Help/Recipes/Researcher or product manager
Research synthesis
What do we already know about it?
- For
- A researcher, product manager, or anyone about to start work on a topic the team has already discussed
- When
- Before new research, a spec or a big decision
- Takes
- A few minutes of agent time
- You get
- A one-page synthesis: the answer in brief, findings with confidence and sources, where the sources disagree, open questions, and a proposed next step
What do we already know about <question>? Synthesize it from onecell.
Save it to <cell>.
1. Search onecell broadly (several phrasings, all cells I can read) and
read the documents that matter: research, specs, reviews, decisions.
Include repo docs if the repo is here. Don't search the web unless I
ask.
2. Draft "Research: <question>" (dry run first):
- the answer in brief, in three sentences
- findings as a table: finding, confidence (high, medium, low) and
source, cited as cell · heading
- where the sources disagree, and what that suggests
- open questions, as a checklist
- the next step as a proposed decision
3. Reply with the link and the one finding that changes what we should
do next.What your agent does
Searches widely
Several phrasings across every cell you can read, because the best note is often filed somewhere unexpected.Rates what it finds
Each finding gets a confidence and a source, so you can see what is settled and what is a hunch.Looks for disagreement
Where two documents pull different ways is usually where the real decision is.Proposes a next step
As a decision you can accept, not a list of ideas.
What you get back
Open the full example →A synthesis for a made-up invoicing app, Tally. The useful part is the tension it found: interviews and usage data disagree, which changes what to test next.
Sample research synthesis: why freelancers chase payments by hand
The question
Tally can send reminders automatically, yet 60% of accounts with late invoices also chase by hand. Why?
Findings
| Finding | Confidence | Sources |
|---|---|---|
| Freelancers worry an automatic reminder will sound rude to a client they need | High | 8 of 11 interviews; 2 support threads |
| They chase by hand when the client is new, and let reminders run for regulars | Medium | 5 interviews; usage data, June |
| Many don't know reminders can be edited | Medium | Support tags, Q2; 3 interviews |
| Price is not a factor | Low | 1 interview; no contrary evidence |
Where the evidence disagrees
Interviews say reminders feel rude; usage data shows accounts with reminders on get paid six days sooner and rarely turn them off. People may dislike the idea more than the effect.
Next step
Test a reminder preview at the moment of sending (“here is what your client will get”) with 10% of new accounts for a month, and compare manual chasing.
Variations
Add outside research
Add: “Then search the web for what others have published on this, keep outside sources in a separate table with links and dates, and don't mix them with ours.”
Interview or feedback notes
Point it at a cell of interview notes and ask for themes, each with how many people said it and one short quote, instead of confidence levels.
Keep it current
Ask again after new work lands: “Update the synthesis, mark what changed since last time, and keep the old findings dated.”
Tips
- Low confidence is useful information. Don't let the agent drop it.
- Ask for the tension explicitly. Agents tend to smooth disagreements over.
- A synthesis is only as good as what the team wrote down. Gaps it finds are worth filling.