Journal entry review with AI: a second-look prompt for risky entries

When to use it

Use it as a second look before close sign-off, or to analyze the whole quarter's JE population the way an auditor will. It doesn't replace your JE approval control. It makes sure the reviewer looks hardest at the entries that deserve it.

Input data

• JE population for the period: JE number, line, posting date, effective date, created by, approved by, source (manual/system), account, amount, description.
• User list with roles (who is expected to post what).
• Chart of accounts with account type.
• Your approval matrix (thresholds by amount and entry type).
• Optional: support documents for a selected sample of entries.

The prompt

Replace the [bracketed] items. Paste it with your files attached. It works in any capable assistant; tool-specific notes are in the steps.

Help me screen the journal entry population for [period] before close sign-off. I will review flagged entries myself.

Attached: JE detail [file]; user roles [file]; chart of accounts [file]; approval matrix: [paste].

1. Completeness check: number of entries and lines, and confirm each JE balances (debits = credits). Report any that don't.
2. Flag entries with any of these characteristics (list every flag that applies):
   - Manual entries (source = manual) over $[X]
   - Posted after period-end / after [close day], or effective-dated into a closed period
   - Created by users who don't normally post entries, or created and approved by the same person
   - Missing approval, or approval not consistent with the approval matrix
   - Round amounts (e.g. ending in 000) over $[Y]
   - Seldom-used accounts (used fewer than [3] times in the last 12 months) or unusual account combinations (e.g. revenue debited against cash or an accrual)
   - Weekend or holiday postings
   - Vague descriptions ("adjust", "per [name]", "reclass" without detail)
   - Entries that reverse or offset within a few days
3. Rank flagged entries by number of flags and amount. For the top [25], give a one-line note on why each warrants a look.
4. If support documents are attached for an entry, compare the support amount, period and account logic to the entry and note any inconsistency.

Do not conclude that any entry is appropriate or inappropriate. Present flags and facts only.

Steps

1. Export the full JE population from the ERP, and confirm the line count and totals match the trial balance activity.
2. Run the prompt. Check the completeness section first.
3. The reviewer examines each top-ranked entry against its support in the ERP. AI comments are pointers.
4. Escalate anything unsupported or outside the approval matrix per your policy (correct, re-approve or reverse through the normal process).
5. Keep the flagged list and review notes with the close binder.

Control map

What the AI may do, step by step. How ratings work.

Audit trail

• Retain the population export (with record count and control totals), the criteria used, the flagged list and the reviewer's disposition of each flagged entry.
• Approval evidence stays in the ERP or JE tool.
• If you use this in support of a SOX control, document it in the control description and agree it with your auditors before relying on it.

Failure modes

• An incomplete population (filtered export, or missing system entries) makes the whole screen meaningless. Hence the completeness check.
• Too many flags. Thresholds set too low produce hundreds of entries nobody reviews. Calibrate on one month first.
• The AI "clearing" entries with generic reasoning ("appears to be a routine accrual"). The prompt forbids conclusions; enforce it.
• Approval-matrix logic misread for multi-line entries. Spot-check a few against the ERP's workflow history.

Data handling

• Use a company-approved AI workspace whose terms say your inputs are not used for model training. Check your plan; don't assume it.
• Strip or mask what the task doesn't need: employee names, bank account numbers, customer PII, SSNs or TINs.
• If you are a public company, treat pre-release results as material non-public information and keep them inside approved systems.