ChatGPT Search Defaults to English

A new study has revealed that ChatGPT often defaults to English when carrying out its background web searches, even when users submit questions in other languages. The findings come from research conducted by Peec AI, which examined how ChatGPT builds answers using what are known as “fan-out queries”.

Fan-out queries are the additional searches ChatGPT runs behind the scenes when it needs to gather information from the web. These searches help the system verify facts and pull in supporting sources before presenting a final response to the user.

While users may ask questions in languages such as Polish, German or Spanish, the research suggests that a large proportion of these background searches are still carried out in English.

What the Research Looked At

Peec AI analysed data from more than 10 million ChatGPT prompts and over 20 million fan-out queries. To ensure accuracy, the company only included cases where the language of the prompt matched the user’s location. For example, Polish prompts from Poland and German prompts from Germany were included, while mixed signals such as German prompts from UK IP addresses were excluded.

Once filtered, the data showed a striking pattern. Around 43% of all fan-out queries were conducted in English, even when the original user question was written in another language.

More significantly, nearly 78% of non-English prompts generated at least one English-language fan-out query during the process of building a response.

Turkish-language prompts had the highest proportion of English fan-out queries at 94%, while Spanish-language prompts recorded the lowest share at 66%. No non-English language group fell below 60%.

How Fan-Out Queries Work

According to OpenAI documentation, ChatGPT Search does not simply send one query to a search engine. Instead, it rewrites the user’s question into several smaller and more targeted searches. These are then sent to search partners to retrieve relevant information.

After reviewing the first set of results, the system may issue further searches that are more specific. This multi-step process helps refine the final answer.

However, OpenAI does not publicly explain how ChatGPT chooses which language to use for these rewritten queries. Peec AI’s findings suggest that English is frequently introduced during this stage, even if the user never asked in English.

Examples From the Study

The report also included several real-world examples showing how this behaviour can influence the results users receive.

In one case, a Polish-language prompt asking about auction websites produced answers focused on global platforms such as eBay, while giving little attention to Allegro.pl, which is Poland’s largest and most popular online marketplace.

Another example involved a German-language question about German software companies. The response failed to list any local firms and instead highlighted international businesses.

A similar issue appeared with a Spanish prompt about cosmetics brands. The results focused on well-known global names rather than Spanish brands. In this instance, Peec AI showed that ChatGPT’s first fan-out query had been written in English. A later query switched back to Spanish but added the word “global”, even though the user had not asked for global brands.

These cases suggest that English fan-out queries may push ChatGPT towards international sources rather than local ones.

Why This Matters for Businesses and SEO

The findings raise concerns for companies and content creators operating in non-English markets. If ChatGPT relies heavily on English-language background searches, it may prioritise English sources when selecting information, even for users searching in their native language.

This could make it harder for local businesses to appear in AI-generated answers, especially if their content is written only in local languages. Global brands with strong English-language websites may gain an advantage simply because their content is more likely to be discovered during fan-out searches.

For SEO teams, this introduces a new challenge. Traditional optimisation strategies focus on local keywords and language-specific content. However, AI search tools may not follow the same logic if they are drawing from English sources during their background research.

Limitations of the Study

Peec AI gathered its data through browser automation using customer-defined prompts. This means the prompts did not come from everyday consumer use, but from controlled testing through its platform.

The company did not publish details about which industries or topics were covered by the prompts, so the results may not fully represent typical ChatGPT usage across all sectors.

The report was authored by Tomek Rudzki, a technical SEO specialist who has spoken at international industry events and contributed to several marketing publications.

What This Could Mean Going Forward

At present, OpenAI’s documentation explains that ChatGPT rewrites and expands user queries but does not clarify how language selection works in this process. It remains unclear whether the preference for English is intentional or simply a by-product of how the system processes information.

The report raises an important question for the future of AI-driven search: will businesses need to produce more English-language content to remain visible, even in local markets? Or will AI systems evolve to better reflect regional languages and local sources?

As AI search continues to develop, language handling is likely to become a key issue for publishers, marketers and users alike. The Peec AI findings suggest that while ChatGPT aims to provide helpful global answers, its underlying search behaviour may still be strongly influenced by English-language data.

For now, companies operating in non-English markets may need to think carefully about how their content is structured and whether multilingual strategies will become essential for visibility in AI-powered search tools.