How do I get cited by AI assistants like ChatGPT, Claude and Perplexity?

In brief: getting cited by AI assistants

Getting cited by AI assistants means that systems like ChatGPT, Claude, Perplexity or Google Gemini use your website as a source for an answer – often with a visible link or a named mention. The discipline behind this is called Generative Engine Optimization (GEO), sometimes Answer Engine Optimization (AEO). It ensures your content not only ranks but is recognised by generative models as trustworthy, verbatim-citable fact.

GEO does not replace classic SEO – it builds on it. If you are already visible in Google search results and Google AI Overviews, you are best positioned to be cited in pure chat answers too. The difference: AI assistants reward clarity, structure and verifiable facts more than raw keyword density.

• SEO targets rankings in a results list; GEO targets a mention inside a generated answer.
• SEO optimises for clicks; GEO optimises for a model lifting your sentence and naming you as the source.
• Both need good content – GEO additionally needs clear answer blocks, entities and machine-readable signals.

How ChatGPT, Claude, Perplexity and Gemini choose sources

Generative assistants ground answers in two ways. First, in knowledge learned during training. Second – and decisive for citations – in live retrieval of web content via Retrieval-Augmented Generation (RAG): the model searches in real time, reads the best hits and summarises them. Perplexity and Google Gemini show visible source links; ChatGPT with search and Claude with web search do the same.

What gets cited tends to be what is easy to extract and easy to verify: a clearly written paragraph that directly answers a specific question, with unambiguous terms and a traceable source. Models prefer passages that can stand as a self-contained statement without having read the whole article.

• A precise, self-contained answer in the first paragraph of a section.
• Consistent naming of people, places, products and concepts (entities).
• Visible publication and update dates plus a named author.
• Machine-readable structure: clean headings, lists, tables and schema.org markup.
• Technical accessibility: the page is readable by AI crawlers and loads without JavaScript barriers.

Answer-first content an AI can lift verbatim

The most effective GEO lever is answer-first writing. Answer the core question in the first 40–75 words of a section before you explain, expand or qualify. That compact passage is exactly the building block a model extracts and attributes to you.

Phrase definitions as complete, context-free sentences. Instead of "It depends …" write: "An llms.txt is a text file in a website's root directory that gives AI systems a guide to its most important content." Such sentences are citable because they make sense without surrounding context.

• Start every section with the core statement, then give evidence and examples.
• Write a 45–75-word summary at the very top that an AI can lift verbatim.
• Use the actual user question as a plain-text heading (H2/H3).
• Prefer short, fact-rich sentences over long nested ones.
• Add an FAQ block with real questions and concise, self-contained answers.

Entities and consistency: the language machines understand

AI models think in entities – clearly identifiable things like people, companies, places, products and concepts – and in the relationships between them. Naming an entity the same way every time and linking it to established references makes it easy for the model to assign your content to a topic and rate you as a competent source.

In practice: define the central entities per topic and use the same labels throughout. Link to authoritative references (for example the official page of a standard or authority) and build internal links between thematically related articles – this creates a recognisable topic cluster.

• Define key entities and write their names identically every time (no synonym-switching mid-text).
• Link entities to recognised references and to your own "About" or author page.
• Build topic clusters: a pillar article plus deep-dive spokes, cross-linked.
• Name relevant technical terms explicitly instead of paraphrasing – models weight exact terminology.

Structured data, llms.txt and technical signals

Structured data using schema.org (as JSON-LD) translates your content into a format machines read unambiguously: what is an article, who is the author, when was it updated, which question does an FAQ entry answer. Suitable types include Article, FAQPage, HowTo, Organization and Person. This increases the chance your answer is interpreted and attributed correctly.

In addition, llms.txt is emerging – a simple Markdown file in the root directory (yourdomain.ch/llms.txt) that gives AI systems a curated guide to your most important content. The standard is young and not evaluated by every provider, but it costs little and signals structure. More important remains: AI crawlers are allowed to fetch your pages at all (check robots.txt) and the core content sits in the HTML, not only after JavaScript runs.

• Use JSON-LD for Article/BlogPosting, FAQPage, HowTo, Organization and Person.
• Keep title, visible text and structured data consistent with each other.
• Provide an llms.txt with links and short descriptions of your most important pages.
• Check robots.txt: AI crawlers (such as the bots used for web search) should be allowed to read your content pages.
• Ensure the main text is delivered server-side in the HTML.

Trust and source transparency – the Swiss advantage

Generative models weight signals of experience, expertise, authoritativeness and trustworthiness (E-E-A-T). They are more likely to cite sources that identify themselves: with a named author and their qualification, publication and update dates, clear contact details and – where useful – references to primary sources. Transparency is therefore not a nice-to-have but a citation factor.

For Swiss SMEs this is an advantage. Referencing the Swiss context correctly – the revFADP (nFADP) rather than the GDPR, the FDPIC (EDÖB) as supervisory authority, prices in CHF, cantonal specifics – delivers precision that generic content lacks. This local accuracy makes you the preferred source for Switzerland-related questions where international websites stay vague.

• Name a real author with a short bio and professional qualification.
• Show "published on" and "updated on" visibly.
• Link primary sources and authorities instead of vague claims.
• Make your Swiss focus explicit: revFADP, FDPIC, CHF, cantonal rules.
• Avoid unsupported numbers – models and readers penalise unverifiable claims.

Done in seven steps (for Swiss SMEs)

GEO works as a repeatable process, not a one-off action. The following order prioritises the biggest levers first and can be implemented with existing resources.

• 1. Collect user questions: use real questions from sales, support and search as headings.
• 2. Write answer-first: open every page with a 45–75-word summary.
• 3. Fix entities: define central terms and name them consistently.
• 4. Add structured data: implement JSON-LD for Article, FAQPage and Organization.
• 5. Show trust: make author, dates, contact and primary sources visible.
• 6. Check the tech: crawler access, HTML delivery and optionally llms.txt.
• 7. Measure and refine: test whether ChatGPT, Perplexity and Gemini name you, and close the gaps.
• If you would rather not set this process up yourself, you can have it accompanied – the Weissmann AI Academy combines SEO, GEO and AI expertise with Swiss data-protection practice.