We Audited 30 of CEE's Must-Follow Startups. None Were Strongly Visible in AI.

Vestbee published its list of must-follow startups from Summer Pitch CEE 2026 on July 21, 2026, including the ten companies its jury selected to pitch live. Making that list puts a company in front of investors and press. But there is a second kind of visibility that increasingly decides whether a buyer ever finds you: whether an AI model names you when someone asks it about your category.

More B2B research now starts with a question typed into ChatGPT, Perplexity, Claude, or Gemini than most founders realize. If the model does not know you exist, you are absent at the exact moment a buyer builds a shortlist. So we ran the jury's ten picks, plus twenty more names from the list, through VisibAI to see how visible they actually are inside AI answers.

The short version: not very. And that is the interesting part.

TL;DR

• Of 30 startups measured, none reached strong AI visibility (a score of 70 or above).
• The highest score was 49 out of 100. The cohort average was roughly 36.
• The ten jury-selected companies averaged about 34, no better than the wider list.
• Being picked as a standout told you almost nothing about whether AI could find the company.
• 18 of the 30 were effectively invisible. 12 were moderate. None were strong.
• We ran our own site through the same audit and scored 49. We are not exempt.

What we measured

Each company was mapped to a category based on how it describes what it does, then run through a set of blind, non-branded category prompts, the kind a real buyer would ask, across multiple AI answer engines including ChatGPT, Perplexity, Claude, and Gemini. We never mentioned a company by name in a prompt. Results were averaged across runs to smooth out the natural variance in model outputs, then rolled into a single visibility score from 0 to 100.

The bands we use throughout: strong is 70 to 100, moderate is 40 to 69, and anything below 40 is effectively invisible. All 30 audits were run on July 21, 2026.

The jury's top 10, scored

Only three of the ten cleared the moderate line. The other seven, despite being chosen to pitch on stage, are effectively invisible to the tools a buyer would ask first.

Twenty more from the list

Across all 30 companies, 12 landed in the moderate band and 18 were effectively invisible. None were strong.

A note on our own scorecard

It would be fair to ask whether we are holding these companies to a standard we do not meet ourselves. So we ran our own site through the same audit. VisibAI scored 49, the same ceiling the rest of the field hit, and comfortably short of strong.

We are describing a problem we are also working on, not standing outside it. That is rather the point. AI visibility is hard-won for everyone. It is earned through coverage and time, not declared.

Why so many good companies score low

A low score almost always means the models have not learned the company yet, not that the product is weak. Three threads ran through the low scorers.

The category is young. For newer deeptech, defence, and climate niches, there is simply less public writing for a model to absorb. Even strong companies go unnamed because the material barely exists.

Coverage is thin. Companies that live mostly on their own marketing site, with few third-party mentions, directory listings, or independent write-ups, give models very little to learn from. A curated list like this one is exactly the kind of source that helps.

Citations are missing. Being mentioned is not the same as being linked. Across this set, even companies that surfaced in an answer were rarely cited with a source. That limits both referral clicks and the repeated exposure that teaches a model to trust a name.

If your startup scored low, here is what to do

A low AI visibility score is a content and distribution problem, not a verdict on the business. The levers that move it:

• Make sure AI crawlers can read your site. Many sites unintentionally block GPTBot, ClaudeBot, or PerplexityBot, or bury their content behind heavy JavaScript. If the crawler cannot read you, the model cannot learn you.
• Earn third-party coverage. Directory listings, editorial mentions, and comparison pages are what models absorb. Your own homepage is the weakest signal you have.
• Publish category-anchored content. State plainly what category you are in and who you are for, so a model can connect a neutral buyer question to your name.
• Add structured data and an llms.txt file. Help machines parse what you do rather than guess.
• Track it over time. Visibility shifts as models retrain and as your coverage grows. Measuring monthly is how you know whether the work is paying off.

For a fuller walkthrough of the tactics, see our 2026 GEO playbook and our guide to AI visibility for startups with no brand history.

The wider point

The gap this data exposes is not between good companies and bad ones. It is between companies the internet has written about and companies it has not. Right now, in Central and Eastern Europe, that second group is most of the ecosystem.

That is a genuine opportunity. The categories where nobody scores well are categories where the first company to build real coverage becomes the default answer. Being early to AI visibility in a young category is worth far more than being incrementally better at it in a crowded one.

Frequently asked questions

What is a good AI visibility score?

On our scale, 70 or above is strong, 40 to 69 is moderate, and below 40 means a brand is effectively invisible in AI answers. In this cohort of 30 CEE startups, no company reached strong and the average was around 36. For context, established brands in mature, well-documented categories routinely score far higher, because there is simply more written about them for models to learn from.

Why do early-stage startups score badly in AI search?

Because generative engines answer from what has been written about a company across the open web, not from what the company says about itself. A young startup usually has no Wikipedia entry, few independent write-ups, and thin domain authority, so the model has almost nothing to draw on. It is a coverage problem rather than a product problem.

Does being featured on a startup list improve AI visibility?

It helps, but slowly and only as part of a wider pattern. A single mention on a curated list is one signal among many. Models tend to surface brands that appear consistently across multiple independent sources over time. One listing rarely moves a score on its own, which is precisely what this data shows: appearing on a respected must-follow list did not correlate with being findable in AI answers.

Which AI engines were used in this study?

Blind, non-branded category prompts were run across multiple answer engines including ChatGPT, Perplexity, Claude, and Gemini, with results averaged across runs to reduce variance.

Can a low AI visibility score be fixed?

Yes. The main levers are making sure AI crawlers can actually read your site, earning third-party coverage on sources models already trust, publishing clear category-anchored content, adding structured data, and then tracking visibility over time as models retrain.

How often does AI visibility change?

It shifts as models retrain and as coverage about a brand accumulates, so a score reflects discoverability on the day it was measured. Monthly tracking is usually enough to see whether the work is having an effect.

Scores captured July 21, 2026. Model outputs vary between runs, so a score reflects AI discoverability on the day it was measured, not product quality.

See where your own brand lands

When a buyer asks AI about your category, does your brand come up, or does a competitor's? You can run the same audit on your own company in a couple of minutes, free.

Run a free AI visibility audit