Cold-Start AI Visibility
The starting state in which a brand has never appeared in an AI answer, and the measurement design for getting from that state to a first appearance. The key variables are a baseline period, time to first citation, and the order in which engines start naming the brand
What is cold-start AI visibility?
Cold-start AI visibility is the starting state in which a brand has never appeared in an AI answer, together with the way you measure and design the path from that state to a first appearance. The term borrows the cold-start problem from recommender systems, where new users or items have no history. It applies to new brands, established brands entering a new category, and brands entering a foreign market for the first time.
How is it different from an established brand?
An established brand's task is to appear more often and higher up. A cold-start brand's task is to enter the candidate list at all. With no third-party content for the AI to draw on, either the brand's own pages are the only evidence or there is no evidence and the brand is never named. In one experiment, a SaaS brand with a zero baseline reached 298 appearances in 30 days by combining third-party listicle placements, an owned listicle and PR — and 85.8% of those came from third-party listicles. The owned listicle took 18 days to be cited for the first time.
Measurement design
In a cold-start state, the measurement design decides how the results can be read.
| Element | Why it matters | How |
|---|---|---|
| Baseline period | You need to confirm that "zero" is not chance before later changes can be read as effects | Run the same question set repeatedly for 2–4 weeks before starting and confirm zero |
| Unbranded queries | Queries that include the brand name return answers even in a cold start and overstate visibility | Use category and commercial-intent queries only |
| Time to first citation | Speed differs by placement type — third-party is fast, owned is slow | Record the placement date and the first observed citation date per placement |
| Per-engine separation | Engines start naming new brands at different times; ChatGPT was last in the experiment | Log runs separately for ChatGPT, Gemini, Perplexity and others |
| Repeated runs | Variance is high right after a first appearance; a single check cannot be told apart from chance | Collect several runs under the same conditions and read the trend |
Measurement notes
- Speed depends on the category — the 30-day case was a 15-keyword experiment. Time to appearance varies with competitive intensity and engine, so measure it in your own category.
- Appearing is not the same as being visited — whether a first appearance turns into traffic has to be cross-checked against referral and AI-channel data.
- The next step is the source gap — after the first appearance, the work continues with source gap analysis: finding the sources that cite only your competitors and getting into them.
Related terms
Further reading
- From Zero to 298 AI Mentions in 30 Days — 85.8% Came From Third-Party Placements — platform distribution, source mix and caveats of an experiment that started from a zero baseline
- The Branded-Query Baseline Trap — how branded queries distort a cold-start baseline
- The Month-over-Month Illusion — run-to-run variance and measurement design
Is your site visible in AI search?
See for free how ChatGPT, Perplexity, and Gemini describe your brand.
Start Free Diagnosis →