Once Thoughts is connected, the most useful questions you can ask ChatGPT or Claude are about your own life: What did I decide about the cabinets? What has Sam added this week? Where is that thought about pricing? Your assistant can search your synced thoughts, read one in full, summarise a project and show you recent changes — turning a pile of captured notes into answers. This guide collects the question patterns that work best, and the small habits that make the answers trustworthy.
Start with orientation questions
If you have been away from a project — or your own notes — begin broad:
Show me my recent thoughts.
What projects do I have, and which were shared with me?
What is my Autumn launch project about?
The project summary is a genuinely good way back into work you half-remember. Ask the assistant to follow it with the project's most recent thoughts, and you have reconstructed your context in under a minute — the same move you would make by opening Thoughts and scanning, except the assistant can then reason with what it found.
Find a specific thought
Search in Thoughts matches the literal words in titles, text and tags. It is fast and predictable, but it is not semantic search: “that thing about money” will not find a note that says “budget”. Help it with distinctive words you actually wrote:
Search my Thoughts for “risotto” in Recipes and read the matching thought in full.
Find thoughts mentioning “boiler” and tell me what the engineer quoted.
Two refinements improve results a lot. Name the project when you know it, to narrow the field. And ask for the full text before conclusions — previews can truncate, and an assistant summarising a preview will confidently miss the second half. “Read the whole thing, then answer” is a phrase worth keeping.
Ask what changed
Recent-activity questions are where a connected notepad beats manual browsing, especially in shared projects:
What has changed in my projects in the last week?
Show me thoughts Sam added, and anything Sam last edited.
Do I have any unread invitations or new shared thoughts?
Understand what “recent” means here: activity shows each thought's latest visible state in the window, ordered by when the server received it. A thought captured offline on Tuesday can appear when it syncs on Thursday. It is a catch-up feed, not an audit log — it will not show you deleted items or a history of every edit. Used with that in mind, it is the fastest stand-up meeting you will ever have. For shared work specifically, see catching up on a shared project with AI.
Make the assistant work with what it found
Retrieval is step one; the payoff is what happens next in the same conversation:
Read my three most recent Launch thoughts and draft this week's update from them.
Compare the two cabinet quotes in Kitchen renovation and list what is missing from each.
Turn those next steps into a checklist and save it back as a new thought in Launch.
That last move closes the loop described in a notepad for ChatGPT: read from the shelf, think, write back to the shelf. The new thought sits next to the sources it came from, in the project where the work lives.
Check answers the way you would check a colleague's summary
An assistant reading your notes can still misread them. Ask it to quote the relevant line when a figure matters. If it says “you decided X”, ask which thought says so. Authorship and last-editorship are tracked separately in Thoughts, so “Sam added” and “Sam last edited” are different questions with different answers — the connector keeps them distinct, and so should your prompts.
Duplicate names are another trap. If you have two projects called “Trip” or two collaborators called Sam, a good assistant will show you both matches rather than silently picking the first. If an answer feels off, ask: “Which project did you read that from?”
Know the edges
Your assistant can only ask about what it can reach: synced, unlocked thoughts in projects you own or belong to, plus synced Unfiled thoughts. Local-only projects, locked thoughts and unsynced captures do not exist as far as it is concerned — ask about a local journal and it will honestly find nothing. It reads thought text, not recordings or attachment files. And it cannot edit or delete anything; corrections happen in Thoughts, by you. The full boundary is spelled out in what MCP means for your notes, and setup takes minutes following the ChatGPT or Claude guides.
Try one question today that only your notes can answer. That is the moment a notes app stops being storage and starts being memory you can interrogate.
Frequently asked questions
What can I ask ChatGPT or Claude about my notes?
You can ask for recent thoughts, a project summary, a search across titles, text and tags, the full text of one thought, recent changes, who added or last edited something, and whether you have unread invitations or notifications.
Why can't the assistant find a note I know exists?
Usually the note is local-only, locked or not yet synced, or your search words do not match its text. Thoughts search is literal word matching, so try distinctive words from the note and name the project.
How do I get accurate answers from my notes?
Name the project, use distinctive search words, ask the assistant to read the full thought before answering, and ask it to quote the source line for figures or decisions you need to trust.
Can the assistant see my recordings and attachments?
No. Through the connector it reads thought text only. Voice recordings and attachment files stay out of reach, as do locked thoughts and anything you keep local.