ChatGPT SEO: SEO and GEO to get your business cited by AI

ChatGPT answers in two ways. For a general question, it often relies on what it learned during training. For a question that calls for recent or local information, such as “Which accountancy firm should I choose in Leeds?”, it searches the web, reads a few pages and writes its answer, citing its sources.

For this search, OpenAI uses its own crawler, OAI-SearchBot, as well as third-party search engines. A business that neither this crawler nor the search engines know about is unlikely to appear.

The answer reads like advice: a few names, a reason for each, and links to the pages consulted. Your prospect often reads this advice and goes no further.

OpenAI is explicit: for your website's content to be used in ChatGPT's summaries and snippets, OAI-SearchBot must not be blocked. If a page is disallowed to it, ChatGPT can at best show its link and title, without using its content.

OpenAI uses another crawler, GPTBot, to collect pages that may be used to train its models. The two settings are independent: you can refuse training and remain visible in search.

Example robots.txt file, if you refuse training but want to be cited:

OpenAI says that a change to the robots.txt file takes about 24 hours to be taken into account. The AI crawlers guide details the settings for each AI platform.

Since ChatGPT also relies on third-party search engines, a well-indexed website has a head start. Check that your important pages are indexed in Google, with Search Console, and in Bing, with Bing Webmaster Tools.

ChatGPT uses the passages that clearly answer the question asked. For each service and each area you cover:

• a dedicated page that says in the very first sentence what you do, for whom and where;
• verifiable facts: lead times, price ranges, certifications, years of experience;
• an FAQ that uses your customers' questions, in their words;
• text that can be read in the page's code, not locked in an image, a PDF or a module that needs JavaScript.

To recommend a business, ChatGPT cross-checks what several sources say about it. “Best … in …” comparisons, your industry's directories, your Google Business Profile, customer reviews, local press articles: these are often the pages it consults. Also make sure your name, address, services and prices are the same everywhere.

An example built around a fictitious business.

Atelier Dumont, a joiner in Manchester, asks ChatGPT the question its customers ask: “Which joiner should I choose for a bespoke kitchen in Manchester?”. The answer cites three competitors, never Atelier Dumont. Looking at the sources shown and at the website, three causes emerge:

• Atelier Dumont's robots.txt blocks OAI-SearchBot, a setting left behind by a security plugin;
• the website has no page dedicated to bespoke kitchens: the information gets lost in an “Our work” page made up of photos;
• the three competitors appear in a comparison of Manchester joiners published by a local website; Atelier Dumont isn't in it.

The fixes follow the same order: unblock the crawler; create a “Bespoke kitchens in Manchester” page with lead times, guide prices and an FAQ; then offer the comparison website photos and verifiable reviews. None of these three fixes involves “talking to ChatGPT”: they make the business readable, precise and verifiable.

• Looking for a sign-up form: there isn't one for ChatGPT's answers.
• Repeating keywords: ChatGPT looks for a clear answer, not repetition.
• Publishing pages in bulk with no new information: they give it nothing to cite.
• Buying reviews: it's illegal in the UK and against review platforms' rules.
• Relying entirely on the llms.txt file: its effect hasn't been demonstrated; it doesn't replace any of the four requirements above.

The principle is the same; the crawlers change.

• Perplexity: its PerplexityBot crawler is used to show and link to websites in its results; Perplexity states that it isn't used to train models. Let it through.
• Gemini: it relies on Google Search. The Google-Extended control changes nothing about your presence in Google Search, but it limits the use of your pages by some other Google AI systems.
• Claude: Anthropic distinguishes ClaudeBot, for training, from Claude-SearchBot, for search, and from Claude-User, which reads a page at a user's request. According to Anthropic, blocking the last two can reduce your visibility in Claude's answers.

Ask ChatGPT your customers' questions, as they would phrase them, with your town or region. Open a new chat for each try, repeat several times, and note the businesses cited and the sources shown. You'll soon see that answers vary: that's why a single try tells you almost nothing.

Questions to test, depending on your business:

• tradesperson: “Which [trade] should I choose in [town] for [type of job]?”;
• online shop: “Where can I buy [product] in the UK?”;
• firm or consultancy: “Which firm should I choose for [need] in [town]?”;
• comparison: “[Your business] or [competitor]: which should I choose?”;
• reputation: “What do you know about [your business]?”, to spot incorrect information.

In your analytics, Google Analytics or Matomo, visits from ChatGPT's links usually carry the tag “utm_source=chatgpt.com”: they show which pages ChatGPT already recommends.

iasyo automates this check: ChatGPT is queried in its real interface, in your language and from your country, with each question asked several times. See the methodology