Hello,
I would like to suggest a feature that could help ChatGPT answer questions about obscure, poorly documented, or rapidly changing subjects where reliable information is difficult to find through ordinary web search.
The idea would be an opt-in community knowledge system based on submitted evidence, rather than simply allowing ChatGPT to believe information because users repeatedly state it.
Users who explicitly consent could allow information from their conversations to contribute to a shared evidence pool. Claims could be supported by materials such as screenshots, videos, logs, files, source code, reproducible tests, official documentation, or other relevant evidence.
The important distinction is that quantity alone should not determine whether a claim is treated as reliable. For example, 10,000 people repeating the same statement should not automatically outweigh one piece of much stronger evidence. The system could instead consider factors such as evidence quality, source independence, reproducibility, software/version information, and whether different forms of evidence support the same conclusion.
Information obtained this way should also be clearly identified as community-derived rather than presented as guaranteed fact. For example:
“Community-sourced information: Multiple independent reports and supporting files indicate this behavior, but it has not been officially confirmed.”
Different confidence levels could also be used depending on the evidence available.
An important part of this system would be handling conflicting evidence. If two explanations have approximately equal amounts and quality of independent evidence, ChatGPT should not arbitrarily choose one. Instead, it could present both possibilities and state that the available evidence is currently insufficient to determine which is correct.
The system could also attempt to determine whether apparently contradictory reports are actually caused by different circumstances, such as different software versions, settings, operating systems, hardware, game configurations, regions, or other environmental differences.
Another issue would be avoiding feedback loops. If ChatGPT tells users that community evidence suggests something, users later repeating that same information should not automatically count as independent confirmation. Ideally, the system would distinguish between original observations/evidence and information that originated from ChatGPT or another existing source.
Versioning would also be important. A fact that is correct for one version of a game, program, or service may become incorrect after an update. Community evidence could therefore be associated with relevant versions, dates, configurations, and other contextual information where possible.
Privacy would require explicit consent and appropriate safeguards. Even when a contributor consents, submitted material could contain information belonging to other people, so personal or sensitive information would still need to be detected, excluded, or redacted.
I also think this could potentially exist as an optional information source rather than replacing ChatGPT’s normal sources. A standard mode could continue using the existing model knowledge, web search, official documentation, and other established sources, while a community-informed mode could additionally consult this evidence pool. Users who prefer not to rely on crowdsourced information could therefore avoid it entirely.
The use case that originally made me think about this was troubleshooting obscure game modifications. During normal ChatGPT conversations, users can sometimes provide screenshots, crash reports, program files, code, and repeated observations that reveal information not documented anywhere publicly. Other users may later encounter the exact same problem, but ChatGPT currently cannot make that accumulated evidence available to them as a structured, clearly labelled knowledge source.
The same concept could potentially be useful beyond games—for uncommon software bugs, hardware problems, compatibility issues, niche applications, undocumented behavior, and other topics where conventional search results are limited.
I understand that implementing something like this would introduce substantial challenges involving moderation, malicious or fabricated evidence, privacy, provenance, storage costs, source independence, version tracking, conflicting reports, and determining appropriate confidence levels. I am not suggesting that crowdsourced evidence should automatically be accepted as truth. The purpose would instead be to make consented, evidence-backed information available with transparent uncertainty and provenance.
Even when something cannot be confirmed with certainty, saying that multiple independent pieces of evidence support a conclusion—and clearly stating its limitations—could sometimes be considerably more useful than having no information available at all.
Thank you for considering the idea.
Disclosure: This proposal is based on my own idea and a discussion I had with ChatGPT while developing it. ChatGPT assisted me with organizing and wording the final proposal, and I reviewed and approved the content before submitting it.
Note: I found an older feature request proposing community-provided information after drafting this idea. This proposal expands on that general concept by focusing specifically on evidence-backed contributions, provenance, uncertainty, conflicting evidence, privacy and consent, and version-aware verification.