Also written as: AI cold-start problem, zero-corpus problem
Last updated: August 18, 2026 - Reviewed by the DataEase AI editorial team
Living through it right now? Read "Why isn't my brand in ChatGPT?" ->
The cold-start problem is a new brand having no third-party corpus for a model to draw on. With nothing written about you anywhere but your own site, an AI assistant cannot name you, however good the product is.
"Our first scan came back empty across every question we cared about. The instinct was to write more blog posts. The actual problem was that nobody outside our own domain had ever written about us, and a model does not take your word for it."
An answer engine composes from text that already exists. Your own website is part of that text, but it is the part with the least weight, because every company describes itself favourably and models are trained on enough marketing copy to know it. What carries weight is corroboration: a directory listing, a review, a comparison article, a roundup, a forum thread, a press mention. Evidence you did not write.
A brand-new company has none of that, by definition. It is not being penalised. There is simply nothing to retrieve. This is why the cold start is different from a ranking problem: you cannot fix it by improving a page, because the missing material is not on your pages at all. It is the seventh pillar of brand readiness, Brand Authority, and it is the only pillar that is evidence rather than a claim.
Every founder's first reaction to an empty scan is to start producing. That is usually premature, because "we are not mentioned" is at least four different situations wearing the same face, and they have different fixes.
| What you see | What it may actually be | First move |
|---|---|---|
| Zero mentions everywhere | No corpus at all, or a missing name variant swallowing real mentions | Register aliases, then baseline a scan |
| Named but never recommended | You exist in the corpus but nothing in it argues for you | Comparison and evidence content |
| Mentioned only on branded questions | People who know you find you; the category does not | Work the non-branded question set |
| Most category answers name nobody | The category itself is unclaimed, not just you | Move first, while the room is empty |
Telling those apart requires a baseline, which is why measurement comes first. Brand discovery in DataEase AI is free and consumes no credits: it works out your competitors, your personas, and the questions buyers actually ask, before you spend anything. Adding a brand domain grants 100 free credits, and every scan shows its exact price before it runs.
No. Models compose answers from what has been written, not from product quality. A brand with 0 third-party citations is invisible to all 5 assistants DataEase AI tracks, no matter how well the product performs for the customers who already found it.
Longer than a content sprint. DataEase AI judges nothing before 7 days, tracks movement across the last 6 scans, and re-audits a single fixed page for 1 credit so you can see whether a change registered before committing a quarter to it.
| Concept | What it covers |
|---|---|
| Cold-start problem | No corpus exists, so there is nothing for a model to retrieve about you |
| Brand Readiness | Whether you are set up to be found once a corpus does exist |
| AI Visibility | What assistants actually say once there is something to say |
| Citation graph | The structure of the corpus you are trying to build |
| Brand alias | A false cold start - the mentions exist but never resolve to your brand |
Brand Alias Share of Model Citation Graph Answer Engine Generative Engine Optimization Brand Readiness Brand Presence Intelligence AI in Branding
For the diagnostic walkthrough, read why your brand is not in ChatGPT, and for what an early-stage audit typically looks like, read our SaaS brand audit findings. For the citation-building side, see the AI citation playbook. To baseline where you stand across the readiness pillars, run the free Brand Analyzer.