What should I include in a prompt for an AI journal review?
By Chart Reporter Reviewed 2 min read
The short answer
Give the assistant a bounded review task: verify account identity, define dates and timezone, state the metric and units, check sample completeness, and separate observations from hypotheses. Ask for supporting records and missing-data notices. Keep credentials and unnecessary private history out of the prompt.
Specify the question and its scope
A useful prompt describes the intended account or hub, date interval, timezone, completed-trade basis and the grouping to compare. Use account labels carefully and have the assistant verify identity through the authorised connection. Do not supply a broker password or reporting key in ordinary chat.
For example: review completed trades for the selected hub over a named calendar window, grouped by symbol, using net P&L in the confirmed account currency. Ask for counts and component sums. That defines a descriptive task without asking the assistant to choose a future trade or infer your intentions.
Add evidence requirements
Tell the assistant to check pagination and report the loaded and eligible counts. Ask it to preserve unavailable metrics rather than replacing them with zero. Require it to identify the timestamp basis of the tool filters and state any mismatch with the requested calendar convention.
For every main finding, request supporting trade identifiers and a short explanation of the calculation. If chart context or notes are needed, first verify that the current tools expose them. Chart Reporter MCP does not currently provide screenshot bytes or journal-note text, so the assistant must not claim to have inspected those automatically.
Ask for a useful report structure
Request five parts: scope and coverage, descriptive findings, supporting examples, uncertainties, and a specific follow-up question. Keep observed facts distinct from possible explanations. If a query fails or the data cannot answer the question, ask for a clear account of that limitation.
Save the prompt and tool settings with the report so the review can be repeated on a later sample. A more elaborate prompt cannot repair missing history, incompatible currencies or an ambiguous definition of R. Those are evidence and scope problems to resolve explicitly before expecting a reliable comparison.
Sources and further reading
Related guides
What evidence should I ask for when evaluating an automatic MT5 journal?
Compare capture requirements, reconciliation, context, privacy boundaries and exportable evidence.
What questions can an AI assistant answer from a trading journal?
Match review questions to available trade records, scope and evidence instead of expecting predictions from history.