The knowledge cutoff, when tools help, and how to cite.
Key takeaways
- After the cutoff, the model needs search.
- Search for the world, connectors for your data, nothing for pure text work.
- Primary source, open the link, cite the source.
The knowledge cutoff
A model's knowledge stops at a date. The world doesn't.
A model learns from data up to its knowledge cutoff. On its own, it doesn't know what happened after that, and it may not notice that it doesn't know.
For anything recent or changing, like rules, prices, versions or news, it needs a tool such as web search.
Remember: After the cutoff, the model needs search.
Tools and connectors
Search, files, calendars: tools let the model fetch instead of guess.
A tool is something the assistant can call: web search, a calculator or code runner, or a connector to your drive, email or calendar. Rewriting a paragraph needs no tool at all.
The app runs the tool and hands the result back, so the answer is only as good as what the tool found. Connectors use your access, so connect only what the task needs.
Remember: Search for the world, connectors for your data, nothing for pure text work.
Cite and check sources
A link is not proof. Open it.
Ask for sources with every claim, and prefer primary sources: the official site, the law itself, the original study.
Open the link and check that the page says what the answer claims; answers sometimes cite a real page that says something else.
When you pass it on, cite the original source, not 'the AI said'.
Remember: Primary source, open the link, cite the source.