How can I make my website appear in ChatGPT answers?

There is no ranking to climb, no submission form, and nothing to buy. Whether ChatGPT names your website is decided by a two-stage process, and understanding which stage you are failing determines everything about what to do next.

How ChatGPT decides

When someone asks a question that needs current or specific information, ChatGPT does not simply recall it. It runs a retrieval step — searching an index and pulling a shortlist of candidate pages — and then the model reads those candidates and writes an answer, quoting and attributing the passages it finds most useful.

That gives you two separate gates with completely different requirements.

You have to clear retrieval like an SEO, then win selection like an editor writing a pull quote.

Failing stage one means you are never considered. Failing stage two means you were considered and someone else was quoted. The remedies share almost nothing, which is why generic advice about AI visibility is so often useless — it addresses one stage while your problem is in the other.

Stage 1: clearing retrieval

Four requirements, all binary, all cheap to check.

GPTBot must be allowed. Check robots.txt, then check your CDN or WAF bot-management settings — which is where the block usually is. Many platforms ship presets that deny AI user agents by default, so the block is inherited rather than chosen. Verify from outside your own network.

The answer must exist in raw HTML. This is the single most consequential and least known requirement. Googlebot renders JavaScript in a headless Chrome engine; AI crawlers do not. An analysis by Vercel and MERJ covering more than 500 million GPTBot fetches recorded zero JavaScript execution — it downloaded JavaScript files roughly 11.5% of the time and never ran them, with the same behaviour observed for ClaudeBot and PerplexityBot.

Test it in five minutes: disable JavaScript in your browser and load your ten most important pages. If the answer text is not there, ChatGPT does not see it, no matter how well the page ranks in Google.

The page must be indexed and reasonably relevant. Retrieval leans on the same signals as organic ranking. You cannot be cited from a page that was never retrieved, and you will not be retrieved if your fundamentals are weak. This is why SEO remains the foundation rather than something AI visibility replaces.

The server must respond quickly. AI crawlers operate under tight timeouts, often one to five seconds. A slow response can mean no crawl at all, with no error anywhere to tell you.

Stage 2: winning selection

Once you are in the candidate set, the model chooses which passages to quote. Four properties decide it.

Self-contained answers. The model lifts passages, not pages. A two- or three-sentence chunk that makes sense without the surrounding paragraphs is far easier to use than one that builds across half a page. Test yours by copying three sentences out of context — if they still answer something, they are extractable.

Question-led structure. Phrase headings the way a person asks. “How do AI engines choose what to cite?” matches a real prompt; “Citation mechanics” matches nothing. Follow each heading immediately with the answer, then add nuance.

Specific, attributable claims. Vague statements get paraphrased and lose the attribution. A dated, sourced number gets lifted with your name attached, because there is nothing else to credit it to. The Princeton and IIT Delhi GEO study found Statistics Addition, Quotation Addition and Cite Sources were the three highest-performing optimisation methods tested, delivering roughly 30–40 percent relative improvements in AI visibility.

Structured data that matches the visible text. Article, FAQPage and Organization markup give the model labelled, unambiguous facts. A mismatch between markup and content is worse than no markup at all.

The part you do not control

The uncomfortable finding, once teams start measuring, is that a competitor ranking below them on Google is frequently named first by ChatGPT. The usual reason is corroboration.

A model weighs what the rest of the web says about a brand far more heavily than what the brand says about itself. Ten independent sources describing you the same way is stronger evidence than any amount of your own copy — and crucially, unlinked mentions count. Reviews, comparison posts, industry roundups, forum threads and news coverage all shape what a model believes.

Reddit in particular is cited disproportionately often, which is why forum presence has become a legitimate part of the discipline rather than a growth hack. Our guide to Reddit visibility covers doing that without being obnoxious about it.

Consistency matters more than volume here. Ten sources describing you identically beats fifty describing you five different ways, because the second pattern gives a model no stable fact to rely on. That means the practical work is often unglamorous: making sure your category description, your product names and your positioning are stated identically everywhere they appear.

A five-step checklist

1. Load your key pages with JavaScript disabled. If the answer is not there, fix rendering. Nothing else matters until this passes.
2. Confirm GPTBot is allowed in robots.txt and at the CDN, verified from outside your network.
3. Give each important question its own page, answered in the first two sentences, with question-led headings.
4. Add one thing nobody else has — a dated statistic, a named source, original data. This is what gets quoted with your name on it.
5. Build consistent third-party description in the places that already get cited in your category, then track whether you start being named.

Steps one and two can be done this afternoon and are frequently the entire problem. Steps three to five are a quarter.

How long it takes

Timelines differ sharply by engine, and confusing them causes most misjudgements about whether the work is landing.

Perplexity often reflects a new or updated page within days, because it cites live retrieval openly. Use it as an early indicator that something landed, not as a measure of your overall position.

Gemini tracks closely to Google’s view of the web, so changes appear on roughly a recrawl cadence and classic SEO strength carries over directly.

ChatGPT is slowest and most dependent on third-party sources catching up. It is also the one most of your buyers actually use, which makes it the least satisfying and most important engine to be patient with.

Expect a readable trend after four to eight weeks of weekly checks. Never judge on a single answer — generated responses vary between identical runs, so one result is noise and only the trend is signal.

Which pages to build first

Not every page benefits equally. Three types earn citations reliably, and they are not the ones most content plans start with.

The single-question page. One buyer question, answered completely, with the answer in the first two sentences and the nuance below it. These are unglamorous and they are what assistants quote, because they map exactly onto how a prompt is phrased. A site with thirty of these outperforms a site with three hundred general pages.

The comparison page you are honest on. “X vs Y” queries are enormously common in assistant conversations, and models heavily prefer sources that read as balanced. A comparison that names situations where your product is the wrong choice is far more likely to be cited than one that does not, because it looks like analysis rather than marketing. It also converts better, for the same reason.

The page with a number nobody else has. Original data — a survey, a benchmark, an analysis of your own operations. It gets cited by assistants and by other writers, which then feeds the corroboration layer. One of these is worth more than a quarter of general publishing.

What to do with pages you already have

Most sites do not need thirty new pages. They need their existing best pages restructured so the answers are extractable, and split where one page is currently attempting six questions.

Take the five prompts where a competitor is named and you are not, find the page on your site that should have been the answer, and ask why it was not chosen. Usually the answer is visible immediately: the page answers the question in paragraph six, or answers it alongside five other things, or answers it without stating anything concrete enough to quote.

Four things that do not work

Submitting your site. There is no form. Anyone selling one is selling something else.

Paying for placement in organic answers. You cannot buy a citation. Google has begun placing advertising inside AI surfaces, but those are labelled ads and separate from cited sources. Any vendor promising guaranteed inclusion in an assistant’s organic answer is describing something outside their control.

Stuffing your page with brand mentions. Models reason about entities, not string frequency. Repeating your name does not make you more citable; it makes the page worse to read.

Publishing volume. Fifty thin pages produce fifty things nobody quotes. One page that answers a question completely and contains a number nobody else has will outperform all of them, because extraction rewards depth on a single question rather than coverage across many.

Sources

• Vercel and MERJ — analysis of 500M+ GPTBot fetches finding zero JavaScript execution; GPTBot downloaded JS files ~11.5% of the time without running them; same behaviour for ClaudeBot and PerplexityBot.
• Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan and Deshpande — “GEO: Generative Engine Optimization”, ACM SIGKDD 2024 (arXiv:2311.09735); Statistics Addition, Quotation Addition and Cite Sources as top-performing methods, ~30–40% relative visibility improvement across ~10,000 queries.
• OpenAI documentation — GPTBot crawler identification and robots.txt directives.

Frequently asked questions