# AI search optimization: how assistants find and cite your brand

[Canonical HTML page](https://gradiently.design/guide/llm-seo)

ChatGPT, Claude, Perplexity and Google's AI answers all have to get their facts about you from somewhere. This is how they find them, and what a marketer can do so the facts they find are right.

## The short version

- AI search optimization is the work of making your brand easy for AI assistants to find, understand and quote accurately.
- Assistants answer from what they learned in training and from pages they fetch live, so both your long-standing reputation and your current pages matter.
- AI crawlers such as GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot are documented as respecting robots.txt, so blocking them can keep your pages out of AI answers.
- Pages that answer the question in the first paragraph, in short sentences that stand alone, are easier for an assistant to quote.
- llms.txt is a proposed standard file that gives AI tools a plain map of a site; it is cheap to publish, but not every assistant reads it.

**AI search optimization** is making your brand easy for AI assistants to find, understand and quote correctly. In practice it means four things: let AI crawlers read your site, write pages that answer the question in their first paragraph, keep your facts identical everywhere they appear, and give machines a plain map of what you publish. Most of it is good SEO done carefully. A little of it is new.

## How AI assistants find information about a brand

An assistant answers from two places. The first is what the model learned in training, a snapshot of the public web and other sources taken months or years before you ask. The second is live retrieval: for many questions, the assistant searches the web, fetches a few pages and writes its answer from them, often with links. Optimising for AI means being findable and clear in both.

| Source | What it means for you | What you can influence |
| --- | --- | --- |
| Training data | Older, broad picture of your brand | Being described consistently, for a long time, in many places |
| Live search | Current pages fetched at question time | Crawler access, clear pages, ranking in ordinary search |
| Tools and connectors | The assistant uses your product directly | An MCP server or app that exposes accurate data |

Three routes from your brand into an AI answer. Live search is where a marketer can make the quickest difference.

## Let AI crawlers read your site

AI companies run their own crawlers, and the major ones say they respect robots.txt. Some sites block them by default through a hosting setting or a security plugin, sometimes without anyone deciding to. If an assistant cannot fetch your pages, it will answer from whatever else it finds, which may be a two-year-old review or a competitor's comparison page.

```text
# robots.txt: allow search and AI assistant crawlers
User-agent: *
Allow: /
Disallow: /account/

User-agent: GPTBot
User-agent: OAI-SearchBot
User-agent: ChatGPT-User
User-agent: ClaudeBot
User-agent: Claude-User
User-agent: PerplexityBot
User-agent: Google-Extended
Allow: /
Disallow: /account/

Sitemap: https://example.com/sitemap.xml
```

A robots.txt that names the main AI crawlers explicitly. Google-Extended is a control token for Google's AI training rather than a separate crawler. Keep private areas disallowed for every agent, and check your CDN or firewall isn't blocking them separately.

> **Training and answering are separate choices** Some companies use one crawler for training and another for live answers. You can allow the answering crawlers while blocking training ones, but read each company's documentation first, because the names and their roles change.

## Write pages an assistant can quote

When an assistant writes from a page it fetched, it lifts the clearest statement it can find. A page that opens with three paragraphs of story before the answer gives it nothing to lift. A page that answers in its first sentence, then explains, gives it exactly what it needs and gives the human reader the same courtesy.

### Hard to quote

- The answer sits below a long introduction
- Facts hidden inside images or sliders
- Sentences that only make sense with the one before
- Vague claims: "industry leading", "best in class"
- Different product names on different pages

### Easy to quote

- The answer in the first paragraph
- Facts written as plain text in the HTML
- Short sentences that stand on their own
- Specifics: sizes, steps, names, dates
- One name and one description, used everywhere

Structured data helps machines read the page too. FAQ, product, organisation and article markup describe what a page is in a form any crawler understands. It is not a guarantee of being cited, but it removes guesswork. If your pages are shared as links, [Open Graph tags](https://gradiently.design/guide/open-graph-tags) and a proper [link preview image](https://gradiently.design/guide/open-graph-image-size) help the human side of the same click.

## Publish an llms.txt file

llms.txt is a proposed convention: a Markdown file at the root of your site that tells AI tools, in plain language, what the site is and where its most useful pages are. It is not an official standard and no assistant is obliged to read it. It takes an hour to write, though, and it forces you to describe your business in a few exact sentences, which is valuable on its own.

```text
# Harbour Bakery

> A neighbourhood bakery at 2 Quay Street, open 7am to 3pm,
> Tuesday to Sunday. Sourdough, pastries and coffee.

## Key pages

- Menu: https://example.com/menu (breads, pastries and drinks)
- Visit: https://example.com/visit (address, hours, parking)
- Orders: https://example.com/orders (celebration cakes, 48 hours notice)

## Optional

- Journal: https://example.com/journal (recipes and seasonal news)
```

A small llms.txt, simplified for reading. The proposal writes each entry as a Markdown link followed by a short note. The opening quote block is the description you most want an assistant to repeat.

## Keep your facts identical everywhere

Assistants compare sources. If your site says you open at 7am, your Google Business profile says 8am and a directory says you closed in 2023, the answer will be hedged or wrong. Pick one name, one line describing what you do and one set of facts, then update every profile you control. [Get your brand mentioned by ChatGPT](https://gradiently.design/guide/perplexity-chatgpt-search-brand) covers the off-site side: reviews, comparisons and the places assistants look beyond your own pages.

1. **Write a one-line description** What you are, for whom, and where. Use it word for word on your site, profiles and press pages.
2. **List your core facts** Name, location, hours, products, founding year, contact. One source of truth, shared with anyone who edits a profile.
3. **Audit your profiles** Google Business, social bios, directories, marketplaces and partner pages. Fix every mismatch.
4. **Ask the assistants** Ask ChatGPT, Claude, Perplexity and Google what they know about you. Note what is wrong and trace where it came from.
5. **Repeat every quarter** Assistants change their sources and models. A short check every few months catches drift early.

## How Gradiently approaches AI search

Gradiently's own Field Guide is written for people and assistants at once. Every article answers its title in the first paragraph and carries a short summary of sentences that are each true on their own, so an assistant can quote one without the rest. FAQs are published as FAQPage data. The site publishes an llms.txt and a full-text llms-full.txt (see `gradiently.design/llms.txt`), offers a Markdown copy of every guide article, and its robots.txt names the main AI crawlers and allows them on public pages.

There is a second route, too. Gradiently works inside ChatGPT, Claude and any assistant that supports remote MCP servers, so an assistant can search Marks and create designs directly rather than describing the product from memory. [What is MCP](https://gradiently.design/guide/what-is-mcp) explains how that connection works, and [connect an AI assistant](https://gradiently.design/guide/connect-ai-assistant) shows the setup.

## FAQ

### What is AI search optimization?

It is the work of making your brand easy for AI assistants such as ChatGPT, Claude and Perplexity to find, understand and quote accurately, through crawler access, clear pages and consistent facts.

### Is LLM SEO different from normal SEO?

Mostly it is the same work done carefully: crawlable pages, clear answers and good reputation. The differences are allowing AI crawlers, writing sentences that can be quoted alone and optionally publishing llms.txt.

### Does llms.txt help with ChatGPT?

llms.txt is a proposed convention, and no assistant is obliged to read it. It is cheap to publish and makes your site easier for any tool that does read it to understand.

### Should I block GPTBot in robots.txt?

Blocking AI crawlers can keep your pages out of AI answers, so most brands that want to be recommended allow them on public pages. Some companies use separate crawlers for training and for live answers, which you can treat differently.

### How do I check what ChatGPT says about my brand?

Ask it directly, along with Claude, Perplexity and Google, using the questions your customers would ask. Note any wrong facts and trace them back to the page they came from.