ChatGPT Prompts for Tax Preparers: Client Emails and Checklists

A tax preparer can use ChatGPT for client emails and checklists without any return data by writing every prompt with placeholders: [client first name], [tax year], [document list] and [deadline]. Ask for reusable templates, then merge the real details in your own practice software. No tax ID numbers, income figures or identity details ever enter the chat.

Redaction comes first because a template needs none of that data. A reminder to send documents reads the same for every client; only the name, the year and the list change. The pack below has eight prompts in four groups, each with a short code (A1, B2 and so on) so staff can find the right one in a shared file: getting documents in, deadline and filing messages, new-client messages, and checklists with an intake script. Every template carries an illustrative output and the edits it needs before staff use it.

Follow me on Instagram@sagnikteaches

The redaction-first rule, and why the pack needs no return data

Most privacy trouble in a practice comes from convenience: pasting a return "just to summarise it", or an email thread that includes a tax ID number. The way to prevent it is to design prompts that cannot need such data. Every prompt in this tutorial asks for a template. The model never learns who the client is, what they earned or what they owe.

Connect on LinkedInSagnik Bhattacharya

It still helps to know what happens to anything a colleague does paste. According to OpenAI's help page on data controls, personal ChatGPT accounts on Free, Plus or Pro start with the training setting ("Improve the model for everyone") turned on and keep it on until the user changes it, whereas OpenAI excludes Business, Enterprise and Edu workspace content from training by default. Feedback you give with thumbs up or down can still make a whole conversation eligible for training. A Temporary Chat is not used to improve models but may be kept up to 30 days for safety. None of that is a reason to paste return data. It is a reason to keep the pack placeholder-only.

Subscribe on YouTube@sagnikteaches

Write the standing rule at the top of your prompt file, and give it to every staff member and seasonal worker:

1. Placeholders only. [client first name], [spouse first name], [tax year], [list], [deadline], [fee]. No surnames, addresses, employers or account numbers.
2. No identifiers. No tax ID or national ID numbers, dates of birth, bank details or portal logins, in any format, even partial.
3. No returns or documents. No pasted forms, statements or uploaded files.
4. Merge offline. Real details go in through your practice software or mail merge, never the chat.
5. Human review. A preparer reads every email before it is sent.

For the cases where you do want AI help with a return, how to redact a return before you paste covers the technique, and what clients must sign when AI tools touch return data covers consent. The pack here deliberately avoids both questions.

The pack index: eight templates in four groups

Keep this table at the top of the shared prompt file. The code is what staff search for, the timing column tells them when the message goes out, and the last column lists the only placeholders each template needs. None of them needs a figure from a return.

Two columns earn their place. The timing column turns the pack into a season calendar, so a seasonal worker knows on day one which message goes when. The placeholder column is the redaction rule in practical form: if a template seems to need something not listed there, it is the wrong template, or the message should be written by hand from reviewed data.

Group A: getting documents in

These two templates do most of the season's work. The request goes to every client once; the reminders go only to those who have not sent everything.

A1. The document request that clients actually answer

The first email of the season sets the pace. A good one is short, gives a deadline for the client's side, and lists documents as a checklist the client can tick. A long email with a PDF of instructions gets ignored.

Write a template email asking a client for their tax documents.
Placeholders: [client first name], [tax year], [deadline for documents], [upload method], [your name], [practice name].
Include: why the deadline matters (so we can file on time without rushing), a checklist grouped as income, deductions, and life changes, and one line saying to tell us if anything is unclear.
Tone: friendly, clear, no jargon. Under 180 words. Do not mention any amounts, penalties or specific rules.

Illustrative sample output: "Hi [client first name], it is time to get your [tax year] return moving. If we receive your documents by [deadline for documents], we can prepare them calmly and file on time. Please send: Income (employment statements, interest and dividend statements, self-employment records); Deductions (receipts for charitable gifts, education costs, childcare costs); Life changes (a new home, a new job, a marriage or a new child). Upload them using [upload method]. If you are unsure whether something matters, send it anyway and we will sort it. Thanks, [your name], [practice name]."

What I would fix: the model sometimes lists specific forms by number, which varies by situation and can be wrong. The prompt bans specifics for that reason; if a form name sneaks in, check it against your own current list or remove it. Keep the email to one screen on a phone.

A2. A three-stage reminder sequence

Some clients need three nudges. Build the sequence once and store it, so the tone escalates without becoming sharp.

Write three reminder email templates for clients who have not sent documents, to go out at 14 days, 7 days and 2 days before [deadline for documents].
Placeholders: [client first name], [tax year], [missing items], [deadline], [your name].
Stage 1 gentle, stage 2 clear about what happens if the deadline passes (we may need to file an extension, which is a normal option), stage 3 direct with a call-to-action to phone or upload today.
Each under 100 words. No threats, no penalties, no amounts, no claims about rules.

The wording about extensions needs care. A template that says "an extension gives you more time to pay" would be wrong in many places, because an extension to file is not always an extension to pay. Keep that sentence out of the model's hands: delete any such claim and replace it with your own approved wording, or nothing. The prompt's ban on claims about rules is there for exactly this reason.

An illustrative stage-2 output: "Hi [client first name], we are still missing [missing items] for your [tax year] return, and the document deadline is [deadline]. If we cannot receive them by then, we may have to file an extension so we can do the work properly. Upload them using the link in my last email or reply with any questions. Thanks, [your name]". That is a neutral, accurate message with no invented consequence.

Group B: deadline and filing messages

Both messages in this group sit next to rules and dates, so the model writes the structure and the tone while every factual sentence comes from you.

B1. An extension-notice email that explains and reassures

An extension notice is both a service message and a risk point. Clients panic when they read the word, and preparers get calls. The email should say what an extension is, what you will do, what the client must do and what stays unchanged. Because the rules vary between places and years, you must supply the factual sentences.

Write a template email telling a client we are filing an extension.
Use these approved sentences exactly where indicated and do not add any claim about rules, dates, penalties or payment:
[PASTE YOUR APPROVED EXPLANATION OF WHAT THE EXTENSION DOES AND DOES NOT CHANGE]
Placeholders: [client first name], [tax year], [new target date for documents], [your name].
Structure: reassure in one sentence, explain using my approved sentences, state what I need from them and by when, and offer a phone call.
Tone: calm and practical. Under 150 words.

Notice what the prompt does: it passes the factual core to the model as fixed text. The model contributes the reassurance and the structure; you contribute the substance. That split is the pattern for every message that carries a rule. Your approved sentences can come from your professional body's guidance or from a fixed paragraph you have used for years. Have the first version checked by a colleague or adviser, then reuse it.

B2. A thank-you and confirmation note after filing

After filing, one short message does three jobs: confirms what was done, tells the client what to keep and offers a way to book next year. Keep it factual.

Write a template email confirming a return has been filed.
Placeholders: [client first name], [tax year], [date filed], [what to keep and for how long, my approved wording], [next-year booking link], [your name].
Include one sentence thanking them. Under 120 words. Do not state any refund amount, balance due, deadline or rule.

Illustrative: "Hi [client first name], your [tax year] return was filed on [date filed]. Please keep [my approved wording]. If you would like to book next year's appointment, use [next-year booking link]. Thank you for sending everything on time. [your name]". The prompt forbids amounts for a good reason: a number in a template could be merged against the wrong client. Amounts belong only in messages you write by hand or merge from reviewed data.

Group C: new-client and engagement messages

These are the messages that decide who you work for this season: one chases a signature, the other says no politely.

C1. A follow-up for the client who has gone quiet

New clients who have received an engagement letter and not signed it are a common drag in the first weeks of the season. The follow-up should be plain and make the next step trivial.

Write two follow-up emails for a prospective client who has not returned the signed engagement letter.
Placeholders: [client first name], [service], [fee], [signing link], [date we hold their slot until], [your name].
Email 1 after 3 days: friendly nudge with the link. Email 2 after 8 days: says we can only hold their slot until [date], offers a short call.
Do not pressure, do not mention competitors, do not add any fee other than [fee]. Under 90 words each.

Illustrative output for email 2: "Hi [client first name], I can hold your slot for [service] until [date]. After that I may not have capacity before the busy weeks. The letter is here: [signing link]. If anything in it is unclear, I am happy to talk it through in ten minutes. [your name]". The phrase "may not have capacity" is true for most practices in season; edit if it is not true for yours.

C2. A polite decline for work outside your practice

Every preparer is asked for something they should not do: a return type they do not handle, a deadline that cannot be met, a client whose records are a mess. A short decline protects your time and your reputation.

Write a template email politely declining new work.
Reason to give: [one true reason: outside our services / no capacity this season / records need work first].
Offer: [a referral to a colleague, or a date after which they can ask again, or nothing].
Under 90 words, warm, no blame, no legal language.

Use it sparingly and always fill the reason with something true. A decline that hides the reason invites argument; one that states it plainly tends to end the conversation well. The model will occasionally add "I wish you the best with your tax affairs", which is fine, or "we strongly recommend you consult a tax professional", which sounds odd from a tax professional. Delete the second.

Group D: checklists and the intake call

The checklist and the script share one source, so the questions asked on the call match the documents clients are asked for.

D1. A year-round document checklist, grouped and tickable

A checklist is the best format for clients, and it is also the safest use of AI in a tax practice because it is general by design. Ask for a client-facing checklist that adapts by client type using a clear switch: an employee, a self-employed person, a landlord, a retiree.

Create a plain-language document checklist for tax preparation for each client type: employee, self-employed, landlord, retired.
Format: one list per type with tick boxes (use "[ ]"), grouped under income, expenses, and "things that changed this year".
Add, for each type, five questions I should ask on the intake call.
Use generic descriptions of documents; do not cite form numbers, rates, thresholds or deadlines.

A sample output for a self-employed client lists income records (invoices, payment platform summaries), expenses (vehicle log, home-office costs, software subscriptions, equipment purchases) and changes (new business bank account, a new employee). The intake questions might include "Did you start or stop any business activity this year?" Ask a colleague to review each list for gaps, because the model cannot know what your jurisdiction and your clients' situations require. Add rows from your own experience: the items clients always forget are the best part of any checklist.

D2. The intake-call script, so no question is skipped

Intake calls drift. A script that follows the checklist keeps them to 20 minutes and produces consistent notes. For the diligence side of this work, due-diligence interview notes and checklists with AI goes into recording answers properly.

Write a 20-minute intake call script for a returning client in a tax practice.
Sections with timings: greeting and purpose (2 min), changes since last year (6 min), documents still missing (4 min), fees and timing (4 min), questions and next steps (4 min).
Under each section, give the exact questions to ask and a line to say if the client does not know.
Do not request any identifying numbers or financial figures in the questions.

The last line matters: the script itself should not train staff to ask for sensitive items over a call that may be recorded or noted in an unsafe place. Use your secure upload for anything with an identifier. Illustrative script line: "Has anything changed in your household, job or home since last year? If you are not sure, tell me what you remember and we will check it against your documents."

The rushed prompt that carried a client's ID digits

Here is the mistake the pack is designed to prevent. An assistant, rushed at the peak of the season, pastes this into a personal-plan chat: "Write a reminder to Mr [real surname] at [street address] who still owes us his employer and bank statements, his income was about [amount] last year and his ID ends in [four digits]." The prompt carries a name, an address, an income figure and part of an identifier, none of which the email needs.

The corrected prompt reads: "Write a reminder template for a client who still owes us documents. Placeholders: [client first name], [missing items], [deadline], [your name]. Under 80 words, friendly, no amounts." The output is identical in usefulness, and the data stays in your software. When you train staff, show them both versions side by side and ask what the model gained from the extra detail. The answer is nothing, and that is the whole lesson.

A good habit is to make the redacted version the only version people ever type. If a template needs a variable, add a placeholder. If a variable has to be discovered from the file, that step belongs in your practice software, not in a chat window.

A filled-in checklist: the landlord client

To show what a finished checklist from D1 looks like after your review, here is an illustrative landlord version. Every line is generic and carries no client data:

• Income: [ ] Rent received by property (bank statements or ledger); [ ] deposits held and returned; [ ] any other income linked to the property such as parking.
• Expenses: [ ] Agent or management fees; [ ] repairs and maintenance invoices (not improvements); [ ] insurance; [ ] mortgage interest statement; [ ] utilities paid by you; [ ] travel to the property, with a log.
• Things that changed: [ ] Properties bought, sold or newly let; [ ] periods when a property was empty; [ ] a change in ownership share; [ ] major improvement works.
• Questions for the call: Did any tenant move out or in? Did you pay for anything that added value, not just fixed something? Do you hold receipts for every repair? Did you rent to a relative at a reduced rate?

Review it with fresh eyes. The model listed "improvements" next to "repairs", which is a genuine distinction in many places and a common source of questions, but it cannot tell you how it is treated. Leave the treatment to the preparer. If your own practice handles furnished lets or short stays, add those rows. A checklist that reflects your clients' real situations is more valuable than any generic one, and the additions are usually the items you have seen go missing before.

Testing a template on three imaginary clients

Before you trust a template, run it against three invented cases that stress it: a client with a long list of missing items, a client with none, and a couple with different first names. For the document request, check that the checklist is still sensible when only one item is missing, and that the greeting works for two people. For the extension email, check it reads sensibly for someone who has already sent everything. If a template fails a case, fix the wording or create a variant. Three test cases take ten minutes and prevent most mail-merge embarrassments.

Record which variant covers which situation in the prompt file: "Reminder A: one or two items missing; Reminder B: more than five; Reminder C: all missing". A few well-labelled variants beat a single template with conditional text, which tends to break when fields are empty.

Merging real details safely: the practical routine

A template is only useful if the merge step is as safe as the drafting. Store the approved templates in your email app's template feature or your practice software's mail merge. Use its fields for [client first name] and [deadline]; the data stays in the system that already holds it. If you have no merge tool, a spreadsheet with one column per placeholder and a standard mail-merge feature in your word processor will work.

Before each batch, send one test to yourself. Check that every placeholder was replaced (search for square brackets), that dates read correctly, and that nothing from a different client slipped into a field. A realistic mistake: a practice sent forty reminders in which the [missing items] field was empty for eleven clients, producing the line "we are still missing for your return". A two-minute test would have caught it. Another: a template said "your return" to a couple who file together; add [spouse first name] where relevant or use "your return" deliberately.

Checking the pack against your practice's obligations

Three checks keep the pack safe. First, security: does your written security plan cover AI tools, and does it say placeholders only? adding your AI tools to a written information security plan gives a structure. Second, consent: if you ever use AI on data that identifies clients, check your consent obligations; this pack avoids that situation by design. Third, accuracy: every sentence that states a rule, date or consequence must come from you or your professional body, not the model.

A cost-and-time sum, illustrative: a practice that sends 300 document-request emails and 400 reminders a season, at about five minutes each to personalise by hand, spends roughly 58 hours on those messages. With templates and a merge, the hands-on time might be nearer 15 minutes of checking per batch of 100. The saving is real, but only if the templates are right, so spend the first hour on review. For wider workflow ideas, how small tax practices use AI through the busy season goes beyond emails.

Keeping the prompt pack current from year to year

Each autumn, review every template against that year's rules, your fee schedule and the lessons from the last season. Note which emails drew questions: a template that prompts three replies out of ten needs rewording. Keep a changelog with the date, what changed and who approved it, so staff always use the current version.

Finally, train every preparer on the five-line rule in under ten minutes, and test it once with a fake case: hand them a mock email thread containing a made-up tax ID number and ask what they would paste. The answer should be "nothing". If you want help setting up the pack, the merge routine and the staff rules together, I cover exactly that in an AI implementation consultation.

Questions tax preparers ask about ChatGPT and client privacy

Can I paste any part of a client's return into ChatGPT?

Not for these prompts. Everything here works from placeholders and generic facts, so no return data is needed. Return figures, tax ID numbers and identity details should never go into a chat, whatever the plan. Check your professional obligations and your client consent forms before any other use.

What does a placeholder-only prompt look like?

It names the situation without the person: [client first name], [tax year], [document list], [deadline date]. The model writes a template; you merge the real details in your practice software or mail merge, where client data already lives and is already protected.

Does a business plan make client data safe to paste?

It changes how OpenAI treats the data, since Business workspaces are not used for training by default, but your duty of confidentiality and any consent rules still apply. Treat the plan as one safeguard, not permission. Redaction-first remains the safest habit.

Should I tell clients I use AI to draft emails?

Check your engagement letter and any professional rules that apply to you. If AI never sees client data, many preparers add a sentence saying drafts of general communications may be AI-assisted and always reviewed by a person. Wording is for you and your adviser to settle.

Further reads

• How to Automate Client Onboarding in a Small Accounting Practice — Automate the onboarding steps around these client emails.
• AI Readiness Checklist for Accountants, Solicitors, Consultants — Check how ready your practice is for wider AI use.
• AI Incident Response Plan for Small Businesses (With Template) — Plan for what to do if client data ends up in the wrong place.
• A Simple AI Risk Register for Small Businesses (With Template) — Log and rate the AI risks in your practice.
• How to Mention AI Use in Client Contracts and Proposals — Add AI wording to engagement letters.
• Tax Preparation AI Software: TaxDome, Canopy and ChatGPT Compared — TaxDome, Canopy and ChatGPT sit in different product categories. Compare their AI, their cost models and the consent step before return data goes in.
• AI Tax Organiser: Build One That Chases Missing Documents — Turn a prior-year return into a client document list, track uploads row by row, and send reminders that name exactly what is still missing.
• AI Tax Return Prep for Small Practices: Extract, Review, Sign Off — A three-stage workflow for AI-assisted return prep: extract with page references, tie out with a spreadsheet, then review and sign off in writing.
• ChatGPT for Sales and Marketing: B2B Battle Cards and One-Pagers — Prompts and a fact-check routine for a competitor battle card and a one-page service sheet in a small B2B firm.
• Circular 230 and AI: Due Diligence When a Tool Drafts the Work — A per-return checklist and sign-off sheet that turns the IRS's June 2026 AI alert and Circular 230 diligence rules into a routine.
• Tax Services AI for a Small Practice: Intake, Chasing and Review Notes — A three-part workflow for a small tax practice: intake checklist, document-chasing emails and reviewer notes, with a named person owning each step.
• Tax Workpaper Prep With AI: Tie the Return Back to Source Documents — How AI can draft a source-to-return tie-out schedule for a tax return, and the reviewer checklist that proves each figure before sign-off.
• Client Tax Organiser: Build One With AI and Send It by Portal — A conditional tax organiser example with client questions, portal instructions, staged reminders and a completeness review.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI Help Center page on data controls in ChatGPT; a national tax authority's page on filing extensions, which states the extension is only for filing (both checked 4 November 2026).