Why AI assistants can't see your brand — and the half you can actually fix

Ask ChatGPT, Gemini or Perplexity to recommend a brand in your category, and one of two things happens: your name comes up, or it doesn't. When it doesn't, the instinct is to assume the whole system is a black box you can't influence. It isn't. There are exactly two reasons an AI assistant fails to name you — and only one of them is in your control. Telling them apart is the first real step to becoming visible.

The two ways a brand shows up in an AI answer

Every mention an assistant makes comes from one of two places:

• Parametric memory — what the model absorbed during training. If your brand was written about widely across the web the model learned from, it "remembers" you, even with no live lookup.
• Live retrieval — what the assistant fetches at answer time, by searching the web or reading pages it's given. This is where citations come from, and where up-to-date recommendations are formed.

The distinction matters because the two behave completely differently. Parametric memory shifts only slowly — it changes when the next model is trained on a newer web. You cannot edit it this quarter. Live retrieval, by contrast, responds to what's on your site and who links to you right now.

Which half you can actually fix

Retrieval is the fixable half. If an assistant would surface you when it looks, but your pages aren't structured to be found, quoted and cited, you lose a mention you had already earned. That's not a branding problem — it's a readiness problem, and readiness is buildable.

Concretely, retrieval favours pages that:

• State a clear, self-contained claim near the top, so a single passage can be lifted into an answer.
• Carry clean structured data (JSON-LD) that tells machines what the page is.
• Are corroborated elsewhere — mentions, citations and consistent facts across the web that let a model trust the entity.

None of that requires waiting for the next model. It's why we treat the fixable half as the priority: it moves now.

Why this is what the AI Visibility Index measures

The AVI Score is built on this split. Sixty per cent is Outcome — how often ChatGPT, Gemini and Perplexity actually name and recommend a brand across sampled buying prompts. Forty per cent is Readiness — how well a site is built to be found and cited when an assistant does look. A brand can have a strong reputation and still score poorly because its pages give retrieval nothing to grab; a smaller brand can punch above its weight by being ruthlessly easy to cite.

We publish the aggregate picture in The State of AI Visibility, and rank brands vertical by vertical on the Index. The pattern is consistent: most brands have never deliberately built for AI retrieval, which means the fixable half is wide open.

What to do this week

1. Pick three buying prompts a real customer would type, and ask ChatGPT, Gemini and Perplexity each. Note whether you're named, recommended, or cited.
2. Where you're absent, check whether the answer used live retrieval (look for citations). If it did and you're missing, that's a readiness gap you can close.
3. Fix the most-cited page first: one clear claim up top, valid structured data, and a link or two from places that already talk about you.

Being found by AI is a discipline, not a lottery. The parametric half you earn slowly, over years of being written about. The retrieval half you can start earning today — and it's the half that decides whether the assistant names you the next time someone asks.

Want a reading of where your brand stands? Run a scan for your AVI Score and a prioritised plan.