ChatGPT optimization

ChatGPT is where a lot of buying research now starts, and it does not work the way a search engine works. It reads a few pages and writes one answer. You are either in that answer or you are not. This page covers the part of our GEO services aimed at that engine specifically, because what works here does not transfer to the others. For the full playbook on what actually earns a citation, see how to get cited by ChatGPT.

How does ChatGPT decide who to name?

By reading, not by ranking.

When a question needs current information, ChatGPT searches OpenAI's own index and the search data it licenses from Google and Microsoft, plus a small set of partner sources for things like local listings. It fetches a handful of results, reads them, and composes an answer from what those pages say.

That changes the target. A ranking system asks which page deserves position one. A reading system asks which sentences are clear enough to quote and confident enough to repeat. A page that buries its answer in the fourth paragraph loses to a page that states it in the first, even if the second page is thinner.

Does ranking in Bing get you into ChatGPT?

It used to be the standard advice. It is no longer true.

Through 2026 the overlap between ChatGPT's citations and Bing's results fell steeply, while overlap with Google's results rose. ChatGPT is not a wrapper on either one. It has its own index, and most of what it cites does not sit at the top of either search engine for the same question. Ranking is not what gets you quoted there.

We still make sure Bing has your sitemap and can crawl the site, because that costs an afternoon. We do not build a plan on it.

What does ChatGPT cite that other engines do not?

The pattern behind the table: ChatGPT trusts pages written by someone other than you when the question is about choosing, and trusts your own pages when the question is about understanding. So we write to be the definition, and we work to be listed by others for the shortlist.

What do we change on your pages?

Structure first, then substance.

Every target page gets a two or three sentence answer directly under the heading, before any preamble. Headings become the smaller questions a buyer actually asks: what it costs, how long it takes, what is included, what happens if it does not work. Sections stay short enough to lift whole rather than summarise. Facts get a named source and a date.

+28% to +41%

Then we check that the text exists in the HTML at all. ChatGPT's fetchers do not reliably run your JavaScript, so a page that paints its content after a script runs can read as an empty page. That is ordinary technical SEO work, and on most sites it is the single change that moves the most.

Which crawlers have to be open?

Three, and they do different jobs.

OAI-SearchBot builds the index that ChatGPT search reads, so it decides citations. ChatGPT-User fetches a page live when a user or a task asks about a specific URL. GPTBot collects training data, which does not affect search directly but does affect whether the model knows your name without looking.

Blocking OAI-SearchBot removes you from ChatGPT search. It is usually blocked by accident, through a firewall preset rather than a robots.txt line, which is why we test each one with a live fetch instead of reading the file and assuming.

The other engines read other indexes and want other crawlers, so a rule that opens one does nothing for the rest. The full six-engine comparison sits on the GEO services page.

Where do we get your name mentioned?

On the pages ChatGPT reaches for when someone asks who to hire.

That means review platform profiles that are complete and actually reviewed, professional company profiles kept current, and comparison lists published by other people. Getting into other people's lists is real outreach work: most list authors gate inclusion on a verified review profile, so the review work has to come first and it takes months, not weeks.

We also publish your own honest comparison content, including the cases where a competitor is the better choice. Engines quote pages that concede something far more readily than pages that claim everything.

What does the work look like?

How do we measure it?

By counting, not by feeling.

Each week we run the prompt set and log three states per question: named, linked, or absent. We record the sources ChatGPT used, so the report names the competitor and the publication that took your place. Over time the useful number is the share of your prompts where you appear at all, tracked against the same list you agreed at the start.

We do not report a single AI visibility score with no method behind it. Scores like that move for reasons nobody can explain, which makes them useless for deciding what to do next.

Related work sits on the Perplexity optimization page, and the starting measurement is the AI visibility audit. If you want to see how we run this for a team in one of our regional markets, our market page walks through the same process.

What does not help here?

Publishing an llms.txt file. No engine uses it for citations, Google said in June 2026 that it has no effect, and almost nobody ever requests one.

Adding FAQ markup to chase a rich result. Google retired FAQPage rich results in May 2026. We keep the markup where it mirrors visible text, and we keep the visible questions, because those were always the part doing the work.

Buying a visibility score without fixing the pages. And writing the same page once per engine, which is how a site ends up with six near-identical pages that none of the engines quote.

How do we start?

Book a free 30-minute consultation. We check whether the OpenAI crawlers can reach you, whether your key text is in the raw HTML, and what your markup says, then name the first three fixes.

If you want the full picture across every engine, the AI visibility audit runs the same checks in depth. Plans are flat monthly retainers with starting prices on the pricing page, month to month, and you can stop at the end of any month.