# AI Blind Test 2026: Analysis of 7,184 AI answers on 449 mid-sized companies in Germany

> Analysis of 7,184 AI answers on 449 mid-sized companies in Germany

Source: https://www.codaai.ai/en/study/
Published: 2026-07-14
Updated: 2026-09-22
May be quoted with attribution and a link to the source.

---

**Author:** Oliver Parrizas ([author page](https://www.codaai.ai/en/author/oliver-parrizas/)) · **Publisher:** CodaAI (AMP Beratung, Gütersloh) · **Version:** 1.0 · **Data as of:** June/July 2026

## Key findings

1. In 54.7% of the 3,592 buyer questions examined (1,965 questions), neither ChatGPT nor Google AI Overviews names the company concerned.
2. For 46.8% of the 449 companies examined (210), the AI systems recommend competitors by name while the company itself is missing from the majority of its buyer questions.
3. 12.9% of the companies examined are not named in any of their eight buyer questions.
4. 34.9% of the 361 companies with more than 100 top-10 rankings on Google are completely invisible in ChatGPT.
5. 25.6% of the companies are not named even in the market-overview question about their own category.
6. Average visibility is 24.9% in ChatGPT and 39.4% in Google AI Overviews; 39.6% of companies are never named by ChatGPT, 15.1% never by Google AI Overviews.
7. For 39.0% of the companies, their own website is not cited as a source in any AI answer.
8. Where companies are named, their average position is 1.8; the company takes first place in 23.8% of all questions.

_Source: CodaAI, AI Blind Test 2026 (n = 449 companies, June/July 2026), https://www.codaai.ai/en/study/_

## Methodology

- **Sample:** 449 mid-sized companies from the German-speaking market, predominantly B2B (among others mechanical engineering, electrical engineering, construction suppliers, industrial services). Selected from our own audit practice — not a random sample (see limitations).
- **Question generation:** Per company, 8 qualified German buyer questions in three archetypes: market overview (2), use case (4), competitor comparison (2). Created on the basis of an automated analysis of the website, the industry and the competitive landscape, with a deterministic quality check. The question sets are frozen and can therefore be repeated exactly for tracking over time.
- **Query:** Every question was put to two interfaces: ChatGPT (model answer without live web search — this measures the structural "knowledge" of the model) and Google AI Overviews (German localisation). Collection period: June/July 2026.
- **Evaluation:** For each answer we recorded: mention of the company (including brand aliases), position of the mention, competitors recommended by name, and whether the company website is cited as a source. Aggregated into visibility metrics per company and across the full dataset.
- **Monetary context:** The opportunity costs quoted are a conservative model calculation based on organic traffic values (SISTRIX data) and discovery assumptions — intended as an order of magnitude, not as a forecast. It is labelled as a model calculation throughout.

## Limitations

- Not a random sample: the companies come from our audit practice and are B2B-heavy. The statements hold for the 449 companies analysed — not for mid-sized companies in Germany as a whole.
- A snapshot: AI answers are volatile; model updates can shift mentions. The data date is stated; the frozen question sets make tracking over time possible.
- ChatGPT was deliberately queried without live web search in order to measure the structural knowledge of the model. With web search enabled, individual answers can differ.
- The monetary context is a model calculation with conservative assumptions, not a revenue forecast.
- Two smaller, independent third-party surveys (24 and 150 companies, spring 2026) arrive at similar orders of magnitude — that supports the direction of the findings, but does not replace representativeness.
- The study measures the German market in German. AI systems work the same way elsewhere, but we have not measured other markets — so we do not claim them.

## How to cite

The aggregated figures may be quoted freely with attribution. Please state the base (share of questions or share of companies) and note that this is not a random sample.

**APA (7th edition)**

Parrizas, O. (2026). AI Blind Test 2026 [KI-Blindtest Mittelstand 2026]: Analysis of 7,184 AI answers on 449 mid-sized companies in Germany (Version 1.0). CodaAI. https://www.codaai.ai/en/study/

**Harvard**

Parrizas, O. (2026) AI Blind Test 2026: Analysis of 7,184 AI answers on 449 mid-sized companies in Germany. Version 1.0, data as of June/July 2026. Gütersloh: CodaAI (AMP Beratung). Available at: https://www.codaai.ai/en/study/

**Short credit for articles and slides**

Source: CodaAI, AI Blind Test 2026 (n = 449 companies, June/July 2026), https://www.codaai.ai/en/study/

**BibTeX**

```bibtex
@techreport{parrizas2026aiblindtest,
  author      = {Parrizas, Oliver},
  title       = {AI Blind Test 2026: Analysis of 7,184 AI answers on 449 mid-sized companies in Germany},
  institution = {CodaAI (AMP Beratung)},
  address     = {G{\"u}tersloh},
  year        = {2026},
  month       = jul,
  note        = {Version 1.0, data as of June/July 2026},
  url         = {https://www.codaai.ai/en/study/}
}
```

Key figures as CSV: https://www.codaai.ai/en/study/key-figures.csv