Considerations
Source quality shapes the result
Generated outputs and chat responses depend on the scope and quality of the selected sources. If key material is missing, or if unrelated material is mixed into the same source set, the result may be limited or unfocused.
Check readiness first
Sources can appear as pending, processing, ready, or error. In general, generation and chat work best when the source is in the ready state. Including sources that are still processing can lead to missing context or failed generation requests.
Files, URLs, and web research can fail
- Upload can be rejected if the file type is unsupported or if the extension does not match the actual file contents.
- The current default upload limit is 50 MB per file.
- URL extraction can fail because of extraction issues, login requirements, access restrictions, or blocking policies.
- Crawl previews can expire over time and may need to be run again.
- Web search and deep research depend on the quality of external results and accessible pages.
Keep limits in mind
The current default limit is 10 sources per notebook. When adding pages through crawl or web search, select only the material you need. In crawl preview, high depth or breadth values can increase processing time.
Verify important content with citations
Before using the output for important decisions, review the original material and the citation badges together. If freshness matters, also check when the material was written. Generated content includes summarization and restructuring, so it may not match the original wording exactly.
Handle sensitive material according to policy
Confirm whether you can upload sensitive material such as personal data, customer information, contract content, or security documents under your organizationβs policy. When creating outputs for external sharing, also check that sensitive details from the source do not remain in the result.