[Chat] Official product research misses learn.chatgpt.com/docs and over-relies on Help Center

Feedback ID

no-active-thread-019fe11f-f59b-7cf3-a977-e3bacab3578c

Product / surface

• Ordinary ChatGPT Chat
• Current ChatGPT web/desktop product-information research behavior
• ChatGPT Pro

What issue are you seeing?

When asking ChatGPT questions about current ChatGPT/OpenAI product behavior and explicitly asking it to research/check official OpenAI documentation, the answer/search path can repeatedly retrieve and cite help.openai.com while failing to discover relevant current first-party documentation under learn.chatgpt.com/docs.

This can materially change the answer. In the observed case, questions about the current unified ChatGPT desktop app and the distinction between Chat, Work, and Codex initially produced Help Center-grounded answers. Relevant and more current product documentation was only found after explicitly instructing ChatGPT to check learn.chatgpt.com/docs.

This is a grounding/source-discovery problem rather than a request to prefer one documentation site universally. The issue is that the official Learn documentation corpus can be omitted unless the user names it.

Why this matters

learn.chatgpt.com/docs is a first-party OpenAI product documentation corpus and contains current documentation for ChatGPT product behavior, including Chat, Work, Codex, desktop/web usage, troubleshooting, changelog/release information, settings, projects/chats, and related product surfaces.

The Learn documentation root also exposes a documentation index (llms.txt) and machine-readable documentation pages, making the corpus suitable for automated discovery/retrieval.

If ChatGPT treats Help Center results as effectively exhaustive for current OpenAI product questions, users can receive incomplete or outdated answers even though relevant first-party documentation is publicly available in the Learn corpus.

Steps to reproduce

1. Open ordinary ChatGPT Chat.
2. Ask a current product-behavior question about ChatGPT, for example a recently changed desktop-app behavior or the current relationship between Chat, Work, and Codex.
3. Explicitly ask ChatGPT to research/check the official OpenAI documentation.
4. Inspect the sources/searches used in the answer.
5. Observe that the research may retrieve/cite help.openai.com while omitting directly relevant pages under learn.chatgpt.com/docs.
6. Tell ChatGPT explicitly to check learn.chatgpt.com/docs.
7. Observe that relevant current documentation is then discovered and can materially revise or clarify the answer.

Expected behavior

For questions about current OpenAI/ChatGPT product behavior, official-source retrieval should discover the complete applicable first-party documentation set, including learn.chatgpt.com/docs when relevant.

The retrieval system should not require the user to already know and explicitly name the Learn documentation domain in order for ChatGPT to find it.

A reasonable discovery path could include the Learn documentation index such as learn.chatgpt.com/docs/llms.txt when current product-documentation lookup is needed.

Actual behavior

ChatGPT can repeatedly search/cite Help Center material while overlooking directly relevant current documentation under learn.chatgpt.com/docs until the user explicitly supplies that domain.

Concrete observed example

While researching behavior in the current unified ChatGPT desktop app, initial searches concentrated on Help Center pages and did not surface the Learn documentation corpus. After the user explicitly pointed out learn.chatgpt.com/docs, the relevant documentation was found and the answer had to be corrected/refined.

This is particularly problematic for "how does ChatGPT currently work?" questions, where first-party product documentation should be the strongest grounding source.

Duplicate search

I searched the public openai/codex issue tracker for combinations involving learn.chatgpt.com, help.openai.com, documentation search/retrieval, missing docs, and ChatGPT product research. I did not find an exact existing report describing this specific Learn-corpus discovery failure.

Additional information

Feedback/diagnostics were submitted from the app before filing this report. The feedback confirmation instructed me to open or reference a GitHub issue if the problem persisted.

Feedback ID for correlation: no-active-thread-019fe11f-f59b-7cf3-a977-e3bacab3578c