Label supported findings, estimates and unresolved questions separately, and mark recommendations as editorial judgment. Readers should be able to see which parts of an AI-assisted brief come from inspected evidence and which depend on assumptions.
Use a three-tier evidence legend
Adopt this practical editorial legend:
- Tier 1 — Supported finding: an inspected source supports the precise statement, within a recorded scope.
- Tier 2 — Estimate: a calculation or projection uses stated inputs and assumptions; it is not a directly observed result.
- Tier 3 — Unresolved: evidence is missing, inaccessible, conflicting or insufficient for the intended conclusion.
These labels describe evidential status, not absolute truth. A supported finding from one small sample remains a finding about that sample. A transparent estimate can be decision-useful without becoming a fact.
Add a separate “Editorial judgment” tag to recommendations so they do not masquerade as a fourth kind of evidence.
Attach labels to individual claims
For each consequential sentence, record its label, source location, scope and review owner. Split sentences that mix categories. “The pilot saved time and will pay for itself next month” may combine an observed timing result with an untested financial forecast.
Keep labels visible in the brief rather than burying them in an appendix. For Indian budgeting decisions, specify rupees, hours and the period covered. Leave unknown costs unknown; do not silently treat them as zero.
Read a worked synthetic brief
The following example is entirely hypothetical; its records and numbers are invented for illustration.
Tier 1 — Supported finding within the synthetic scenario: A fictional public pilot note records 40 tasks taking an average of 12 minutes before a workflow change and nine minutes afterward. This establishes the note’s reported timing, not causation or future savings.
Tier 2 — Estimate: At 400 comparable tasks per month, an assumed three-minute reduction would equal 1,200 minutes, or 20 hours. Comparability and persistence are assumptions.
Tier 3 — Unresolved: Review time, error rates, access costs and whether harder tasks were excluded are unknown.
Editorial judgment: Investigate quality and review time before assigning a rupee value to the projected hours. Evaluate the brief by whether a reader can reproduce the arithmetic and identify every unverified input.
Give uncertainty an action owner
Stop a purchase or rollout recommendation when an unresolved item could reverse it. Assign a specific next step: obtain the missing cost definition, inspect the methods or measure review time using synthetic tasks.
If the next step involves the AI tools catalog, check access format and save terms before buying. Marketplace account, code or token offers are not proof of official subscription equivalence.
Sources and scope
Source material checked 29 October 2026; features/policies may change. The supplied Perplexity citations reference concerns source links for inspection. Direct fetching was blocked, so no evidence-labelling feature is claimed. The legend and synthetic brief are editorial guidance.