LearnGrokNothing here is financial advice, and none of it replaces your own checks.
Use this pack after results are out and before the follow-up list becomes a mixture of hunches, repeated questions and uncited claims. It is for equity research analysts, finance managers and investment teams who need a short, defensible set of management questions. Nothing here is financial advice, and none of it replaces your own checks.
The four prompts work as a sequence. Start with source evidence, then create questions, allocate the work, and audit the pack before it leaves the team.
Key point
Start with evidence, not questions
A well-phrased question is still weak if nobody can point to the release, transcript or prior commitment that caused it.
1. Assemble the source set
Collect the final versions of three documents:
- The current earnings release, including tables and guidance language.
- The current call transcript, with timestamps if available.
- Your prior-quarter notes, especially recorded commitments, open questions and forecast assumptions.
Keep document names and headings intact when pasting. A citation such as call transcript, location not supplied is better than a plausible timestamp that nobody can verify. If you are using file upload or document handling features, behaviour can vary with the model and product configuration. Check the relevant xAI documentation before relying on a particular workflow.
Run Evidence register from results materials first. It separates what management reported from what it previously said or implied. This matters most where a headline result is positive but a driver, definition or timing statement has changed.
Do not treat the register as a conclusion. It is a working record. Review every quoted figure against the supplied document, including units, period labels and whether a percentage refers to growth, margin or mix.
Watch out
Do not merge unlike measures
Revenue growth, organic growth, constant-currency growth and volume growth may all appear in the same results set. Keep their labels separate until the company provides a reconciliation.
2. Turn gaps into answerable questions
Paste the completed register into Rank management follow-up questions. The prompt produces ten questions so the team has to rank rather than merely collect issues.
A useful management question asks for one thing. It should normally seek one of these:
- A bridge between reported results and an underlying driver.
- The basis, scope or duration of a guidance assumption.
- The timing and owner of a stated operational action.
- A reconciliation of a new statement with a prior-quarter commitment.
- A definition for a KPI, adjustment or segment measure.
Reject questions that contain their own conclusion. For example, do not ask whether a margin decline shows that a programme has failed. Ask which cost categories changed, by how much, and which were expected to reverse.
The priority label is not a view on the company. It is a way to direct limited follow-up time towards points that could alter a forecast input, reporting judgement or risk discussion. A critical question should have both material decision use and unresolved evidence.
3. Give every question a route and an owner
Use Assign owners and decision use once the ranked list is stable. Supply actual team roles, even if you do not include names. The output identifies who prepares the question, where it should be raised, and the internal decision that the response informs.
This step stops a common failure: asking management for detail that no one will use. If the decision is not defined, retain the question only as a monitoring item or remove it from the immediate follow-up list.
For a board-pack query, write the decision in board language. For example, determine whether the risk section needs updating is more useful than understand risk. For research work, name the assumption or sensitivity that needs review rather than implying an investment action.
Check
The ownership plan worked if
You can read each row and know who will ask the question, what evidence would close it, and what internal document or assumption changes if the answer differs from expectations.
4. Check what comes back
Run Check the final question pack before circulation. This is not a cosmetic edit. The audit should expose citations that do not support the question, repeated issues phrased differently, and questions that cannot be answered without a vague management response.
Use this table to resolve common audit findings:
| If you see | Do this | Do not do this |
|---|---|---|
| A quote has no location | Find the page, heading or timestamp, or mark it for source re-check | Invent a location |
| Two documents give different figures | Preserve both and ask for reconciliation | Select the figure that fits your expectation |
| A question has no decision use | Move it to monitoring or remove it | Label it critical by default |
| Management language is vague | Ask for a measure, range, timing or definition | Treat the wording as a commitment |
Check numerical comparisons yourself. Confirm the denominator, reporting period, currency basis and whether management is discussing reported or adjusted results. Where prior-quarter notes record an internal interpretation rather than a direct company statement, label them as internal notes and do not cite them as management evidence.
Stop
Do not circulate unverified quotations
A misquoted transcript line can undermine the whole follow-up pack, even when the question itself is reasonable.
When the pack does not work
If the evidence register is thin, do not compensate by writing broader questions. Request the missing table, appendix, transcript section or KPI definition first. If the transcript and release conflict, retain the conflict, cite both sources and assign human review.
If the ranked list is too long, keep the questions tied to a named decision and move the rest into a dated monitoring log. If the model produces unsupported citations, rerun the audit with smaller source sections and require exact quotations. The qualified reviewer should decide what the evidence means and whether any resulting work product is fit for use.
