GEO Forensics: The Invisible Layer of AI Brand Visibility

For two decades, marketers optimized for search engines. Keywords, backlinks and snippets decided who got found and who disappeared. But discovery is now shifting to a different layer. When someone asks ChatGPT, Claude or Gemini which brand they can trust, the model does not simply pull up Google’s index. It draws on its memory: a compressed reflection of everything it learned from years of web data, news, reviews and conversations.

That shift confronts every brand with a new, fundamental question: does the model even remember you? That is no small detail. More and more often, the buying journey does not end on a results page, but in an answer the assistant formulates itself. If your brand is missing from that answer, you simply do not exist for that buyer, no matter how much you invest in advertising or classic visibility.

Check your own page: see how ready your page is to be cited by AI with our free GEO check.

From SEO to GEO

SEO was about visibility in searches. Generative Engine Optimization (GEO) is about visibility in conversations. Where SEO trains algorithms to rank, GEO trains language models to remember your brand and weave it naturally into their reasoning. If you want to understand how this discipline works, read our complete guide on what GEO is and how it works in 2026.

Classic analytics tell you about traffic, clicks and rankings. They say nothing about the question that really matters now: is your brand anchored in the model’s memory, that invisible layer which steers answers, recommendations and perceptions? This is where GEO Forensics™ comes in.

What is GEO Forensics™?

GEO Forensics™ is the first structured method to audit a brand’s presence inside large language models. It is not about prompt tricks or one-off searches. It is a forensic framework: a repeatable research process that makes your brand memory measurable instead of guessing at it.

The approach is deliberately forensic. A single prompt in ChatGPT gives you a snapshot that can already sound different the next day. GEO Forensics therefore asks the same thing multiple times, from different angles and across different models, building a pattern that is actually reliable. You are not measuring what the model happened to say once, but what it structurally knows about you.

What GEO Forensics measures

The method examines your brand along five dimensions. Together they form the picture a language model has of you.

1. 🧭 Brand Recall: Does the model remember your brand internally, or does it have to retrieve you via the web? The difference is crucial. What sits in the model’s weights surfaces spontaneously; what only comes in through retrieval depends on which source happens to be found at that moment.
2. 🔍 Context and positioning: How does the model describe you? Which tone, adjectives and associations are attached to your name? Are you cast as a market leader, a niche player, or categorized completely wrong?
3. ⚔️ Competitive visibility: Who else appears when a user asks about solutions like yours? Your real competitor in the AI layer is not always your competitor in the market.
4. 🧠 Source and retrieval diagnostics: Where does your brand data come from? Wikipedia, trade media, review sites, your own blog? This shows you which sources the model truly trusts, and therefore where your influence is greatest.
5. 📊 Sentiment and topic mapping: Which emotional and thematic patterns does the model link to your name? Are you associated with reliability, with innovation, or with unresolved complaints?

The steps of a GEO Forensics audit

An audit runs in a fixed order, so that the outcome is repeatable and comparable, including over time.

Step 1, set the scope. You define which brand, which product category and which buyer context you are investigating. “Best fleet management software” is a different question than “reliable supplier for mobile teams”, and the model answers each question differently.

Step 2, run the prompts. You compile a set of research prompts that cover the five dimensions, and run them consistently across the major models. By varying the phrasing, you surface the difference between genuine memory and chance hits.

Step 3, code the answers. Each answer is scored: are you mentioned, in what position, with what description, and with which source? This turns free text into data you can add up.

Step 4, score and visualize. The coded answers flow together into the core outputs (see below), so a pattern becomes visible that a single conversation can never show.

Step 5, repeat and benchmark. Because models keep learning, a single measurement is a photo and not a film. By repeating the audit periodically, you see whether your work is having an effect and whether competitors are gaining ground.

Important: treat every step as research, not as marketing. The value of a forensic approach lies in the discipline. Use the same prompts, the same scoring rules and the same models on every round, so that differences between measurements reflect real shifts and not noise. Just as you would not change your measurement method every month during an SEO audit, here you stick to a fixed methodology. Only then does your Brand Memory Report become an instrument you dare to base decisions on, rather than a collection of loose observations.

The core outputs

Every GEO Forensics test produces four results that together form one report:

• A Brand Recall Index (0 to 100) that shows how often and how accurately the model remembers you.
• A Visibility Heatmap that sets your presence against that of competitors.
• A Source Breakdown that reveals which publications feed your AI footprint.
• A Sentiment Graph that tracks how positively or negatively you are described.

Together they form a Brand Memory Report: the modern equivalent of an SEO audit, reimagined for language models. The difference with a vanity figure is that each of these outputs points to a concrete action. A low Recall Index calls for more and better sources. A skewed Source Breakdown tells you exactly where your PR and content efforts pay off most.

Why brand memory is market share

When a user asks “What is the best app for fleet management?” or “Which platform helps manage mobile teams?”, the model draws on what it remembers to formulate an answer. If your brand is not in that memory layer, you do not appear, regardless of your advertising budget or your classic SEO.

That is the new frontier of brand value: being remembered by the model, not just by the market. In a world where AI assistants steer buying decisions, memory is market share. And that is precisely why GEO Forensics is not an academic exercise but a commercial measurement instrument. It does not measure whether you score well on a vain list, but whether you are present at the moment a buyer makes a choice.

What this means for marketers

The insights from a Brand Memory Report translate directly into strategy in an AI-mediated world:

• PR teams can deliberately feed the data sources models rely on, instead of guessing where coverage counts.
• Content strategists can write for recall instead of only for clicks, with clear, citable phrasing that a model easily picks up.
• Leadership teams can measure their influence within exactly the systems their customers consult.

For those who lead their sector, GEO Forensics becomes the competitive advantage that keeps you anchored in the model’s long-term memory. And because that memory changes slowly, whoever invests now builds a lead that competitors will find hard to close later. The work you put into citable sources today keeps paying off every time a model is retrained.

From research to practice

GEO Forensics did not start as a marketing gimmick, but to help brands win back their visibility in a world no longer mediated solely by search engines. Since then, marketers, founders and agencies have started running LLM visibility audits and discover, often for the first time, whether their brand exists at all in the AI consciousness. The confronting side is that many brands that are strong in the market turn out to be barely present in the models. The encouraging side: once you know how a model sees you, you can teach it to see you better.

In the coming years, AI brand optimization will grow from experiment into a mature discipline. GEO Forensics™ is the diagnostic layer beneath it: the starting point that tells you where you stand before you invest. At Customer Impact, the agency behind the book The End of Search, we build this measurement into our generative engine optimization (GEO), so that you do not optimize on gut feeling but on data.

De Keyzer, T. (2025). GEO Forensics™: A Framework for Measuring Brand Memory in Large Language Models. Customer Impact Research Publication, 2025.

Want to know whether the models remember your brand and what you can do to steer that memory? Schedule a conversation with our team and we will walk through your first Brand Memory Report together. Want to test it yourself already? Then download the GEO Forensics Toolkit and run your first audit.

Further reading

• What is a foundation model? An explanation for B2B marketers