Over 800 million users per month now use AI-powered search. ChatGPT, Gemini, Perplexity, and Google AI Overviews do not display a list of 10 blue links — they generate a single synthesis, recommend one company, and cite one source.

For Polish companies, this means a fundamental change in the rules of the game. Previous investments in SEO, content marketing, and performance do not guarantee visibility in this new channel. AI-search is a separate ecosystem with its own rules — and most Polish companies do not even know they do not exist in it.

That is why we conducted the most extensive study to date of Polish business visibility in generative AI model responses. 400+ companies, 5 sectors, 14 crawlers, 6 analysis categories. The results show the scale of the problem — and the window of opportunity for those who act first.

AIVS™ (AI Visibility Score) is a proprietary metric by AIVS measuring company visibility in generative AI model responses. The 0–100 scale aggregates results from 6 weighted categories, calculated by 14 crawlers based on over 400+ Polish companies.

Each company is evaluated across the following categories:

• AI Crawlability — whether AI bots can crawl and index the site
• Structured Data & Schema — quality of structured data (JSON-LD, schema.org)
• Content Readability — content readability for language models (SSR, semantic HTML)
• Entity & Brand Signals — strength of entity and brand signals in the AI ecosystem
• LLM-specific Files — presence of llms.txt, llms-full.txt, and dedicated resources
• Reputation & Trust Signals — review sentiment, E-E-A-T, external citations

The average AIVS™ score for 400+ surveyed Polish companies is 40.6 out of 100. This is the „Closed Fortress” level — companies are partially visible, but technical and content barriers block their potential in AI responses.

Score distribution:

• 66% of companies scored below 50 points
• 34% of companies in the 50–79 range
• 0% of companies above 80 points (AI-Native level)

Three most alarming findings:

• 95% of companies lack an llms.txt file — a dedicated information resource for AI crawlers
• 100% of companies have no deliberate GEO strategy — even those with the highest scores achieve them „by accident” thanks to good technical architecture
• WAFs and firewalls block AI crawlers — many companies actively prevent AI bots from accessing their content

AIVS™ scores differ significantly between sectors. Banking achieves the highest average (47.1), while insurance has the lowest (36.5). The spread between the best and weakest company in each sector reaches 30–50 points, showing that differences stem from individual technical and content decisions, not from industry characteristics.

It is worth noting that all sectors fall within the „Closed Fortress” range (35–59). Differences within sectors are greater than differences between sectors — meaning that an individual company can significantly improve its position regardless of industry.

The telecommunications sector — four major players serving tens of millions of customers — achieves an average of just 39.5 points. A paradox: companies spending hundreds of millions on marketing are poorly visible in a channel that is capturing an ever-growing share of purchasing decisions.

Key finding: When a user asks ChatGPT about the best mobile plan in Poland, the model does not cite any operator directly. Instead, it refers to comparison sites, Reddit, and tech blogs. Operators — despite billion-dollar advertising budgets — cede the narrative about their offering to intermediaries.

T-Mobile and Orange lead thanks to better technical architecture (SSR, schema.org), but neither pursues a deliberate GEO strategy. Their advantage is „accidental” — and easy for competitors to close once they start acting deliberately.

E-commerce is the sector where AI visibility has a direct impact on sales. When a user asks AI „Where should I buy a laptop for 3,000 PLN?”, the cited platform gets the customer — the rest do not exist.

The Allegro Paradox: Poland’s largest e-commerce platform scores 62 points — highest in the sector, but still below the AI-Native threshold. Allegro benefits from an enormous content base and strong entity signals, but does not deliberately optimize for AI responses.

The CSR (Client-Side Rendering) problem: Media Expert with a score of 29 points is an example of a company whose site was built with the user in mind, but renders content on the client side. AI bots see an empty template — not the offering. A classic „Invisible Giant”: a large brand, invisible to AI.

Banking achieves the highest average among the surveyed sectors (47.1), which stems from higher technical and regulatory standards. Despite this — no bank comes close to the AI-Native level.

The YMYL problem: Banking falls under the „Your Money or Your Life” category — AI models are especially cautious with financial recommendations. This raises the entry threshold: to be cited, a bank must have not only good technical architecture, but also strong authority signals (E-E-A-T), up-to-date structured data, and positive sentiment in external sources.

PKO BP — the largest bank in Poland — scores just 49 points. Santander leads thanks to a better technical layer, but the difference is small and stems from individual architectural decisions, not a GEO strategy.

The insurance sector achieves the lowest average of all surveyed industries. This is especially concerning given that insurance questions are among the most common commercial queries directed at AI assistants.

The Uniqa case: Uniqa blocks all AI crawlers in robots.txt. This is a deliberate decision that effectively eliminates the company from generative model responses. In an era where more and more customers search for insurance through AI assistants, blocking bots is the equivalent of closing the store during peak hours.

PZU — the leader of the Polish insurance market — scores just 38 points. Allianz leads thanks to its international technical infrastructure and stronger entity signals in the global ecosystem.

The media sector is a special case — content is their core business, and at the same time it is precisely content that is „borrowed” by AI models to generate responses. The copyright vs. AI visibility dilemma is sharpest here.

The copyright vs. AI dilemma: Media face a fundamental choice. Blocking AI crawlers protects content from „borrowing,” but simultaneously eliminates the brand from generative responses. Onet and Benchmark achieve the highest scores in the sector because their technical architecture favors indexing — but that does not mean they deliberately optimize for AI.

90minut with a score of 12 points is the lowest result in the entire media sector study. The site is practically unreadable for AI crawlers due to intensive Client-Side Rendering and lack of basic structured data.

Below is a comparison of the 10 companies with the highest and 10 companies with the lowest AIVS™ scores among all 400+ surveyed entities across 5 sectors.

Top 10 — highest AIVS™ scores

Bottom 10 — lowest AIVS™ scores

The spread between the best (T-Mobile, 64) and the weakest score (90minut, 12) is 52 points. Importantly — even the ranking leader is far from the AI-Native level (80+). This shows that the entire Polish market is in a pre-GEO phase.

1. The Polish market is in a pre-GEO phase. None of the 400+ surveyed companies pursues a deliberate Generative Engine Optimization strategy. Even the highest scores (60–64 pts) are the result of good architectural decisions, not deliberate optimization for AI. This means the window of opportunity is open — companies that start first will build an advantage that is hard to match.

2. Technical barriers block visibility more than lack of content. Most companies have good content, but technical barriers — Client-Side Rendering, lack of SSR, AI crawler blocking by WAF, lack of llms.txt — prevent models from reaching it. Fixing these barriers is the fastest path to improving an AIVS™ score.

3. Differences within sectors are greater than between sectors. Sector averages differ by about 10 points (36.5–47.1), but the spread within each sector reaches 30–50 points. This proves that individual technical and content decisions matter more than industry specifics.

4. llms.txt is low-hanging fruit — and nobody is picking it. 95% of companies lack an llms.txt file. This is a dedicated information resource for AI crawlers, analogous to robots.txt. Its implementation is simple, inexpensive, and sends an immediate signal to models that the company is „AI-aware.” Yet — almost nobody does it.

5. Online reputation is becoming an algorithmic factor. AI models use reviews on Trustpilot, Google Reviews, and other platforms as a trust signal. Negative sentiment not only lowers the AIVS™ score but actively blocks recommendations. Reputation management is no longer PR cosmetics — it is becoming a key visibility factor in AI.

GEO is today exactly where SEO was in 2010. Few players understand its significance, entry costs are low, and the potential for building lasting competitive advantage — is enormous. But the window is closing.

In the next 12–24 months, AI models will become increasingly precise in their recommendations. Companies that by then build strong entity signals, implement llms.txt, optimize their architecture for AI crawlers, and take care of their online reputation — will be „encoded” in the models as default references in their categories.

The rest will catch up — at many times the cost and with much lower chances of success.