LearnGrokNothing here is financial advice, and none of it replaces your own checks.
Use these prompts to test whether a committee pack says what it needs to say, and whether the evidence supplied supports it. They are for finance directors and company secretariat teams preparing papers for review. Nothing here is financial advice, and none of it replaces your own checks.
Start with the complete pack where possible. A challenge log is only useful if it points the paper owner to the page, number, action or schedule that needs attention.
Key point
Review the evidence trail
Ask the model to trace each material claim back to a schedule, action or authority. A well-written narrative is not evidence by itself.
1. Assemble one review set
Before you paste anything, collect the documents that establish what the committee is being asked to consider:
- The current draft board or committee pack, including appendices.
- Prior minutes, action log and decision record.
- The management accounts, forecast, KPI dashboard, cash flow, reconciliation or other supporting schedules cited in the pack.
- The relevant terms of reference or delegated-authority extract.
- An assumption register or model output where the paper relies on forecasts or scenarios.
Give each item a clear label in the prompt. Include page numbers if the source material has them. If a schedule is absent, say so. Do not ask the model to fill the gap from general knowledge.
Stop
Do not paste confidential material into a tool unless your organisation has approved that use.
Apply your information-handling policy first, including any redaction, access and retention requirements.
The service behaviour and available features can vary. Check the relevant documentation before setting a repeatable process: xAI documentation overview.
2. Choose the prompt that matches the review stage
Use the first prompt to create the broad challenge log. It is the best starting point when you have not yet tested the pack against its evidence.
Use the other prompts to deepen a specific area rather than asking one very large question to do everything.
| If you need to check | Use this prompt | Expected output |
|---|---|---|
| The whole pack against actions and schedules | Build the first challenge log | Prioritised issue register |
| Forecast or scenario assumptions | Trace assumptions to evidence | Assumption-to-source table |
| Numbers repeated across the pack | Reconcile key figures | Figure discrepancy log |
| What the committee is actually being asked to do | Test committee decisions | Decision-readiness log |
| A response list for authors | Prepare owner questions | Concise query tracker |
Run the broad review first. Then use the figure, assumption or decision prompt only for sections that are material or unclear. This keeps the follow-up work focused and makes the resulting questions easier for the paper owner to answer.
3. Read the log as a review tool, not a verdict
A challenge log should separate four different problems:
- Missing evidence: the pack makes a claim but no source schedule, action update or authority is supplied.
- Inconsistent figure: the same measure differs between pages or between the pack and a source.
- Unsupported assumption: a forecast or scenario depends on a basis that is not stated, sourced or reconciled.
- Unclear decision: the committee cannot tell what it is being asked to approve, note or endorse.
Treat each row as a question for the owner. Do not label a point as an error unless the supplied documents demonstrate it. A mismatch may result from timing, scope, currency, accounting basis, rounding or a valid update after the source schedule was produced. The owner needs to explain which applies.
Watch out
A repeated number is not necessarily reconciled.
Check the reporting period, entity, currency, units and whether the figure is actual, forecast, budget or scenario before accepting a match.
Prioritise High items where a missing answer could prevent a valid decision, misstate a material matter, leave a prior action unresolved, or make a key figure impossible to reconcile. Use Medium for significant clarity or evidence gaps. Use Low for drafting and presentation points that do not change the committee's understanding.
4. Check the output before sending it on
The model can miss content, combine distinct issues or attach a challenge to the wrong source. Check the log against the documents before you issue it.
Start with every High item. Open the cited pack page and schedule. Confirm that the quote, figure and period are exact. Then check that the question asks for something the owner can provide.
Check
The log is ready when each High item has a location, an evidence statement and one answerable owner question.
If any of those is absent, rewrite the row before circulation.
Look for these warning signs:
| If you see this | Check this | Do this |
|---|---|---|
| A figure is called inconsistent | Units, period, entity and reporting basis | Confirm the comparison is like-for-like |
| A prior action is marked overdue | Minutes, due date and later updates | Check whether it was closed elsewhere |
| A decision is marked unclear | Exact recommendation and authority | Ask for replacement decision wording |
| Evidence is said to be absent | Appendices and referenced papers | Locate the source or request it explicitly |
Keep the final challenge log factual. Remove speculative explanations. Keep source references even where the issue is resolved, because they show the review trail.
5. When the review does not work
If the output is too general, paste fewer documents and name the specific pack sections and schedules to compare. If it invents a source, rerun the prompt with the instruction to use only supplied documents and to write "Not evidenced in supplied documents" where necessary.
If the log is too long, ask for High items first, then run a second review for Medium and Low items. If page references are unreliable, add page labels to the pasted text or ask for section headings instead. Where a decision or accounting treatment remains uncertain, return it to the paper owner and the appropriate qualified reviewer rather than relying on the generated log.
