How to show up in ChatGPT answers

To show up in ChatGPT answers, your pages have to be reachable, quotable and trusted, in that order. Most businesses that are invisible in AI answers fail at the first step and spend their budget on the third.

Here is what each step actually involves, and what the vendor documentation says rather than what the market says.

Step 1: can the crawler reach you at all?

This is the step people skip, and it is the one that most often explains the whole problem.

Assistants do not browse your site the way you do. They send a crawler, take whatever HTML comes back, and work with that. If your content is assembled by JavaScript after the page loads, the crawler may receive an almost empty document. The page looks perfect in your browser and contains nothing an engine can quote.

The check takes a minute: view the page source, not the rendered page, and search it for a sentence from the middle of your content. If it is not in there, an assistant cannot see it either.

A static-first build avoids the problem entirely, which is why this site is built that way.

Step 2: are you blocking the crawler that actually cites you?

This is the expensive mistake, and it is very common.

Each of the major AI companies runs more than one crawler, and they do different jobs. One collects content that may be used to train a model. Another fetches pages to answer a question and link the result. They have different names and are controlled separately in robots.txt.

OpenAI’s documentation describes them separately:

OAI-SearchBot is used to surface websites in search results in ChatGPT’s search features.

That is the one that produces the link in the answer. The training crawler is a different agent:

GPTBot is used to make our generative AI foundation models more useful and safe.

Anthropic splits them the same way, with ClaudeBot for training and Claude-SearchBot for search quality. Perplexity does too:

PerplexityBot is designed to surface and link websites in search results on Perplexity. It is not used to crawl content for AI foundation models.

Here is why this matters. A blanket rule aimed at “AI scraping” usually catches the retrieval crawler as well as the training one. The site owner keeps their content out of model training, which may well be what they wanted, and also removes themselves from the answers that would have linked to them. They then conclude AI search does not work for their business.

Vendor	Trains models	Fetches to answer and cite
OpenAI	GPTBot	OAI-SearchBot, ChatGPT-User
Anthropic	ClaudeBot	Claude-SearchBot, Claude-User
Perplexity	None stated for foundation models	PerplexityBot, Perplexity-User

The names change. They have changed several times in the last two years, so a robots.txt written from a blog post in 2024 is probably wrong now. Check the vendor documentation, not a listicle, and re-check when you next touch the file.

Step 3: is there anything on the page worth quoting?

An assistant assembling an answer is looking for a passage it can lift and present as its own summary. That has consequences for how you write.

Put the answer first. Open each section with the direct answer in one or two sentences, then expand. Burying the point three paragraphs down means it gets skipped in favor of a competitor who did not bury theirs.

Make each passage stand alone. Take any paragraph out of the page and read it cold. If it only makes sense with the paragraph above it, quoting it would produce something misleading, and engines tend to avoid passages that do that.

Use real questions as headings. A heading is a strong boundary in the text. A heading phrased as a question makes what follows an answer.

Be specific. Vague claims are not quotable, because quoting them commits the assistant to nothing. Concrete processes, named tools, real constraints and actual numbers are.

The test for a quotable page is simple. Could a stranger read one paragraph out of context and repeat it accurately? If not, an engine will not risk it either.

Step 4: are you the same business everywhere?

Assistants have to work out which entity you are before they can describe you. That is harder than it sounds when a business appears under three variations of its name, with two different service descriptions and an old address.

Use the same business name, the same description of what you do and the same contact details everywhere you appear. Add Organization schema so the machine readable version agrees with the visible one. Consistency is what lets an engine resolve you confidently rather than hedging or confusing you with someone else.

This is also where a distinctive name earns its keep. A business whose name reads like a common phrase is genuinely harder for an engine to disambiguate, and there is no markup that fixes it.

Step 5: are you mentioned anywhere the engine already trusts?

This is the slow one, and it is the part no on-page work can substitute for.

Assistants draw on sources they already rely on. A new domain with no mentions anywhere is often not in the candidate pool at all, regardless of how well its pages are written. Being written about, listed, reviewed or referenced elsewhere is what changes that, and it takes months rather than weeks.

Anyone quoting you a timeline to first citation is quoting you a feeling.

What about llms.txt?

Be skeptical. A lot of agencies sell llms.txt as an AI visibility tactic and the evidence does not support it.

Google states the position plainly:

Doing so will neither harm nor help your site’s visibility or rankings in Google Search, as Google Search ignores them.

Large log studies point the same way. One analysis of over 137,000 domains found that 97% of published llms.txt files received no requests at all in a month, and that AI retrieval crawlers, the ones that produce citations, accounted for around 1% of the few requests there were. A separate study across roughly 300,000 domains found no correlation between having the file and being cited.

This site publishes one anyway, because it costs nothing to generate and the specification is still moving. It is not sold as a service and it is not presented as something that moves the needle. If someone quotes you for llms.txt implementation, ask them for the evidence.

What nobody can promise you

A citation. The engine chooses, the answer varies between sessions and phrasings, and no agency can prove a specific mention was caused by its own work rather than by something else that changed that month.

What is within anyone’s control is the list above: be reachable, be unblocked, be quotable, be consistent, and be mentioned elsewhere. That is the honest version of the offer, and it is worth being suspicious of a more confident one.

How to check where you stand

Ask the assistants the questions your buyers would ask, not your own brand name, and write down what comes back. Record whether you were named, how you were described, and which sources the answer used. Run the same prompts again in a few weeks.

One check is an anecdote. A fixed set of prompts, repeated, is a measurement.

For the fuller explanation of how AI answers get assembled and what influences them, read the AI search explainer. If you want the technical base handled, that is SEO services and on-page work.