10 Prompts Every CPA Should Have Saved (With Examples)

Most CPAs are using AI badly because they're prompting badly. Not because they're not smart. Because the prompt patterns that work for marketing don't transfer.

Here are 10 prompts I use or have given to client CPAs that earn their keep every week. Copy them. Tweak the placeholders. Use them.

1. IRS notice classifier

Use case: triage incoming notices before staff time goes into them.

``` You are an IRS notice classifier for a CPA practice.

Given the text of an IRS notice, return JSON with: - notice_type: one of [refund_held, balance_due, audit_initiation, math_error, identity_verification, info_request, penalty_assessment, levy_warning, other] - severity: one of [routine, attention_needed, urgent] - recommended_action: one sentence - estimated_response_time_days: integer - partner_review_required: true/false - confidence: 0.0 to 1.0

If confidence is below 0.85, set notice_type to "needs_human_review".

Notice text: {NOTICE_TEXT} ```

Run this on every incoming notice. Route by severity automatically. Cuts triage time per notice from 15 minutes to under 30 seconds.

2. Document chase reminder

Use case: nudge clients for missing documents without sounding like a robot.

``` You are drafting a polite document reminder for a CPA client.

Client name: {CLIENT_NAME} Missing documents: {DOC_LIST} Days since engagement opened: {DAYS} Client tier: {TIER (1, 2, or 3)} Prior year filing complexity: {COMPLEXITY (simple/moderate/complex)}

Write a 4-6 sentence email reminder. Match the tone to the tier: - Tier 1: warm, personal, references something specific from prior year - Tier 2: friendly, professional, direct - Tier 3: brief, businesslike, clear deadline

End with a single CTA: upload via secure portal. Do not list the documents in the email body — link to the checklist page instead. ```

Tier 3 emails auto-send. Tier 1 and 2 route to the partner's inbox as drafts.

3. Engagement letter pre-flight

Use case: catch scope issues before the engagement letter is signed.

``` You are reviewing an engagement letter draft for scope clarity.

Engagement letter text: {LETTER_TEXT} Prior year scope and fees (if any): {PRIOR_YEAR_DETAILS} Client's intake form answers: {INTAKE_DATA}

Identify: 1. Any service mentioned in the intake but not in the engagement letter 2. Any service in the letter that the client didn't specifically request 3. Any fee discrepancy from prior year that isn't explained in the letter 4. Any state filing implied by intake (e.g. moved states) not addressed 5. Any business entity in intake but no business return scoped

Return as a bulleted list of "potential issue: ..." with each item under 20 words. Then a one-sentence recommendation: ready / needs revision / partner review. ```

This catches the engagement-letter mismatches that cost firms days of unbilled work mid-busy-season.

4. Client status update

Use case: weekly automated status without the partner writing each one.

``` Write a 3-paragraph status update for a tax client.

Client: {CLIENT_NAME} Current return status: {STATUS (intake/in-prep/in-review/ready-to-sign)} Documents outstanding: {OUTSTANDING_DOCS} Next milestone: {NEXT_STEP} ETA to next milestone: {ETA}

Write in a calm, professional tone. Avoid jargon. Avoid promising specific dates if ETA confidence is low. End with one CTA appropriate to current status (upload docs / no action needed / schedule signing).

Sign as: {PARTNER_NAME}, on behalf of {FIRM_NAME} ```

Partners review the drafts. Most send verbatim. The few that need editing edit faster than they would have written from scratch.

5. Quick tax research helper

Use case: pre-research a tax position before billable hours hit it.

``` You are researching a tax position for a CPA. You are not providing tax advice — you are surfacing the issues for the human CPA to verify.

Position: {POSITION_STATEMENT} Tax year: {YEAR} Jurisdiction: {STATE_OR_FEDERAL}

Return: 1. The 2-3 most relevant code sections or regs 2. Any recent (last 24 months) revenue ruling, PLR, or court case that touches this position 3. The strongest counterargument to this position 4. A short risk assessment (low/medium/high) on the position 5. What additional facts you'd want to verify before signing off

Cite specific code sections, not general topics. ```

Always cross-checked against Lexis or CCH before any decision. The prompt is for orientation, not authority.

6. K-1 explanation for client

Use case: explain confusing K-1 items in plain language without writing the same email 40 times.

``` You are explaining a K-1 item to a non-CPA client.

K-1 box and line: {BOX_NUMBER} Amount: {AMOUNT} Pass-through entity type: {S-CORP, PARTNERSHIP, ETC} Client's likely tax situation: {SITUATION_NOTES}

Write a 2-paragraph explanation: - What this line means in plain language - How it likely affects their personal return - Whether they need to do anything

Avoid tax-code citations unless absolutely needed. Aim for 8th-grade reading level. End with: "I'm happy to walk through this on a quick call if it helps." ```

The "happy to walk through" line drives a 6x increase in client calls (which become billable) at this practice.

7. Busy season triage prompt

Use case: prioritize the daily backlog without the partner spending an hour deciding what to work on.

``` You are triaging today's task list for a tax-prep partner.

Open returns and their status: {RETURNS_LIST} Open client questions: {QUESTION_LIST} Partner's strengths and preferences: {NOTES} Filing deadlines in next 14 days: {DEADLINE_LIST}

Produce a prioritized list of 5-7 tasks for today with: - Task summary - Estimated time (in 15-min increments) - Why this is today's priority - Risk if deferred to tomorrow

Sort by deadline pressure first, then by client tier, then by simplest-first when the day is otherwise full. ```

Used at 7 AM. Saves 30 minutes of "what should I work on first" thrash.

8. Drake Tax data verification

Use case: catch data-entry errors before the return is filed.

``` You are verifying tax-return data for inconsistencies.

Source documents summary: {DOC_SUMMARY} Drake Tax entries summary: {DRAKE_ENTRIES} Prior year final return: {PRIOR_YEAR_DATA}

Flag any of: 1. Amount in Drake that doesn't match source doc (within $10 tolerance) 2. Material change from prior year without obvious explanation 3. Missing schedule that prior year had 4. State implications of any new income source 5. Carryforward items not pulled from prior year

For each flag, include: severity (low/med/high), source doc line, Drake field, recommended action. ```

The amount mismatch check alone catches 1-2 errors per return on average.

9. Year-end planning conversation

Use case: prep a personalized year-end planning email by November 1.

``` You are drafting a year-end tax planning outreach for a client.

Client profile: {PROFILE} This year's notable events (from intake or notes): {EVENTS} Carryforward items: {CARRYFORWARD} Likely strategies based on profile: {STRATEGIES}

Write a 4-paragraph email: 1. Open with reference to a specific year event (sold a property, started a business, etc.) 2. Highlight the 2 most impactful planning moves for their situation, briefly 3. Mention 1 deadline-driven action (contribute to retirement, distribute, etc.) 4. CTA: 30-minute planning call before Dec 15.

Avoid lists of generic strategies. Tie each suggestion to their actual situation. ```

The personal reference in paragraph 1 is the entire game. Without it, clients ignore. With it, response rates are 3-4x.

10. New-client onboarding script

Use case: capture the right intake info from a new client without 4 rounds of follow-up.

``` You are interviewing a new tax-prep client. Generate the next question to ask.

Conversation history so far: {HISTORY} Information still missing: {GAPS} Engagement type: {INDIVIDUAL or BUSINESS or BOTH}

Return: - One specific question to ask next - Why this question is the next priority - A range of likely answers and what each would mean for scope

Stop generating after the question. Wait for client response before next prompt. ```

Run this conversationally in a chatbot or a Notion form. Cuts onboarding from 3-4 email rounds to 1 conversation.

The pattern across all 10

Every working prompt has the same shape: - Role assigned (you are X doing Y) - Inputs explicitly named with placeholders - Output format specified (JSON, paragraphs, list) - Constraints and edge cases included - Confidence handling for the ones that need it

The bad prompts are 1 sentence. The good prompts are 8-15 lines. The extra structure earns its keep in consistency.

What to do with these

Don't paste these into ChatGPT and stop. Build them into your workflow.

Use Claude (or your model of choice) via API with these as system prompts. Wire them to your CRM, your practice management software, your secure portal. Let them run on inputs as inputs arrive.

The prompts are 5% of the work. The plumbing is 95%. If you're a non-technical CPA, the plumbing is what to outsource. Not the prompts.

The prompts are the cheap part. The system around them is the moat.