Module 3, Writing, Memos & Internal Deliverables

⚡ The 30-second version. Use it for: SOPs, internal memos, reports, executive summaries, meeting notes, and checklists. The method: give the AI your bullet points and a structure; it returns a clean draft you finish in minutes, and it never invents facts or numbers. Start here: the 15-minute starter at the bottom.

This is the "everything else I have to write" bucket, the writing that isn't a client email (Module 1) and isn't tax-law research (Module 2): SOPs, internal memos, reports and report narratives, executive summaries, meeting notes, training material, agendas, proposals, checklists, and firm content. High-volume, time-consuming, and the fastest place to feel AI earn its keep.

The one-sentence method: You bring the substance and the judgment; AI brings the structure and the sentences. You never outsource what only a CPA can decide, you outsource the blank page.

What "good" looks like

The win isn't "AI writes my report." It's "I hand AI my rough inputs and a structure, and it returns a clean draft I refine in minutes instead of writing from scratch." Three jobs AI does reliably here:

Draft from bullets, turn your notes or an outline into a polished SOP, memo, or narrative.
Restructure & summarize, compress a long input (a meeting transcript, prior memos, GL notes) into a tight summary or executive summary.
Standardize, apply a consistent format and tone; build reusable templates, checklists, and SOPs once and reuse them forever.

What AI is not doing: forming the professional conclusion, supplying the numbers, or vouching for accuracy. That's you.

The lever: feed it structure + your real inputs + your conclusion

Bland, generic output is the #1 complaint, and it's always the same cause: the model had no substance to work with, so it filled the gap with boilerplate. Fix it by giving AI three things:

A structure (your template, or ask it to propose one you approve).
Your real inputs (the bullet points, the facts, the figures, scrubbed of client identifiers; see guardrails).
Your conclusion / intent (what the deliverable should say or recommend, you decide this).

Then AI does what it's good at: turning that into clean, consistent prose.

Set a reusable "deliverable standards" block (do once, reuse forever)

Like a house style for documents. Paste this at the top of a drafting session, or save it as a Claude Project / custom instruction:

You help me draft internal and professional deliverables for my CPA firm. Standards:
- Audience first: state who it's for and lead with the bottom line / recommendation.
- Plain, precise, professional. Short paragraphs and headers. No filler, no hype.
- Never invent facts, figures, dates, names, or citations. If you need something I
  didn't give you, leave a clearly marked [BRACKET] for me to fill in.
- Don't state a professional-standards requirement (AR-C, AU-C, GAAP, IRC) as fact
  unless I provided it, flag it [VERIFY] instead.
- Match the structure I give you; if I don't give one, propose an outline first and
  wait for my OK before drafting.
Give me the draft only, ready to edit, no preamble.

That [BRACKET] / [VERIFY] instruction is what keeps a deliverable that goes out under your name from carrying confident, fabricated content.

Copy-paste prompt library

1. SOP from a process you can describe

Draft a Standard Operating Procedure for [process, e.g., monthly bank reconciliation for a
bookkeeping client]. I'll describe how we do it: [your steps, rough]. Produce a clean SOP with
purpose, who's responsible, step-by-step instructions, and a final review/sign-off step. Flag
anything I left ambiguous with [BRACKET].

2. Meeting notes → summary + action items

Here are my raw meeting notes / the transcript: [paste, scrub client identifiers]. Give me:
(1) a 4-6 line summary, (2) decisions made, (3) action items with owner and due date where
stated. Don't infer commitments that aren't in the notes.

3. Bullets → report narrative / management letter

Turn these points into a [management letter / report narrative / cover memo]: [bullets].
Audience: [client management / partner / staff]. Lead with the key message. Keep it factual, do not add findings, figures, or recommendations I didn't give you.

4. Executive summary of a long document

Summarize this [document] for a [busy partner / client] in [N] bullets, leading with the single
most important takeaway: [paste document]. Note any open questions or items needing follow-up.

5. Build a reusable template / checklist

Create a reusable [engagement checklist / workpaper index / client-onboarding checklist] for
[engagement type]. Make it generic and fill-in-the-blank so I can reuse it across clients.

6. Tighten and clarify an existing draft

Here's my draft: [paste]. Keep my meaning and all my facts/numbers exactly, but make it tighter,
clearer, and consistently formatted. Don't add new claims.

Tool picks

Claude (Lab default), best for long-context summarizing (feed it a whole transcript or several prior memos) and for tone/structure control. Use Projects to store your deliverable standards + firm templates so every draft starts on-brand.
Copilot in Word / Google Docs Gemini, lowest friction when you're already in the document; good for first drafts and tightening, weaker on structure control.
A saved deliverable-standards block beats a fancier tool you have to leave your workflow for. Friction kills reuse.

Guardrails

Internal deliverables are sneaky on data and accuracy, read these before you draft.

Client data hides in deliverables. Reports, management letters, and memos often carry client financials and names. Scrub identifiers or use a firm-approved tool, the redactor tool works here too, not just for tax research. The data rules in Guardrails and Regulatory Foundation (§7216, FTC Safeguards/WISP, AICPA Confidentiality Rule) apply to any deliverable you paste.
Never let it invent facts, figures, or standards. A deliverable goes out under the firm's name. Use the [BRACKET] / [VERIFY] trick; check every number against your source; and confirm any AR-C / AU-C / GAAP / IRC reference against the actual standard (AI misstates these confidently). For attest deliverables, follow the report language in the standards, not AI's paraphrase.
You are the author and reviewer of record. AI drafts the prose; you own the conclusion, the numbers, and the professional judgment (SSTS §1.4; for attest work, the applicable SSARS/GAAS).
Watch for confident blandness. If a draft reads smooth but says nothing, you under-fed it, add real substance, don't accept boilerplate.

Your 15-minute starter

Pick a process you've explained out loud a hundred times (a recurring reconciliation, a close checklist, an onboarding step).
Paste the deliverable-standards block, then run Prompt 1 and describe the process in rough bullets.
Refine the draft once. You now have a reusable SOP, and a standards block worth saving for the whole firm.
Next time you face a blank report or memo, your instinct becomes "give AI my bullets and structure," not "stare at the cursor."

Win condition: the blank-page tax disappears for every internal document, and your firm starts accumulating a library of clean, consistent SOPs and templates as a byproduct.

Next module: Workflow Automation & Building Skills.