When using ChatGPT with browsing enabled (GPT-4o), I consistently encounter a problem:
Even when I explicitly instruct the model to search only in English-language websites — such as “Search in English only, exclude all Russian content” — it continues to return results from clickbait-style Russian-language websites, even though my request is very specific.
Example:
Prompt:
“Search for recent discoveries in the Giza pyramids. Use only English-language scientific or journalistic sources like National Geographic, Smithsonian Magazine, Science News, or PubMed.”
Actual response:
Includes translated or speculative content from non-English, sensationalist websites, clearly not matching the request.
This happens even when I use detailed filters, such as site: constraints. The model still adds unrelated results from questionable or non-English sources.
Why this matters:
- It contradicts clear user instructions
- It pulls in unreliable or mistranslated content
- It breaks the value of web search for users needing accurate, language-specific information
- It undermines trust for academic, scientific, and technical users
Suggestions to improve:
- Strictly enforce user instructions when filtering by language or domain
- Fully respect
site:or source-based filters - Provide a user setting to exclude specific languages or content regions
- Prioritize results from high-quality, verified sources in the requested language
This issue has been happening for several months despite repeated prompts and tests. I hope the dev team can review it seriously — it’s not a one-time glitch, but a persistent problem that affects many multilingual users.
Thanks!