position two. There is no leaderboard behind the sentence.

You can prove that to yourself in about a minute. Open three fresh chats, ask the same buyer question in each, and compare what comes back. If the businesses differ, or their order differs, there was never a fixed position to hold. The mechanism behind that variation, including the sampling parameters both providers document, is set out in why AI answers change every time you ask.

There is also no single engine to be number one in. Across the 137 businesses in our 30 July 2026 scan with complete records from both engines, Gemini scored the same business higher than ChatGPT 77% of the time, and ChatGPT scored higher just 5% of the time. Two engines reading the same web produced different verdicts on the same business, on the same day. A rank would have to be a rank somewhere in particular, and there is no somewhere.

Three different questions hide inside this one

Three, and only one of them has no answer worth selling.

• Why is my competitor named when I am not? This is the real question most of the time, and it does have an answer. It is worked through in why ChatGPT recommends your competitor.
• How do I get named at all? Also answerable, and it is the rest of this page.
• How do I get listed first when the answer does give a list? This one we will not answer, because we have not measured it. Order inside a generated list is not a stored ranking, it moves between runs, and we have run no study on what shifts it. A firm quoting you a method for it is quoting a method for something nobody has published a measurement of.

What actually decides whether you are named

What independent sources say about you decides it, far more than what your own website says about itself. Our own data says that bluntly, and it is not the answer we would have preferred to find.

Across the 300 Australian businesses we scanned on 30 July 2026, the average on-site AI readiness score was 74.9 out of 100. Most of these websites are technically fine. Yet readiness barely tracked being named: across the 137 businesses with complete records from both engines, site readiness correlated with per-engine visibility at Pearson r = 0.116. A tidy site is close to no predictor of a mention.

What did track was being a known entity. The 36 recognisable anchor brands in the sample averaged 52.5 out of 100, against 36.4 for everyone else. And 41% of the 137 with complete two-engine records were never named once, by either engine, across every question asked.

Recognition on its own is still not enough. 26% of those same 137 were recognised when asked about directly and never named when a buyer asked who to use. The engine knew they existed and had no reason to put them forward. Why the two come apart at all is the difference between an engine recognising you and recommending you, and what to do about it is ChatGPT knows my business but never recommends it.

There is a mechanical floor underneath all of this, worth clearing because it is cheap rather than because it is a lever. Of the 300 sites, 13% block AI crawlers outright in robots.txt, 24% carry no JSON-LD structured data and 71% have no llms.txt. Only the first of those can single-handedly stop you being read at all. See should I block AI crawlers and does schema markup help with AI search.

Named in 4 of 6 buyer questions is the target that replaces rank

A rate is the unit that survives the variance: how many questions from a fixed set named you, out of how many were asked. Named in 4 of 6 is a fact somebody else can reproduce from stored answers. Number one is not a fact at all.

It is also the unit our own scoring is built on. Seventy per cent of every engine score we publish is exactly that count, mentions divided by questions asked, matched with a regular expression rather than by asking a model for a number. The remaining 30% is model-graded recognition and sentiment, and we say so rather than presenting the score as fully mechanical. The formula is published in full on the technology page.

Two rules make a rate useful rather than decorative. Pin the questions, so the next run measures the same thing. And require a change to survive a second consecutive run, because a move that appears once and vanishes was never a move. What counts as a good number once you have one is covered in what is a good AI visibility score.

What to do instead, in order

Work in the order the answer is assembled: be readable, be corroborated, then be measured.

1. Clear the mechanical floor. Check that robots.txt does not block the AI crawlers, that the homepage carries a title and a meta description, and that there is at least one JSON-LD block. This is a floor, not a lever, and it is the cheapest hour in the whole exercise.
2. Get corroborated off your own domain. Engines assemble answers from sources you do not own: reviews, directories, publications, comparison pages, the ordinary web. This is the part that moves recognition and the part you control least. The practical version is how to get recommended by ChatGPT, and the reviews question specifically is do Google reviews affect AI recommendations.
3. Correct what the engines already get wrong about you. Wrong prices, wrong hours, a suburb you left years ago. Fixing a fact an engine repeats is concrete work with a checkable result: ChatGPT has wrong information about my business.
4. Pin a question set and measure the rate. Then measure it again before you believe it. How to tell whether the work is working covers what a real change looks like against a run-to-run wobble.

None of those four are ranking work. They are the work that decides whether the sentence gets written with your name in it, which is the only outcome available.

What we will not claim

We have not measured the order businesses appear in inside a generated answer, so we will not sell you a method for changing it.

The limits of our own study sit on the record for the same reason. Each question in the 30 July 2026 scan was asked exactly once, which means the study measures whether a business was named, not how reliably it would be named across repeated runs. Two engines were queried, not five. Nothing about it is realtime, and we have no access to anyone's real ChatGPT sessions. Those constraints are listed on the technology page and inside the study itself.

If a firm answers this question with a method rather than a correction, that is the most useful thing you will learn in the meeting. The neighbouring question, can you guarantee ChatGPT will recommend my business, has the same shape and the same answer.