Research

Source-Grounded Research Brief

Produces a research brief where every claim is tagged with its source and confidence, and unsupported assertions are quarantined instead of blended into the text.

Prompt template

You are a research analyst writing a briefing for a decision-maker who will
act on it. Being wrong is far more costly than being incomplete.

## Question
{{QUESTION}}

## Sources
{{SOURCES}}

## Decision this informs
{{DECISION_CONTEXT}}

## Output structure — use these exact headings

### Bottom line
Three sentences maximum. Answer {{QUESTION}} directly. If the sources do not
support an answer, say that instead.

### Established
Claims supported by the sources. Format each as:
- Claim. [S1, S3] (confidence: high | medium | low)
Confidence is high only when two or more independent sources agree, medium
when one credible source supports it, low when it is inference from
indirect evidence.

### Contested
Points where sources disagree. State each position, who holds it, and what
would settle the disagreement. Do not average conflicting positions into a
middle view.

### Unknown
What {{QUESTION}} requires that the sources do not cover. Be specific about
what would need to be found.

### Implications for the decision
Map the findings onto {{DECISION_CONTEXT}}. Name the finding that most
changes the decision, and the assumption that would most damage it if wrong.

## Constraints
- Every claim in "Established" carries a source tag. No tag, no claim.
- Never supplement the sources with your own background knowledge. If you
  believe something relevant is missing, put it under "Unknown" as a
  question, not under "Established" as a fact.
- Do not include numbers, dates, or named quotes that are absent from the
  sources.
- Do not soften "Contested" into false consensus.
- Prefer omission to speculation.

Why this works

Two grounding techniques stack here. Mandatory per-claim citation tags make ungrounded statements structurally hard to produce, because an invented claim has no source ID to attach. Separating Established, Contested, and Unknown gives the model somewhere to put information that does not meet the evidence bar — without that outlet, weak material gets laundered into confident prose. The confidence rubric is deliberately mechanical so it can be applied consistently rather than by feel.

Example output

Example

### Bottom line
The sources support migrating the search layer, but not on the timeline in
the proposal. Two of four sources report integration work materially larger
than vendor estimates. No source addresses our data volume directly.

### Established
- Hybrid retrieval outperforms dense-only retrieval on keyword-heavy
  queries. [S1, S4] (confidence: high)
- Migration effort is commonly underestimated at the reindexing stage.
  [S2] (confidence: medium)
- Operating cost scales with index size rather than query volume on this
  architecture. [S3] (confidence: low — inferred from a pricing page)

### Contested
S1 reports reranking as essential to the quality gain; S4 reports it as
marginal. The disagreement tracks corpus type — S1 used long documents,
S4 short records. Testing on our own corpus would settle it.

### Unknown
None of the sources covers behavior above roughly ten million documents,
which is the range that matters for us.

Usage tips

Label your pasted sources [S1], [S2] and so on before running it. Numbered sources give the model a clean identifier space and make spot-checking a claim a two-second job.

The DECISION_CONTEXT field changes the output more than people expect. The same evidence briefed for “should we buy this” and “should we build this” surfaces different implications, and stating the decision keeps the brief from becoming a neutral summary nobody can act on.

Verification pass

For high-stakes briefs, run a second turn: “For each claim under Established, quote the exact sentence from the cited source that supports it. Mark any claim where you cannot produce a supporting quote.” Claims that survive that pass are the ones worth acting on.

Browse library

Competitive Landscape Matrix

Builds a comparison matrix across competitors using criteria derived from buyer needs, with explicit unknowns rather than plausible-sounding filler in empty cells.

The AI signal without the hype — new models, tools worth your time and what actually shipped. One email, no vendor pitches.

Choosing an AI stack for your team?

We help companies pick the right AI tools, wire them into existing systems and avoid the ones that quietly do not scale. Tell us what you are trying to build and we will tell you what we would use — no charge for the conversation.