The Question Isn't Whether AI Sees You. It's What AI Thinks You Are

Ask an AI system what a company does, and within seconds you will usually get an answer that sounds remarkably certain. It may tell you what industry the company operates in, give you a brief history, describe its principal services and name a few competitors. The answer may even contain the kind of specific details that make it feel researched rather than generated.

Now imagine that the answer is wrong.

Not spectacularly wrong. The company has not been confused with a hospital or a software company or a restaurant. Its name is correct, its broad description is plausible and most of the details may even be accurate. But the company has been placed in the wrong category. An old service is described as though it were still offered. A piece of work belonging to another company has somehow been attributed to it. An outdated criticism is repeated without the context that once accompanied it. Two businesses with similar names have been blended into one.

Someone who works at the company would spot the problem almost immediately. Someone who has never heard of the company would have no reason to do so.

That distinction matters because it points to a problem in the way we are beginning to think about Generative Engine Optimization, or GEO. Much of the discussion so far has focused on a relatively simple ambition: getting artificial intelligence systems to mention a brand. Companies want to know how to appear in ChatGPT, Gemini, Claude or Perplexity when someone asks a relevant question.

But being mentioned is not the same as being understood. The more important question may be what the machine believes the company actually is, and what evidence led it to that conclusion.

The Mechanics of Interpretation

For much of the internet era, digital marketing was built around the problem of being found. Search engines made it possible for someone looking for a product, service or company to discover relevant pages. Businesses responded by improving their websites, building links, publishing content and trying to make themselves more visible in search results. The search engine returned a set of possibilities, and the user did the rest. They opened pages, compared information and eventually formed an opinion.

Social media changed the mechanics of discovery, but not the basic relationship between information and interpretation. A person might encounter a company through a recommendation from a creator, a post in a feed or an advertisement, but there was still a person on the receiving end deciding what that information meant.

Generative AI introduces a different intermediary.

When someone asks a generative system a question about a company, the system often does not simply return a collection of documents. It constructs an explanation. To do that, it has to determine what the company is, connect information about it from different places, weigh apparently conflicting accounts and decide which details belong in the final answer.

The machine is no longer merely helping someone find information. It is helping them interpret it. That creates a different kind of reputational risk.

The Consequence of the Quiet Error

The obvious AI error is easy to recognise. If a system invents a company's founder or claims that it operates in an industry in which it has never worked, the mistake is conspicuous. A reader may question the answer or search for another source. The more consequential error can be much quieter. It is the answer that is approximately right but subtly misleading.

Imagine a company that has spent years repositioning itself from a conventional digital marketing agency into a strategic communications consultancy. Its own website reflects that change. Its leadership talks about it. Its recent work reflects it. But much of the company's older online footprint still describes it as a digital agency. Perhaps industry directories use that language. Perhaps old articles use it. Perhaps several third-party websites copied the description from one another years ago.

No single source is necessarily wrong. Yet the cumulative impression is.

A generative system encountering all of this information has to make a judgment. It may decide that the company's most defensible description is the one that appears most consistently across the available evidence. The result could be a perfectly fluent answer that places the company in a category it has spent years trying to leave behind.

That is not a trivial branding problem.

A prospective client who asks an AI system about the company may never know that the description is outdated. The client may assume that the company is simply another digital marketing agency and evaluate it accordingly. They may compare it with the wrong competitors, ask it the wrong questions or decide that its capabilities are not relevant to their needs.

The company has not become less capable. It has become less accurately represented.

The Two Identities: Designed vs. Inferred

This is one reason it may be useful to think of a brand as having two identities.

The first is the identity the organisation deliberately creates. It is expressed through positioning, advertising, corporate communications, websites, presentations and the language the company uses to describe itself. The second is the identity that can be reconstructed from the wider internet.

That version includes everything the company controls and everything it does not: news coverage, directories, reviews, interviews, government records, old websites, third-party articles, industry listings, competitor comparisons and countless pieces of information that may have been published without the organisation ever paying much attention to them.

The first identity is designed.

The second is inferred.

And generative AI systems are becoming increasingly good at making that inference.

Why GEO Is More Than Tactics

This is where the current conversation around GEO can become too tactical. Much of the advice being offered to companies is about creating more content, answering more questions, adding structured data and making information easier for AI systems to retrieve. Those practices may be useful. But they do not necessarily solve the deeper problem.

A company can make itself easier to find without making itself easier to understand.

It can publish hundreds of pages and still leave behind a contradictory digital record. It can describe itself consistently on its own website while dozens of other sources describe it differently. It can make a claim about its expertise without having enough independent evidence online to support that claim.

Generative systems do not have to accept the company's preferred description simply because the company wrote it.

They have to make sense of the evidence available to them.

That changes the nature of the task. GEO is not only about visibility. It is also about the structure, consistency, credibility and freshness of the information from which an organisation's identity can be reconstructed.

Invisibility vs. Misrepresentation

There is another reason this matters: the difference between being invisible and being misrepresented.

If an AI system does not mention your company, you have a problem that is relatively easy to recognise. You can test for it. You can measure how often the company appears in relevant answers. You can identify the questions for which it is absent and work on improving its visibility.

Misrepresentation is harder to see.

A company may not know that an AI system has been describing it inaccurately. There is rarely someone inside an organisation whose job it is to ask, every week, what different AI systems believe the company does, who they think its competitors are, what they think it is best known for or which criticisms they associate with it.

The absence of an answer is visible.

A misleading answer can remain invisible precisely because it sounds reasonable.

Machines are increasingly becoming part of the way people understand organisations. The stakes are not simply about traffic. They are about representation at the moment a person is forming an opinion.

A potential customer may ask an AI system about a company before visiting its website. A journalist may ask for a quick background briefing before making contact. An investor may ask for an overview. A procurement team may use AI to create an initial shortlist. A prospective employee may ask what the organisation is known for.

In each case, the answer may become the starting frame through which the organisation is understood. And once that frame has been established, everything that comes afterwards is interpreted through it.

Companies have spent decades becoming deliberate about how they present themselves to the world. They have built brands, written positioning statements, managed public relations, refined websites and agonised over individual words on a homepage.

There is now another participant in that process.

It does not ask the company who it is. It looks at the evidence that exists and tries to work it out. Sometimes it will get the answer right. Sometimes it will not.

The challenge for companies is therefore not simply to make themselves visible to machines. It is to make the available evidence coherent enough that when a machine tries to explain who they are, the most accurate explanation is also the easiest one to reach.

That may be the real work of GEO.

Not teaching AI to say your name.

Making sure that when it does, it knows who it is talking about.