GEO Ranking: How AI Systems Decide Which Brands They Recommend
Dominik Breitbach founded taismo GmbH and works as lead SEO strategist on SEO retainers and visibility inside AI answers. He has advised companies on search since 2010 and runs taismo as an SEO and GEO agency from Munich.
A GEO ranking is the position a brand holds in the answers of AI systems such as ChatGPT, Gemini, and Perplexity: Who gets named, cited, or recommended when someone asks a question like “Which provider is right for me?” Unlike Google, there is no public list of ten results, just one generated answer that usually names two to five brands. This guide shows which sources AI systems actually pull those recommendations from, how that differs from a classic Google ranking, and which seven actions improve your own GEO ranking.
👉 Want to know how visible your company already is in AI answers today? Visit our GEO services page
What a GEO ranking is
GEO stands for Generative Engine Optimization: The practice of improving your visibility in generative search systems. The term comes from a research paper by Aggarwal et al. (Princeton, Georgia Tech, IIT Delhi) that in 2023 first measured systematically which content generative engines prefer to pull into their answers. A GEO ranking is the outcome of that optimization: How often and how prominently your company appears when AI systems answer questions in your field.
The decisive difference from a Google ranking: There are no positions, only mentions. Where Google shows ten blue links per page and even position 8 still collects clicks, an AI system writes a single answer. If you are not in it, you are invisible for that query. The selection is harsher than any first page of Google, and it follows its own rules. The same discipline also runs under the label LLMO (Large Language Model Optimization), and practitioners increasingly speak of an LLM ranking or AI ranking; all of these describe the same goal of showing up in the answers of large language models.
One caveat about the term itself: In SEO tooling, geo ranking has meant something different for years, namely geographic rank tracking, which checks your Google positions from different locations or map grid points. Both meanings are legitimate, and search results for the term currently mix them. This article covers the meaning that drives the term’s recent growth: Your ranking inside generative AI answers.
How AI answers are built: Training knowledge and live retrieval
To understand how a GEO ranking comes about, you need to know that an AI recommendation can arise in two completely different ways. Each way responds to different signals.
Path 1: Training knowledge (parametric memory). During training, language models learn from enormous text corpora which brands appear in which context. If your brand keeps showing up next to your service in industry articles, ranked lists, and forums, that association gets baked into the model itself. When ChatGPT answers without browsing, it draws on exactly this stored knowledge. This knowledge is slow moving: It only changes with the next model training run, and you can only build it over months through many external mentions.
Path 2: Live retrieval. ChatGPT search, Google’s AI Overviews and AI Mode, and Perplexity fetch current web pages from a search index at answer time, read them, and assemble an answer with source citations. Here it matters whether your content is crawlable, clearly structured, and citable. OpenAI runs a dedicated crawler for this called OAI-SearchBot, which operates independently of the training crawler GPTBot: If you block GPTBot, you stay visible in ChatGPT search; if you block OAI-SearchBot, you drop out of it. Google in turn feeds its AI features from the regular search index and states in its documentation that they require nothing beyond normal search engine optimization. How to work this second path in practice is covered in our guide to ChatGPT SEO.
Where AI systems get their recommendations
When someone asks an AI for a recommendation, the system does not cite at random. In practice, six source types dominate, and together they determine your GEO ranking.
1. Ranked lists and listicles
Questions like “the best CRM tools” or “recommended agencies in Munich” get answered with a striking preference for existing best-of content: Comparison articles, industry rankings, and top 10 posts. The reason is simple: A ranked list is the perfect template for a language model because it already contains exactly the structure the answer is supposed to have. If you appear in several independent lists in your category, you get picked up disproportionately often. If you appear in none, you barely exist for recommendation queries.
2. Review platforms and ratings
Review platforms such as Google reviews, Trustpilot, G2, Capterra, or Clutch give AI systems two things at once: Social proof from real customer voices and machine-readable structure in the form of stars, counts, and categories. For local and service-related queries, review platforms therefore show up regularly among the cited sources. A well-maintained, consistently rated profile is no longer just a reputation topic; it is a direct ranking factor for AI recommendations.
3. Brand mentions: Citations without a link
Brand mentions are references to your brand in other people’s content, even without a hyperlink. Classic link building treated them as second choice; for language models they are gold. A model learns relationships from the text itself, not from the link graph. Every mention of your brand next to terms like your service, your industry, and your region strengthens the association the model later recalls. That is why guest articles, podcast appearances, press coverage, and expert forums pay directly into your GEO ranking, even if not a single backlink comes out of them. What counts is the context of the brand mention: Your name next to your core service helps, a mention without any topical connection evaporates.
4. Structured data and entities
Schema.org markup (JSON-LD) turns a text page into a fact base: Who the organization is, what it offers, who writes here, and how the pages relate to each other. Retrieval systems and the search indexes behind them use this structure to resolve entities cleanly. A brand marked up with structured data for its organization, people, services, and reviews can be cited far more confidently by an AI than an unstructured pile of text. Ambiguity is the silent enemy here: If two similar company names exist on the web, data quality decides which one the AI credits with which properties.
5. Consistent business data (NAP)
NAP stands for name, address, phone: Your company’s core data across directories, portals, and profiles. If spelling, address, or category differ between sources, an AI system sees contradicting facts and, at worst, two half entities instead of one whole one. Consistency across all platforms is therefore the groundwork of every GEO ranking: Unspectacular but measurably effective, because it makes every other source more credible.
6. llms.txt: A briefing for AI crawlers
The llms.txt file is a standard proposed in 2024 by Jeremy Howard (Answer.AI): A Markdown file in the web root that offers AI systems a curated view of a website’s most important pages and facts. To be clear: No major AI system has officially confirmed that it evaluates the file. The effort is minimal though, the format forces you to sharpen your own core statements, and individual crawlers demonstrably already request it. We treat it as a sensible low-cost bet, not as a mandatory signal.
GEO ranking vs. classic Google ranking
GEO does not replace search engine optimization, it extends it. Why we treat GEO as part of modern SEO is laid out in detail in GEO vs. SEO. For day-to-day work, four differences matter most:
- Position vs. mention: Google sorts results, an AI selects. There is no position 7 in a ChatGPT answer; there is named or not named.
- Link graph vs. text context: Google weighs links heavily. Language models additionally learn from plain mentions and their surroundings, link or no link.
- Your page vs. other people’s pages: In a Google ranking, your page ranks. In a GEO ranking, third-party pages often decide about you: Lists, portals, press. A large share of the work therefore happens outside your own website.
- Fast effect vs. slow knowledge: An on-page change can take effect on Google within days. A model’s training knowledge only changes with the next training run; only the retrieval path reacts quickly.
Important for prioritization: The retrieval systems of AI search draw on classic search indexes. If you rank well on Google, you carry a structural advantage into AI Overviews, ChatGPT search, and Perplexity. Solid search engine optimization remains the foundation every GEO ranking stands on.
How to improve your GEO ranking: 7 actions
The Princeton GEO paper compared different optimization tactics experimentally, and its result is remarkably concrete: Adding citations, quotations from relevant sources, and statistics increased visibility in generative answers by up to 40 percent in the paper’s benchmarks, while classic keyword stuffing achieved practically nothing. From those findings and our own project work, seven actions follow, ordered by effort:
- Answer in the first sentence. Build every important page so it answers the core question immediately and citably: Definition first, reasoning second. Our guide on how to get cited by AI shows what that looks like in practice.
- Back up your claims. Numbers with sources, named studies, dated facts. Exactly the elements the GEO paper measured as the most effective tactic.
- Clean up your business data. One pass through all directories and profiles: Same name, same address, same category, same short description.
- Build out your structured data. Organization, people, services, and FAQs as one cleanly connected JSON-LD graph, so your entity is unambiguous.
- Collect reviews systematically. An active profile on the platforms that matter in your industry, with real, current voices.
- Earn brand mentions. Guest articles, expert pieces, podcasts, industry lists: Every context-rich mention feeds the training knowledge of future models. Also check which best-of lists in your category you are missing from, and work on getting included.
- Open your website to AI crawlers. Check your robots.txt for accidental blocks on OAI-SearchBot and friends, keep core content available as crawlable HTML, and add an llms.txt.
For us, GEO is not a separate product but a fixed part of every SEO retainer: Google rankings and mentions in AI answers grow out of the same groundwork. If you want to tackle both systematically, have a look at how our monthly SEO service works.
How to measure your GEO ranking
You cannot read a GEO ranking off a single live query: AI answers vary with phrasing, account context, and the model’s daily form. The measurement only becomes reliable through repetition with a fixed method, which is the core idea behind tracking your AI search visibility. Three building blocks have proven themselves:
- Define a prompt set: 10 to 30 realistic questions from your target group (recommendation questions, comparison questions, problem questions) that you reuse unchanged.
- Measure on a schedule and with tooling: Specialized tools query the major AI systems at fixed intervals and log who gets named and which source gets cited. That produces a time series instead of a snapshot.
- Analyze the cited sources: The cited domains are your target list. If a review platform or a best-of list keeps appearing as a source and you are missing from it, that is the most concrete GEO task there is.
We measure how often AI systems recommend you today with a fixed prompt set, analyze the cited sources, and prioritize the actions with the biggest leverage. No strings attached, and with a clear recommendation on what is worth it for you and what is not.
Frequently asked questions about GEO rankings
What is a GEO ranking?
A GEO ranking describes how often and how prominently a brand appears in the answers of generative AI systems such as ChatGPT, Gemini, or Perplexity. Unlike a Google ranking, there is no list of positions, just the binary question: Does the generated answer name and cite the brand or not?
Is a GEO ranking the same as geographic rank tracking?
No. Geographic rank tracking checks your classic Google positions from different locations or map grid points and is a local SEO method. A GEO ranking in the sense of Generative Engine Optimization measures your presence in the answers of AI systems. The two share an abbreviation but belong to different disciplines.
Can you buy a recommendation in AI answers?
No, there is currently no bookable placement in the organic answers of the major AI systems. The mentions arise from training data and sources retrieved live. Ad formats around AI search are emerging, but they are labeled as ads and do not change whom the model recommends in its answer.
How long does it take for GEO actions to work?
That depends on the path. Actions aimed at live retrieval (citable content, structured data, crawler access) can become visible within weeks once the pages are recrawled. Actions aimed at training knowledge (brand mentions, list placements, reviews) work over months, because they only take full effect with new model versions and refreshed indexes.
Does GEO replace classic SEO?
No. AI search systems draw on classic search indexes, so good Google rankings directly raise your chance of being cited as a source. GEO extends search engine optimization with signals like brand mentions, reviews, and entity maintenance, but it does not replace it.
Sources
- Aggarwal, P., Murahari, V., Rajpurohit, T., Kalyan, A., Narasimhan, K., Deshpande, A.: “GEO: Generative Engine Optimization”, arXiv (preprint, published at KDD 2024), November 2023.
- Google Search Central: “AI features and your website”, developers.google.com, accessed August 2026.
- OpenAI: “OpenAI crawlers (GPTBot, OAI-SearchBot, ChatGPT-User)”, developers.openai.com, accessed August 2026.
- OpenAI: “Introducing ChatGPT search”, openai.com, October 2024.
- Howard, J. (Answer.AI): “The /llms.txt file”, llmstxt.org, September 2024.
