(Originally published on LinkedIn)

Imagine a potential customer asking Perplexity about the best suppliers in your industry. Is your name on the list? Have you been mentioned?

It's a clear problem that affects which purchasing decisions are made right now. As more customers turn to AI instead of Googling, it's the AI's answer that determines which suppliers end up on the list. Not being visible there can be even worse than being on Google's page two since the AI answer often only mentions two or three names.

GEO (Generative Engine Optimization) is the working term for what's required to be visible in this context. This discipline is roughly where SEO was in 2008: many talk about it but few know how to do it correctly.

Here's what we've learned from building RAG-based chatbots while also trying to make our own site citable.

What GEO Actually Measures

SEO is about improving your ranking and increasing click-through rate. A high ranking on Google attracts more visitors to your site. Tools for this include the right keywords, good links, fast page loading, and content that matches the user's search intent.

GEO helps measure citations and references. When you ask a question to ChatGPT or Perplexity, you often get answers that are linked to one or more sources, which adds credibility. GEO tries to be one of them. The tools are structured data, source-referenced text, authentic authorship, and clear fact representation.

Both remain simultaneously. SEO helps those searching on Google, while GEO is for those asking questions to AI. More and more people use AI as their primary search tool each quarter. Gartner predicted that traffic to traditional search engines will decrease by 25 percent by 2026.

GEO is not a replacement for SEO, but a valuable addition that becomes even more important as more customer journeys begin in a chat instead of in the search box. The term comes from a 2024 Princeton study, which showed that targeted text optimizations can improve a website's visibility in generative AI responses by up to 40 percent.

How AI Engines Choose Sources

To truly understand GEO, it's good to learn how it works, since that's often where many mistakes can happen. Understanding this can make a big difference in how you use it.

When a user asks a question, it's converted into a vector which helps the model find similar examples in its training data and recently indexed web pages to provide an accurate answer. The best matches are used as context for the answer. If your site is well-structured, correctly embedded, and seen as a reliable source, the chances increase that it will end up in the context window, making it much easier for the user to find the right information.

Four signals influence the choice of source.

Vector similarity. How well your text semantically matches the question. The text should be formatted so that it can be divided into meaningful parts. Long abstract paragraphs lose against clear, declarative answers.

Authority. What other established sources link to and cite. A link from the IMY website or Wikipedia carries more weight than a hundred links from unknown blogs.

Recency. Date is important. AI engines prioritize fresh sources when information may have changed. Clear publication and update dates in HTML and in the text facilitate this.

Structured data. JSON-LD, FAQ schema, and semantic HTML make it easier for AI to extract facts accurately without interpreting the layout. When you publish pricing information or opening hours as structured data, you reduce the risk of AI misinterpreting it.

Another important mechanic many overlook is that AI crawlers must be able to reach your website. If your robots.txt blocks GPTBot, ClaudeBot, or the Perplexity crawler, you're invisible to them regardless of content quality. It's a deliberate decision. OpenAI's documentation lists three user-agents (GPTBot, OAI-SearchBot, ChatGPT-User) that you can allow or block separately depending on purpose.

Seven Concrete GEO Techniques

1. Direct answers in the first sentence. "Our platform takes fifteen business days to integrate" beats "When it comes to integrations, the time depends on several factors" every time. AI cites the declarative sentence. Write as if every heading is a question and the first paragraph is the answer.

2. Specific numbers with sources. "62 percent of a work day is spent on repetitive tasks (Asana 2023)" is citation-friendly. "Many hours are lost" is not. Source-referenced numbers make the text self-validating.

3. Q&A-structured copy. Use <h2> questions and clear answers under each question. It matches how people ask AI. Add FAQPage schema in JSON-LD so the AI engine gets an explicit question-answer mapping without having to guess.

4. Clear authorship. Use bylines with real names and link to the author's LinkedIn profile. AI engines prefer sources with clear accountability. An article signed "admin" or without a name is considered less credible than the same text with a named author.

5. Brand mentions in canonical sources. When the AI engine searches for your category, it's good if your name appears on pages it already trusts like Wikipedia or recognized industry publications. It's more about building trust than using SEO tricks.

6. llms.txt. A markdown-based standard (llmstxt.org) proposed by Jeremy Howard in September 2024 and the equivalent of robots.txt but for LLMs. A simple file in your site root where you point AI engines to your best content and briefly describe your business. Anthropic, Stripe, Cloudflare, Vercel, and Next.js publish llms.txt for their documentation today. Mintlify-hosted documents have had the file auto-generated since November 2024.

7. Embeddings-friendly chunking. Short paragraphs of three to five sentences with clear headings. This matches how RAG pipelines divide text. Long text passages give worse matches because the relevant parts drown.

What Works Against GEO

Certain practices that worked on Google during the 2010s are now penalized.

Keyword stuffing is still common, but AI engines detect it. A page that repeats "AI for small businesses" twenty times won't rank better and instead gets a lower quality score.

AI-generated copy that sounds like AI is the problem many underestimate. Em-dashes, parallel three-lists, "X, not Y" constructions, and "seamless" intensifiers are common signs. Modern LLMs can recognize their own output and often cite authentic voices. While there's no official standard for AI detection, a clear trend shows that text that appears to be generated is often considered less credible when cited.

Made-up statistics spread until someone verifies them. When no source can be verified, AI loses trust in the entire page. Numbers with clear sources are cited, while uncertain claims are often ignored or flagged.

Content without context. Short blog posts without substance get no citations. Long, fluffy texts without concrete information don't work either. What works instead is medium-length texts that provide clear answers to clear questions.

How You Measure

Manual question tests remain the simplest method. Ask ChatGPT, Perplexity, Claude, and Gemini about your industry every week. Note whether you're mentioned, where you land in the list, and in what context.

The tool landscape is changing rapidly. Otterly.ai, Profound, Athena HQ, and Peec.ai offer different types of brand monitoring for AI responses. Most are still in early stages and pricing varies. Check what's available at the time of publication.

UTM parameters on links that AI engines find in your text allow Google Analytics to identify traffic as AI-driven. It's not entirely perfect since many read the AI response and search your name separately, but it gives an overview.

Indirect measurement through brand monitoring on LinkedIn and news sites is also important. AI systems scan these platforms and form an impression of your authority based on what others say.

Conclusion

GEO is about being the obvious source when AI needs to answer questions in your field. It's not possible to trick the system. It requires that you write with substance and have a clear author who stands behind the content.

Do you want to do a GEO audit on your own site? Send me a DM or email mats@gothiaai.se and we'll talk.