Before a first call with a potential vendor or partner, plenty of people no longer stop at a search engine. They ask ChatGPT, Perplexity or Gemini directly: What does this company do? What is it known for? Who does it compete with? The answer arrives in seconds and sounds remarkably confident.
How widely people use AI assistants to research companies is difficult to measure reliably. For communications teams, the more important change is the possibility itself: someone can form a first impression of a business without ever visiting its website. That makes it worth examining which information surfaces and how it is presented.
A website is only one source
Depending on the provider, AI systems draw on training data, a live web search, or both. How much weight each source gets isn't something providers disclose in detail. A company's own website is rarely the whole story. Trade press coverage, interviews, business media, association pages, platform profiles and old press releases all feed the public record.
That becomes a problem when those sources don't line up. If a company has moved from reselling hardware to selling software while much of the information available online still describes the old business, an AI answer may paint an outdated picture. Systems can also make mistakes, omit context, or draw unsupported conclusions. Answers vary with the tool, the wording, and the timing. They cannot be fully controlled.
What GEO can and can't do
This is the territory of generative engine optimization, or GEO. The term was coined in a 2023 research paper by researchers from Princeton University, IIT Delhi and independent collaborators. SEO is about ranking high in a list of links. GEO is about showing up in AI-generated answers and being described accurately there.
GEO cannot guarantee a particular result, and it is not a bag of tricks. It makes most sense as part of communications that are public, credible, and consistent. That puts PR in the picture: media relations can help establish verifiable information about a company and contribute to how its expertise is discussed in the trade press. A stronger public information base is valuable, even though it cannot dictate what an AI system will say.
A first check doesn't take much: ask several AI systems the questions customers, candidates or reporters would ask, repeat them over a few weeks, and note what's outdated or contradictory. It isn't a measurement, but it often shows where the public record has gaps.
In the end, asking what AI says about a company is an old question in a new form: what does the public know about us, and is it still true?