This guide gives you a reusable starting pattern. It is designed to help you see the work more clearly; it is not a substitute for judgment, source checking, or responsibility for the result.
Prepare the inputs before you ask for output.
The model only sees what you give it. Spend a few minutes naming the reader, desired result, and uncertain information. This makes a first draft easier to assess and reduces the need for decorative rewriting later.
Give the task a useful brief.
Build an evidence matrix from the labeled notes below. Decision question: [what must be decided] Audience: [who will review it] Source notes: [paste each note with source label and date] Create a table with these columns: claim, supporting source, source date, evidence type, strength of support, counter-evidence or limitation, confidence, and next verification step. Keep claims separate instead of merging similar statements. If a claim is not supported by the supplied notes, mark it “Not supported in supplied notes.” Do not invent facts, citations, dates, owners, or confidence. After the table, list the three claims most likely to change the decision and explain what would verify them.
Replace every bracketed field with your real context. Read the output before reuse.
Four steps that keep the result usable.
- 1
State the decision question before collecting claims.
- 2
Label every note with its source and date.
- 3
Ask AI to keep claims, support, and limitations in separate columns.
- 4
Review the highest-impact gaps against the original sources before recommending an option.
Use a matrix when a summary would hide the gaps
A summary is designed to read smoothly. An evidence matrix is designed to be challenged. It keeps a claim beside the source that supports it, the limit that weakens it, and the check that could change your view. That makes it useful before a decision memo or recommendation.
Separate evidence type from evidence strength
A direct measurement, a participant observation, a reported opinion, and an assumption are not interchangeable. Ask the model to name the evidence type first, then describe how strongly the supplied record supports the claim. This avoids turning a neat label into false precision.
Prioritize gaps by decision impact
Not every missing detail deserves the same research effort. Mark the claims that could reverse the preferred option, change a constraint, or alter the responsible owner. Verify those first; leave low-impact uncertainty visible rather than spending time making the whole table look complete.
Keep the original record close
The matrix is a navigation layer, not a replacement for the source. Preserve short source labels and dates, open the original record for important claims, and let the decision owner correct the matrix before it becomes part of a formal recommendation.
A worked example: claims about a competitor feature
Notes claim a competitor “ships export in Q3”. The matrix rows might be: claim, source (“Product blog, 05 Aug”), evidence type (reported announcement), strength (medium — no confirmed release date), and counter-evidence (a customer Q&A saying the timeline is not firm). The decision owner can then see that the claim changes a scope decision only if confirmed, and the next verification step is named.
Know when the matrix is overkill
If the decision is small and the sources are few, a matrix adds ceremony rather than clarity. Use it when a summary would hide the gaps — several claims, mixed evidence, or a choice that depends on which source to trust. When the record is thin, the matrix should say so in the confidence column instead of looking complete.
Run a human check.
- Does every material claim have a source label or a visible unsupported status?
- Are evidence type, strength, confidence, and limitation kept distinct?
- Which three gaps could change the decision, and is each verification step specific?
- Has a human reviewed the original sources before the matrix supports a recommendation?
- Did the draft avoid inventing citations, dates, owners, or certainty?
AI is strongest here when it makes missing information, structure, and options easier to see. The moment an output becomes a claim, commitment, or decision, bring a person back into the loop.