ChatGPT Search often automatically selects English as its response language when users input prompts in any language except English.

The AI search analytics company Peec AI conducted research that discovered that English functions as a major component of the hidden retrieval methods used by ChatGPT Search.

The company investigated how ChatGPT Search creates background queries, which it calls fan-outs, to develop its ultimate response. The data shows that English plays a large role in that hidden retrieval process.

How ChatGPT Search Uses Fan-Out Queries

OpenAI explains in its ChatGPT Search documentation that when a user asks a question, the system typically rewrites your query into one or more targeted queries and sends them to search partners.

The system uses its initial results to decide whether to search with additional specific queries for other search providers.

The system generates fan-out queries, which function as rewritten queries. Users never see them. The system uses them to decide which sources to examine before creating its initial response.

OpenAI describes how rewriting works, but it does not explain how the system selects the language for those background searches.

43% of Fan-Out Queries Ran in English

Peec AI analyzed more than 10 million prompts and over 20 million fan-out queries from its platform. The company found that 43% of fan-out queries triggered by non-English prompts ran in English.

The research found that 78% of non-English prompt runs contained one or more English-language fan-out queries.

The pattern appeared across languages. 94% of Turkish-language prompts included English fan-outs as required. The Spanish-language prompts showed a 66% occurrence.

The dataset contained non-English languages that maintained a minimum 60% presence.

Peec AI filtered its data to remove mixed signals. The research included only cases that showed a match between the IP address and the prompt language. The example involved Polish-language prompts.

The system processed only those cases that matched the prompt language through the corresponding IP address. Polish-language prompts had to come from Polish IP addresses according to the company policy. The company excluded mismatched combinations.

Peec AI asserts that ChatGPT begins its fan-out process through the original language before expanding its response in English.

Example from Poland’s Leading Marketplace

Peec AI shared case studies to show how this pattern can shape results.

A Polish user used a Polish IP address to perform a Polish-language search about auction portals. The answer either omitted or downplayed Allegro.pl.

Peec AI describes Allegro.pl as Poland’s dominant ecommerce marketplace. The platform enables millions of buyers and sellers to participate in its marketplace while it maintains strong market control.

The response dedicated more space to eBay and other worldwide platforms instead of discussing Allegro.pl. Peec AI states that English-language fan-out queries probably led to more international source material.

The company shared similar examples from other countries. The German-language prompt did not result in a German software company list from ChatGPT during the German testing.

A Spanish test about cosmetics brands started with an English first fan-out query. The second query used Spanish but introduced the term “globales”, which the original prompt did not need.

However, it is important to understand that Peec AI only presents structured test results through testing. They do not represent every ChatGPT Search session that exists in the system.

How Peec AI Collected the Data

Peec AI works within the field of AI search analytics. The company documents their operations, which involve running customer-specific prompts through browser automation on a daily basis.

The organization uses web interfaces to connect with AI platforms instead of using APIs.

The 10 million prompts in this report came from Peec AI’s own platform. They did not come from a panel of consumer ChatGPT sessions.

The company has not published a full industry breakdown of the prompts in the dataset. The findings reflect controlled testing conditions, not general consumer usage.

Why Language Choice Matters

Data from W3Techs shows that English dominates much of the indexed web. It states that more than half of the top 10 million websites use English as their main language.

ChatGPT uses English as its retrieval language to access more sources. The method increases coverage. It directs focus towards international companies and media organizations.

The research shows how language impacts trust relationships. CSA Research reports that 76% of consumers prefer to buy products in their own language.

The study found that 40% of people would not shop from websites that lack proper localization.

ChatGPT uses English language content for its fan-out process. This process creates answers that first rely on global information before any local details shape the response.

Open Questions for AI Search

ChatGPT Search uses query rewriting together with follow-up searches as its fundamental features, according to OpenAI. The company does not state whether language switching is a design choice or a side effect of multilingual systems.

Peec AI’s findings raise a key question. Should AI search systems prioritize the language and location of the user?