ChatGPT SEO: how to get your business recommended by ChatGPT

Start with scale. OpenAI reported roughly 900 million weekly active users in February 2026, more than double the figure a year earlier. ChatGPT accounts for something close to 87% of all AI referral traffic reaching websites. If you are going to care about one AI platform, this is the one.

Now the other half of the picture.

ChatGPT’s per-query citation rate sits at around 0.7%. Across a large sample of queries, the chance of any given site being cited is tiny. When browsing is triggered, ChatGPT typically selects between three and ten sources for a response. Not thirty. Not a page of results. Three to ten.

And research into local business visibility found ChatGPT surfacing just 1.2% of business locations, against 35.9% appearing in Google’s traditional three-pack for equivalent searches.

Huge reach, brutal selectivity.

Which is exactly why a citation here is worth more than almost anything else you can earn in search right now, and why most businesses have never had one.

One more detail that changes how you should think about success: ChatGPT mentions brands roughly 3.2 times more often than it cites them with a link. Being named without a link still influences the buyer. It just will not show up in your analytics, which is why measuring this properly needs a different approach entirely.

ChatGPT runs on two layers, and which one fires changes everything about what you can influence.

The training layer is static between model releases. It is what ChatGPT knows without looking anything up. Broad, deep, and impossible to update quickly.

The retrieval layer is live. It queries a web index, reads candidate pages and synthesizes an answer with citations.

The obvious question is when each one fires. The answer is more predictable than most people expect.

One study found commercial-intent prompts trigger web search 53.5% of the time, against just 18.7% for informational queries. Semrush’s analysis of 80 million ChatGPT queries put the overall search trigger rate at around 46%.

Terms that reliably pull live results: “best”, “reviews”, “comparison”, “latest”, “pricing”, “near me”, or any year. Broad definitional questions tend to be answered from training alone.

When search does fire, the sequence runs roughly like this:

Step 3 is the one nobody optimizes for. It is also the one that decides everything.

You will read everywhere that ChatGPT runs on Bing. That was straightforwardly true, and it is now an oversimplification worth correcting, because the correction is actionable.

The evidence for the Bing relationship is strong. Seer Interactive analyzed over 500 citations and found 87% of ChatGPT search citations matched Bing’s top-10 organic results, against only 56% matching Google’s. That is not a coincidence, it is an architecture.

But by mid-2026 the most defensible reading is a blended stack rather than a single upstream index. OpenAI’s own crawler contributes. A reranking layer sits between retrieval and the final answer, optimizing for properties that classical search ranking does not measure at all.

What this means in practice:

• Bing indexation is a floor, not a ceiling. A page absent from Bing and absent from OpenAI’s crawl is very unlikely to be retrieved. So Bing indexation remains the first thing to check.
• Bing rank position correlates only loosely with citation. Being number one in Bing does not get you cited. It gets you into the pool.
• So run two checks, not one. Are you indexed in Bing? And separately, when the model does retrieve your page, is it structured well enough to survive the rerank?

Practical starting point: verify your site in Bing Webmaster Tools, submit your sitemap, and implement IndexNow. Most businesses have never opened Bing Webmaster Tools. It takes an afternoon and it is the single highest-return technical task on this page.

The 15% problem

Here is the finding that should reshape how you approach this.

An AirOps study analyzing 548,534 pages across 15,000 prompts found that ChatGPT cites only around 15% of the pages it retrieves. Zyppy’s earlier research landed on a similar figure.

Read that carefully. Eighty-five percent of pages that ChatGPT pulls in, reads and evaluates never appear in the answer. They cleared the hardest technical hurdle and then lost anyway.

This gives you two distinct problems that require different work:

Almost every ChatGPT optimization guide you will find addresses the left column and stops. That is why so many brands do the recommended work, get indexed properly, and still never appear.

The right column is where the discipline actually lives.

No published ranking factors exist. OpenAI has never released a list, and anyone presenting you with one has assembled it from inference.

What does exist is a substantial body of citation research, and it converges on a fairly consistent picture.

A pattern runs through all of it. ChatGPT is not assessing whether your page is good. It is assessing whether your page is easy to quote and safe to trust.

Those are different tests, and most content fails the first one on formatting alone.

OpenAI runs three crawlers with separate jobs. They can be controlled independently in robots.txt.

Two things worth knowing.

Blocking GPTBot does not block OAI-SearchBot. These are separate decisions. A publisher can legitimately opt out of training while remaining fully visible in ChatGPT’s answers. Very few realize this is available.

OpenAI warns that robots.txt rules may not apply to ChatGPT-User, since it is user-initiated rather than a crawl. This differs from Anthropic, where all three bots honour robots.txt. Do not assume consistent behaviour across vendors.

And the thing we find on roughly half the sites we audit: the block was never intentional. Someone added a blanket Disallow for GPTBot after reading an article about AI training. Or Cloudflare’s bot-fight mode is on and silently dropping AI crawlers regardless of robots.txt. Or a security plugin did it by default.

Nothing reports this. It is invisible in every analytics dashboard. The only reliable evidence is server logs, which is why that is step one of our process. The full list is in our AI crawler reference.

This is the objection we hear most, usually from businesses who have invested heavily in SEO and reasonably expect it to carry over.

It carries over partially. The data is fairly blunt about how partially.

• Pages ranking first on Google are cited by ChatGPT around 3.5 times more often than pages outside the top 20. So ranking helps.
• But only about 12% of URLs cited by ChatGPT rank in Google’s top 10 for the same query.
• And a 2026 study found only 12% of pages ranking number one on Google get cited by ChatGPT at all.
• EMGI’s April 2026 research found 44% of SaaS brands with strong Google rankings had no ChatGPT visibility whatsoever.

Strong Google SEO is a contributing factor. It is not a guarantee, and for a meaningful minority of businesses it delivers nothing at all on this platform.

The reason is structural rather than mysterious. ChatGPT reads a different index, applies a reranking stage that optimizes for extractability rather than relevance, and weights third-party corroboration far more heavily than Google’s algorithm does.

Different test, different result.

The named-versus-cited gap is the one to watch. Given ChatGPT mentions brands around 3.2 times more often than it links them, a brand with a high named rate and low referral traffic is not failing. It is being recommended in conversations that never touch your analytics.

Reporting only on referral traffic would materially understate the result, which is exactly the mistake most agencies make. Our full methodology is documented in our AI visibility measurement guide.

ChatGPT is the largest single platform and it is still one of six.

Because ChatGPT and Copilot both lean on Bing, work done for one benefits the other. That is the cheapest cross-platform win available and almost nobody takes it.

Beyond that pairing, transfer is poor. Studies consistently find around 80 to 91% of cited URLs appear on only one engine. Winning ChatGPT tells you very little about the rest, which is why we run all six through one AI optimization program built on technical SEO and content strategy foundations.