A founder asked me last month whether their company showed up in ChatGPT when someone searched for their category. They had no idea. Neither did their head of marketing. Neither did their SEO agency.
That gap is now a real business problem.
LLM-powered surfaces like ChatGPT, Perplexity, Google's AI Overviews, and Bing Copilot are not a future scenario. They are where a growing share of discovery happens today. And most brands are flying blind inside them.
Classic SEO is about ranking. You target a keyword, you publish content, you earn links, you measure position. The feedback loop is tight. Tools are mature. The logic is legible.
LLM search doesn't work like that. There is no rank. There is no position one. The model pulls from its training data and from live retrieval, weighs it against the prompt, and synthesizes an answer. Your brand either appears in that synthesis or it doesn't.
The criteria that make you appear are not fully published by anyone. But they are not random either. Models tend to surface brands that are mentioned frequently across credible, varied sources. Brands whose descriptions are consistent. Brands whose category language matches what the model has learned to associate with a given problem.
If your content is thin, inconsistent, or written primarily for keyword density rather than genuine explanation, you are likely invisible. And you won't find out from a rank tracker.
Your brand either appears in the synthesis or it doesn't. There is no position two to fall back on.
The good news is that you can probe this yourself, right now, without a specialist tool. The bad news is that most teams don't have a consistent method for doing it, so the results are noisy and hard to act on.
Here is the basic structure we use with clients.
Pick the prompts that match real buying intent. Not your brand name. Prompts like "what's a good agency for X", "how do I choose a tool for Y", "who are the main players in Z". These are the surfaces where you either exist or you don't.
Run the same prompts across multiple models. ChatGPT, Perplexity, Gemini, Bing Copilot. Each model has a different training corpus and a different retrieval layer. Your visibility varies across them. Treating them as one surface gives you a false read.
Document the output verbatim. Not a screenshot. Copy the text. You need to be able to grep it for your brand name, your competitors' names, and the category language the model uses. That language is a signal worth reading carefully.
Vary the prompt framing. A model might mention you when the prompt is broad and miss you when the prompt is specific. Understanding the shape of your visibility tells you where the gap is.
Repeat on a cadence. Models update. Retrieval layers change. A snapshot from three months ago is probably stale.
After running this kind of audit across a range of clients in agency, SaaS, and professional services, some patterns hold up.
Brands that appear consistently tend to have one thing in common: their content explains things. Not at a surface level. They have published material that a model can draw on to understand what they do, who they serve, and why someone would choose them. Detailed case studies. Specific methodology descriptions. Named frameworks explained in plain language.
Brands that are invisible tend to publish content that performs for humans scanning a page but doesn't give a model much to work with. Bullet lists of benefits. Vague service descriptions. A blog that covers trending topics without ever going deep on their own point of view.
There is also a signal strength problem. If your brand is only mentioned on your own domain, models treat that as weak evidence. Mentions in third-party editorial, in niche publications, in community discussions, in podcast transcripts, these build the kind of distributed signal that models weight more heavily. This is not new as a principle. It rhymes with link-building logic. But the mechanism is different, and the content types that generate it are different.
Bullet lists of benefits don't give a model much to work with. Detailed explanation does.
This is not an argument for abandoning SEO. Organic search still drives volume. But LLM surfaces are no longer edge cases. For high-consideration purchases, for B2B buying research, for category education, they are increasingly where the first shortlist gets formed. If you're not on that shortlist, you may not get a chance to pitch.
A few things worth doing in the next 90 days.
Audit your visibility the way I described above. Do it across four models, across ten to fifteen real buying-intent prompts. Treat it as a baseline. You need to know where you stand before you can move.
Read what the models say about your category. The language models use to describe your space tells you what the training data skews toward. If that language doesn't match how you describe yourself, you have a positioning gap that is costing you visibility.
Invest in content that explains, not content that performs. Long-form articles that take a position. Case studies that describe the actual decision-making process. Methodology pages that go beyond a three-step diagram. This content serves both LLMs and the humans who want depth before they reach out.
Build third-party mentions deliberately. Contributed articles, podcast appearances, community participation, earned coverage in niche publications. Not for the backlink. For the distributed signal.
This is the part of the job that gets treated as optional because its effects are slow and hard to attribute. But the brands that establish signal now will be harder to displace later. That is not a prediction. It is just how training data compounds.
Most marketing teams are not set up to measure this. Their KPIs are clicks, impressions, and conversions. LLM visibility doesn't produce a clean number to drop into a dashboard. So it doesn't get prioritized.
That is the real risk. Not that the technology is too complex. Not that the playbook is unclear. The risk is that a measurability problem gets confused with an urgency problem, and the work doesn't happen until it's significantly harder to catch up.
If you want to know whether your brand is visible to AI search, the first step is embarrassingly simple: go ask ChatGPT what companies it recommends in your category. See what it says. That answer will tell you more in thirty seconds than a month of internal debate about whether this matters.
