ChatGPT Search Often Switches To English In Fan-Out Queries: Report via @sejournal, @MattGSouthern

When ChatGPT Search builds an answer, it can generate background web queries to find sources. A new report from AI search analytics firm Peec AI found that a large share of those background queries run in English, even when the original prompt was in another language.

Peec AI analyzed over 10 million prompts and 20 million fan-out queries from its platform data. Across all non-English prompts analyzed, the company reports that 43% of the fan-out steps were conducted in English.

What Are Fan-Out Queries

OpenAI’s ChatGPT Search documentation describes fan-out queries. When a user asks a question, ChatGPT Search “typically rewrites your query into one or more targeted queries” and sends them to search partners. After reviewing initial results, “ChatGPT search may send additional, more specific queries to other search providers.”

Peec AI refers to these rewritten sub-queries as “fan-outs.” The company’s report tracked which languages ChatGPT used when generating them.

OpenAI’s documentation does not describe how language is chosen for rewritten queries.

What Peec AI Found

Peec AI filtered its data to include only cases where the IP location matched the prompt language. Polish-language prompts from Polish IP addresses, German-language prompts from German IPs, and Spanish-language prompts from Spanish IPs. Mixed signals, such as German-language prompts from UK IP addresses, were excluded.

The filtered data showed that 78% of non-English prompt runs included at least one English-language fan-out query.

Turkish-language prompts included English fan-outs most often, at 94%. Spanish-language prompts were lowest, at 66%. No non-English language in Peec AI’s dataset fell below 60%.

Peec AI’s data showed a consistent pattern across languages. ChatGPT typically starts its fan-out queries in the prompt’s language, then adds English-language queries as it builds the response.

Examples From The Report

Peec AI’s blog post included several examples showing how the pattern can play out in practice.

When prompted in Polish from a Polish IP address about the best auction portals, ChatGPT either omitted or buried Allegro.pl in favor of eBay and other global platforms. Peec AI describes Allegro as Poland’s dominant ecommerce platform.

When prompted in German about German software companies, Peec AI reported the response listed no German companies. When prompted in Spanish about cosmetics brands, no Spanish brands appeared.

In the Spanish cosmetics example, Peec AI showed ChatGPT’s actual fan-out queries. The first ran in English. The second ran in Spanish but added the word “globales” (global), a qualifier the original prompt never used. The system appears to have interpreted a Spanish-language prompt from a Spanish IP address as a request for global brands.

These are individual examples from Peec AI’s testing, not necessarily representative of all ChatGPT Search behavior.

Why This Matters

SEO and content teams operating in non-English markets may face a disadvantage in ChatGPT’s source selection that may not map cleanly to traditional ranking signals. In Peec AI’s examples, English-language fan-out queries surfaced English-language sources that favored global brands over local competitors.

We’ve been covering ChatGPT’s citation patterns for over a year now, from SE Ranking’s report on citation factors to the Tow Center’s attribution accuracy findings. Those earlier reports showed which signals predict whether a source gets cited. Peec AI’s data suggests the language of the background query may filter which sources are even considered, before citation signals come into play.

Methodology Notes

Peec AI is a vendor in the AI search analytics space. The company’s documentation describes its data collection method as running customer-defined prompts daily via browser automation, interacting with AI platforms through their web interfaces rather than APIs. The 10 million prompts in this report came from Peec AI’s platform, not from a panel of consumer ChatGPT sessions.

The report didn’t detail the composition of those prompts, what categories or industries they covered, or how representative they are of broader ChatGPT usage patterns.

Tomek Rudzki, the report’s author, is presented by Peec AI as a “GEO Expert” on its blog. He is a well-known technical SEO practitioner who has spoken at BrightonSEO and SMX Munich and contributed to publications such as Moz.

Looking Ahead

OpenAI’s public ChatGPT Search docs describe query rewriting and follow-up queries but don’t explain how language is chosen for those queries. Whether the English fan-out pattern Peec AI identified is an intentional design choice or an emergent behavior of the system remains unclear.

The report raises a question worth monitoring. Will building English-language content become part of AI search optimization strategies, or will AI search platforms adjust their source selection to better reflect local markets?

Featured Image: arda savasciogullari/Shutterstock