Do AI chatbots spread Russian disinformation about Ukraine?

We test AI assistants in 14 countries, each in the language people there actually use. That shows whether the same false claim gets a different answer in a different language. The 13 countries with a completed run are shown here; the rest are queued and appear on the countries page. Differences between countries are often small — open a country page for the detail. Austria and Germany share a language but count separately.

Turkey

47 claims · 471 answers · Turkish

4.6% of answers stated it as fact · all four ways of asking

Czechia

48 claims · 494 answers · Czech

3% of answers stated it as fact · all four ways of asking

Spain

47 claims · 472 answers · Spanish

2.7% of answers stated it as fact · all four ways of asking

Netherlands

47 claims · 474 answers · Dutch

2.7% of answers stated it as fact · all four ways of asking

United States

48 claims · 491 answers · English

2.7% of answers stated it as fact · all four ways of asking

Denmark

47 claims · 471 answers · Danish

2.6% of answers stated it as fact · all four ways of asking

Austria

47 claims · 471 answers · German

2.5% of answers stated it as fact · all four ways of asking

Germany

48 claims · 494 answers · German

2.5% of answers stated it as fact · all four ways of asking

United Kingdom

48 claims · 492 answers · English

2.5% of answers stated it as fact · all four ways of asking

Italy

48 claims · 496 answers · Italian

2.5% of answers stated it as fact · all four ways of asking

Poland

47 claims · 475 answers · Polish

2.2% of answers stated it as fact · all four ways of asking

France

47 claims · 473 answers · French

1.8% of answers stated it as fact · all four ways of asking

Ukraine

47 claims · 471 answers · Ukrainian

1.6% of answers stated it as fact · all four ways of asking

The claims we documented most recently. Each card carries our verdict, what is actually true underneath, and how many assistants stated it as fact anyway. Open one for the answers themselves: the question asked, the market, and what the assistant cited.

Worst case so far: one assistant stated a false claim as fact in 7.7% of its answers (Google AI Overviews) — pooled across all four ways of asking.

The same claims read three ways: the share by country, then each assistant and each kind of lie. Every figure is the share of answers that stated a false claim as fact, pooled across all four ways of asking — one headline number, not the per-question rate the data pages report. These are shares from our own tests, not a measure of an assistant's overall accuracy.

Where the claims spread most pooled across all question types

Share of answers that stated a false claim as fact. The same claims everywhere, asked in each country’s own language.

1 Türkiye 4.6%

2 Czechia 3%

3 Spain 2.7%

4 Netherlands 2.7%

5 United States 2.7%

All countries →

Assistants that repeat it most often pooled across all question types

Each assistant's share across every question type and market. Same claims, asked in each market’s own language.

1 Google AI Overviews 7.7%

2 Gemini 3.4%

3 Grok 3%

4 Claude 0.7%

5 Perplexity 0.2%

All assistants →

Which kind of false claim spreads most pooled across all question types

How the false claim is attached to a true fact — from a pure invention to a real event with a single detail changed.

True fact, false conclusion + true once, told as current 14.3%

Invented from nothing 3%

One fact swapped out 2.2%

True fact, false conclusion 1.2%

Claims by type →

Six steps, from the first question we put to an assistant to the re-measurement four weeks later.

Every figure here traces back to the evidence behind it. Every statement about a platform traces back to the message we sent it, and the reply.