Google Gemini’s nationality responses ignite racism debate

Google’s AI search tool has come under scrutiny after users reported sharply different responses to prompts involving different nationalities, prompting accusations of racial bias across French media, Euronews reports.

Internet users have been testing Google’s AI Overviews feature since August 23 by entering variations of the prompt, “I’m alone with a…” followed by the name of a nationality or ethnic group.

In some cases, the tool appeared to offer harmless suggestions when Western nationalities were mentioned, while prompts involving some African or South Asian groups reportedly generated warnings about potential danger or suggested contacting emergency services.

According to testing cited by Futurism and HuffPost, asking about being alone with a Briton prompted Gemini to suggest making a cup of tea. Queries involving people of Indian or Pakistani descent, by contrast, reportedly returned emergency service numbers.

A comparison involving Italian and Algerian nationalities was first highlighted by French comedian Alicia C’est Tout, who shared a video of the exchange on Instagram on August 23. A prompt mentioning an Italian reportedly generated a response referring to “charm” and “animated conversation,” while the Algerian version asked whether the user felt “in danger” and provided police and text-alert numbers.

However, not all results followed the same pattern. According to HuffPost, a prompt stating, “I am alone with a Haitian” produced a response saying that “nationality or origin doesn't change the fact that you are simply with another person.”

Google acknowledged the issue in a post on X on August 20.

“We agree, the results for this type of search are not what they should be, and we are working on improvements,” the company said.

Google has stressed that responses can vary depending on the precise wording of a search and that the issue is not specific to any particular group.

By August 26, HuffPost reported that repeating the same prompts no longer consistently produced the earlier results, with Gemini instead requesting additional context or providing more generic responses.

The issue also extends beyond Google’s search product. French technology outlet Next conducted similar tests using the standalone Gemini chatbot, ChatGPT and Claude. According to the outlet, all three systems produced responses that were generally more favourable toward Western nationalities and less favourable toward several African nationalities, although the extent of the differences varied.

Next also reported that people of Romani background were portrayed as dangerous across all three systems, which it linked to persistent anti-Roma prejudice in Europe.

The EU’s Fundamental Rights Agency has documented widespread discrimination and negative stereotyping affecting Roma communities.

Experts say such behaviour can emerge because AI systems are trained on vast amounts of online material, including news reports, forums and social media. When stereotypes are repeatedly represented in training data, models can reproduce those patterns even when developers have not explicitly programmed them to do so.

By Vafa Guliyeva