How to appear in ChatGPT

ChatGPT blends two information sources: what it learned during training (with a cutoff date) and, when it browses live via its crawler or Bing (Search mode), what it retrieves in the moment. When a user query triggers real-time search, the model reads several pages, synthesises them and produces an answer with citations.

Cited sources are not picked at random: the model favours pages with clear answers, verifiable data, clean semantic structure and recognised topical authority. If your URL meets those criteria and ranks well in Bing, citation probability shoots up.

OpenAI uses different user-agents to crawl the web. GPTBot crawls content to train future models, OAI-SearchBot indexes for the ChatGPT Search feature and ChatGPT-User accesses pages requested by users during a conversation. Blocking any of them in robots.txt is the fastest way to disappear.

If you want to appear in answers, check that your robots.txt isn't excluding these bots. This is especially common in companies that activated blanket blocks for "all AI crawlers" without understanding the consequence: total invisibility in ChatGPT.

Generative models reward content that solves the user's question in the first sentences. A page that hides the answer behind 800 words of intro is less attractive to cite than one opening with a clear, verifiable synthesis paragraph.

Apply a TL;DR pattern at the beginning of every important piece: a direct 2-3 sentence answer the model can copy almost verbatim. Then develop the topic in depth. This doesn't hurt classic SEO (it actually improves dwell time) and dramatically increases GEO citation chances.

LLMs learn and validate information by cross-referencing multiple sources. A brand only present on its own website has little credibility for the model: it needs external corroboration. That's why presence in Wikipedia, Wikidata, industry aggregators and reference media weighs so much.

If your brand still has no Wikipedia entry or structured Wikidata profile, that is probably the highest-ROI GEO investment. It isn't always easy (Wikipedia has strict notability rules), but digital PR aimed at citable sources helps enormously.

Schema.org isn't just for Google. LLM crawlers also read JSON-LD markup and use it to understand what your page is, which entity it represents and which relationships it has. A correct implementation of Organization, Product, Article, FAQPage and BreadcrumbList helps the model categorise you unambiguously.

Especially useful for GEO is the "sameAs" property inside Organization, linking your brand to its Wikipedia entry, LinkedIn, Crunchbase and other authority nodes. This reinforces entity modelling and reduces identification errors.

Without measurement there is no strategy. The first operational step is to build a list of 30-50 prompts representative of your industry: questions your customers would ask ChatGPT before or during the buying process. Run them manually or with tools like Otterly, Profound or AthenaHQ and log whether your brand appears, in what position and with what accuracy.

Repeat the process monthly to detect changes, identify prompts where you're gaining or losing ground and understand which content actions are working. Without this loop, any GEO action runs blind.

Three errors keep repeating: blocking OpenAI crawlers in robots.txt thinking it protects the brand, relying only on your own website without reinforcing presence in external sources and publishing generic content with no proprietary data or verifiable numbers. Any of the three is enough to disappear.

A fourth, subtler mistake is not measuring. Without periodic tracking it is impossible to know whether actions taken are moving the needle. ChatGPT shifts answers week to week and the only way to manage visibility is continuous monitoring.