GEO instead of SEO. Getting cited in ChatGPT and Perplexity

Classic SEO targets Google rankings. GEO targets the answer itself. When someone asks ChatGPT who builds AI solutions for small businesses in Lüneburg, I want to be part of the answer. Not third place in a list on page two. This article is my entry into the topic. What GEO is, how I approach it myself and what I watch out for while building this site. More articles on the individual sub-topics will follow.

What sets GEO apart

Search engines deliver links. LLMs deliver answers. Anyone who wants to optimise has to understand how these answers come about. Three mechanisms are decisive for me.

• Training data. Whatever was in the pretraining sits in the model.
• Retrieval Augmented Generation. During live search, current sources are pulled in, often via the Bing or Google index.
• Browsing tools. Perplexity, ChatGPT with Search, Claude. They crawl pages in real time and cite them as a source.

What works for me in 2026

A lot about GEO is still experimental. I keep testing things myself and form an opinion based on what I observe. I currently consider the following levers the most important.

Clear statements instead of marketing fluff

LLMs love sentences that make a concrete statement. Whoever writes "We offer innovative solutions" gets ignored. Whoever writes "I build AI chatbots for Twitch streamers with their own voice and live web search" gets cited.

Structured data

Schema.org is not just for Google. LLMs also use structured data as a hint for what a page is and what it talks about. FAQ schema, Person schema, Article schema. It all helps.

Quotable content

Write so that a paragraph from your text works verbatim inside a ChatGPT answer. Short, self-contained statements. No pure filler sentences without substance.

Be technically clean

Before any LLM cites your page, it has to be able to crawl it. Sounds trivial, but it is not. Cloudflare blocks AI crawlers in its default configuration. Anyone using Cloudflare who has changed nothing may be invisible to GPTBot, ClaudeBot and PerplexityBot.

For the robots.txt a simple userAgent: '*' with allow: '/' is enough. That allows all bots including AI crawlers. No reason to list each one individually. robots.txt is not a technical blocking tool, it is a signal to cooperative crawlers.

llms.txt: nice, but no silver bullet

Many GEO guides sell llms.txt as a big lever. My impression is more sober. So far I have found no hard evidence that the major LLM providers really use it for answers. I still have it on this site, because it takes almost no effort and has a clear use case. AI agents and developers who want to use my services programmatically can see at a glance what I offer. View my llms-full.txt.

Beyond your own site

Anyone who wants to be found locally, meaning when someone asks ChatGPT "Who builds AI solutions for small businesses in Lüneburg?", has to understand where ChatGPT gets that information. It is not primarily Google.

Local presence beyond Google

ChatGPT has no direct integration with Google Business Profile. For local recommendations it pulls primarily from the Bing index and from Bing Places for Business. On top of that come aggregated local business data from map and POI databases, review platforms and booking portals. Google Business Profile stays important, but more for Google AI Overviews and Gemini. So if you only maintain your Google listing, you are blind to ChatGPT at first.

NAP consistency

Name, address and phone number have to be identical across all platforms and your own website. Not similar, identical. If the website states "Lüneburg" as the location but the Bing profile only the postcode, the model registers a contradiction and moves on. AI bots do not read to get to know you. They read to verify what other sources already say about you.

Reddit is more relevant than you think

Reddit shows up suspiciously often as a source in LLM answers. From what I observe it is among the most frequently cited platforms, often even ahead of Wikipedia. The reason is fairly simple. OpenAI and Google have licensing deals with Reddit for training data, and real discussions between real people are more valuable to models than any marketing page.

Self-promotion in subreddits does not work. It gets downvoted and tends to hurt. What works is answering real expert questions honestly, sharing your own experience and only pointing to your own resources when explicitly asked.

What no longer works

Keyword stuffing, backlink farms, thin content with a pure search-volume focus. LLMs judge content differently from Google PageRank. Backlinks lose direct importance in the GEO context. What becomes decisive is whether content is clearly quotable and semantically unambiguous.

My conclusion

Write for the reader, structure for the machine. Be technically reachable, be present in the relevant directories, and be visible where real discussions happen. Whoever establishes this now will be hard to catch up with in two years.

This article is meant as an overview on purpose. For each of the topics only touched on here, I am planning my own deeper articles. Specifically on llms.txt, on building a good robots.txt, on local directories and on Reddit as a GEO lever. Anyone who comes back here will find them gradually in the GEO & SEO category.