Product documentation
Ask better questions
Good questions make the intended subject, measure, scope, and output clear. They do not guarantee a correct answer, but they reduce avoidable ambiguity.
Include the necessary scope
For analytical questions, specify what matters:
- Subject: product, customer group, document, table, or business entity.
- Measure: revenue, count, rate, balance, or another defined metric.
- Time: exact date range and, when relevant, timezone.
- Filters: region, status, channel, category, or other boundaries.
- Comparison: baseline, previous period, target, or peer group.
- Output: short answer, table, chart, ranked list, or file.
For example, prefer “Compare monthly net revenue for Region A and Region B from January through June 2026, using CNY, and show a table plus a line chart” over “How is revenue?”
Make the answer verifiable
Ask the Agent to state the applied definitions, filters, and time range. When source-backed evidence matters, request citations—but remember that citation availability still depends on the Agent and source.
Break down complex work
Use one Session for a coherent line of work. Start with the main question, verify the first answer, then ask focused follow-ups. If the Agent asks for clarification, answer the missing detail instead of opening a duplicate Session.
Avoid unsafe or misleading prompts
- Do not include passwords, access tokens, or unnecessary personal information.
- Do not ask the Agent to guess missing values and present them as facts.
- Do not combine unrelated analyses into one vague request.
- Do not rely on polished wording without checking evidence, units, and scope.