Alright, let’s see how this actually works in practice. Everyone’s raving about AI summaries, but I’ve found most pre-built templates are either too generic or try to do too much. I wanted something lean that actually forces the AI to be useful, not just decorative.
My goal was a meeting notes template that:
* Captures raw notes and action items in a standard format.
* Uses Notion AI to generate a *concise* summary, but only from a dedicated "Key Points" field I manually populate post-meeting. This avoids the AI hallucinating from irrelevant chatter.
* Auto-formats follow-ups with assignees and dates.
Here’s the basic property and block setup I landed on:
**Properties:**
- Meeting Title (Title property)
- Date (Date property)
- Attendees (Multi-select)
- Key Points (Text field — this is the only field the AI summarizes from)
- AI Summary (Text field, populated by the AI)
- Actions (Relation to a separate Actions database)
**Template Blocks:**
1. **Agenda & Objective** (Simple bulleted list)
2. **Raw Notes** (A toggle with a blank page inside for freeform notetaking)
3. **Key Points** (That dedicated text property, displayed as a paragraph. I force myself to distill 3-5 bullet points here *after* the meeting.)
4. **AI Summary** (The text property where I trigger "Summarize" from the `Key Points` text.)
5. **Action Items** (A linked view of the Actions database, filtered to this meeting.)
The trick is the self-imposed discipline: the AI only gets my curated `Key Points`. I’ve compared letting it summarize the entire `Raw Notes` field, and the output is consistently worse—it gets lost in tangents. This way, it’s a sanity check on my own distillation, not a magic wand.
Has anyone else tried constraining the AI’s input source like this? The vendor claims of “understanding context” are fine, but I find giving it less context actually produces more reliable results. Prove me wrong.
Really smart to have a dedicated Key Points field as the AI's only source. I did something similar but I use a checkbox property called "Ready for Summary" that I tick only after I've cleaned up my raw notes. It stops me from accidentally summarizing half-baked thoughts.
How do you handle the trigger for the AI summary? Do you run it manually, or did you find a way to automate it?
Not sponsored, just curious
That's a clean setup, and I think the dedicated Key Points field is the key move here. Most people trust the AI to skim the whole raw note blob, which leads to exactly the hallucination problem you mentioned. Forcing yourself to distill 3-5 bullet points before hitting summarize is a great discipline check too.
One caveat though: I've seen some teams over-index on that constraint and end up with summaries that miss important nuance because the Key Points were written too narrowly. A possible middle ground: keep your Key Points field for the AI source, but also maintain a separate "Context" text block (not fed to AI) that you can glance at when reviewing the summary for sanity. That way the AI stays on a short leash but you don't lose the bigger picture.
What do you do for recurring meetings - do you duplicate the template each time or set up a database view that resets the fields?
Stay factual, stay helpful.