Accounting September 22, 2026 8 min read

Automate Client Tax Return Checklists With ChatGPT

A practical build for Australian accountants: turn a client's circumstances into a personalised, ATO-aware document-request checklist using a saved ChatGPT prompt and a simple Zapier flow.

To automate client tax return checklists with ChatGPT, you build one reusable prompt that turns a few facts about a client — occupation, income types, whether they own a rental or run a business — into a personalised list of the documents and substantiation you need from them. You then connect that prompt to a Zapier flow so a new client record automatically generates the checklist and emails it out, without anyone retyping the same request list forty times in July.

This is written for Australian accountants and tax agents, and it uses ATO terminology — deductions, substantiation, work-related expenses, rental schedules — rather than US tax concepts. If you are a sole practitioner or a small firm drowning in "can you send me your documents" emails every tax season, this is the workflow to build first.

Why generic tax checklists waste everyone's time

Most firms send every individual client the same PDF checklist. It lists everything: motor vehicle logbooks, rental income, dividend statements, crypto, work-from-home hours, private health details. The client scrolls a two-page document, decides most of it does not apply to them, and either sends nothing or sends the wrong things.

A personalised checklist flips that. A salaried nurse with a mortgage gets a short list: PAYG summary, private health statement, work-related expense receipts relevant to nursing, and their working-from-home record. A client with a rental property and a small sole-trader business gets a longer, different list. Same effort from you once it is built, far better response rate from the client.

This sits alongside the other admin wins covered in AI for Accountants: 5 Admin Tasks You Can Automate With ChatGPT — this article goes deeper on one of them.

Step 1: Write the reusable ATO-aware prompt

The whole build rests on one well-structured prompt. You are not asking ChatGPT to know your client — you are giving it the facts and asking it to assemble the right document request. Write it once, test it, and save it as a Custom GPT or a saved prompt so staff use the same version every time.

Here is the core prompt to paste in and refine:

You are assisting an Australian registered tax agent to prepare a document-request checklist for an individual tax return. Based on the client details below, produce a concise, plain-English list of the documents and records we need from this client, grouped under headings. Use Australian tax terminology (PAYG payment summaries / income statements, work-related deductions, substantiation, rental schedules, CGT). Only include items relevant to the details given. For any deduction that requires substantiation, note briefly what evidence the ATO expects. Do not give tax advice or estimate refunds. Client details: [occupation], [income types], [owns rental: yes/no], [runs a business or is a sole trader: yes/no], [investments: shares/crypto/managed funds], [private health insurance: yes/no], [other notes].

Run it with a made-up client first. When the first output is wrong — and it usually is in one of these ways — you correct it in plain English:

  • If it invents items that do not apply, add: "Only list items directly supported by the details provided. Do not include speculative categories."
  • If it slips into giving advice or estimating outcomes, add: "Do not comment on eligibility, amounts or likely refund. Only list documents to collect."
  • If it uses American terms like "W-2" or "1099", add: "Use only Australian terms. Never reference US tax forms."
  • If the tone is too technical for a client, add: "Write for a client with no accounting knowledge."

Two or three rounds of this and you will have a prompt that produces a clean, client-ready list every time. Save that final version. That saved prompt is the asset — the automation around it is just plumbing.

Step 2: Connect it to a Zapier flow

Once the prompt is reliable, Zapier turns it into a hands-off process. The pattern is the same one we describe in building a client intake form with Notion, ChatGPT and Zapier, pointed at tax season instead of onboarding.

  1. Trigger: a new or updated row in your intake source. This could be a Google Form, a Microsoft Form, an Airtable base or a Notion database that captures the client's occupation, income types and the yes/no flags from the prompt.
  2. Action — ChatGPT (OpenAI) step: Zapier passes those fields into your saved prompt and returns the generated checklist as text.
  3. Action — draft the email: insert the checklist into a Gmail or Outlook draft, or a template, addressed to the client.
  4. Human review step: route the draft to a staff member for a quick check before it sends. Do not let it fire automatically to the client on day one.

For a small firm running one or two seasons of this, Zapier's paid "Professional" tier handles the multi-step flow comfortably. If your logic gets complex or you want it running inside your own infrastructure, weigh it against the alternatives in n8n vs Zapier for professional services firms.

What has to stay human

The checklist is a request for documents, not a return. Keep a person in the loop for anything that requires judgement:

  • Reviewing the checklist before it goes out. Early on, read every one. Once you trust the prompt for standard cases, spot-check and only fully review the unusual ones — rentals, businesses, capital gains events.
  • Anything the client's situation flags as messy. A marriage separation, a deceased estate, a first year of business — these deserve a phone call, not a form.
  • The actual advice. The prompt is deliberately barred from commenting on eligibility or amounts. That is your work and your professional obligation as a registered agent, not the model's.

Automation that removes the human review step is where firms get burned. The goal is to delete the retyping, not the judgement.

Handling client data and the Privacy Act

The moment a client's name, occupation and financial circumstances go into an AI tool, the Privacy Act applies. The OAIC's general guidance on commercially available AI products is clear that personal information should not be fed into public chatbots, and that principle governs this build.

In practice:

  • Use the OpenAI API tier via Zapier, or a Custom GPT on a business plan — not the free public ChatGPT — so inputs are governed by contract and not used for training.
  • Feed the prompt only the fields it needs. The occupation and income-type flags do most of the work; you do not need to paste the client's tax file number or address to generate a document list, so do not.
  • Prefer a tool your firm already licenses. If you run Microsoft 365, Copilot inside your tenant may be the cleaner option — see Microsoft Copilot vs ChatGPT for accounting firms.
  • Where processing happens overseas, treat it as a cross-border disclosure under Australian Privacy Principle 8 and record it. Keep a simple written data-flow note for the build.

For the full picture, read is AI safe for client data under the Australian Privacy Act before you connect anything live.

What it costs and who owns it afterwards

The running cost is modest: a Zapier paid plan plus API usage measured in cents per checklist. The real cost is the few hours to write and test the prompt and wire up the flow.

Ownership is the part firms forget. Decide up front:

  • Where the saved prompt lives. Store the final prompt text in a shared firm location, not in one person's head or personal account. If they leave, the asset stays.
  • Who maintains it. ATO expectations and your standard document requests shift year to year. Someone should own reviewing the prompt each June.
  • How to avoid shadow IT. This flow should sit inside your firm's AI policy and use firm-controlled accounts — not a staff member's private ChatGPT login. If you do not have a policy yet, the structure in our AI policy template adapts easily to an accounting practice.

When not to bother automating this

If most of your clients are complex — SMSFs, trusts, multi-entity businesses — a generated checklist saves less, because each one needs a conversation anyway. The payoff is largest for firms with a big book of straightforward individual returns where the same requests repeat hundreds of times.

And if you have never built a Zapier flow before, start with the prompt alone. Just having a saved, tested prompt that a staff member runs manually and pastes into an email captures most of the time saving with none of the integration risk. Add the automation once the prompt has earned your trust across a season.

If you want a broader view of which admin tasks are worth automating before you commit, our free guide, 10 AI Workflows to Save 10+ Hours a Week, walks through the highest-value candidates for professional services firms — the tax-checklist build is a natural place to start.

Common questions

Can I use the free version of ChatGPT for client tax checklists?

Not with real client information. The free public tool can enter your data into training and does not offer contractual protection, which conflicts with the OAIC's guidance and your Privacy Act obligations. Use the OpenAI API through Zapier, a business-plan Custom GPT, or Copilot inside your Microsoft 365 tenant, and only include the fields the checklist actually needs.

Does the checklist need to be reviewed before it goes to the client?

Yes. Keep a human review step, especially early on. The prompt generates a document request, not advice, but it can miss context or occasionally include an item that does not apply. Read every checklist while you are building trust, then move to spot-checking standard cases and always reviewing complex ones like rentals or businesses.

What Zapier plan do I need for this automation in Australia?

A multi-step flow that triggers on a form, calls an AI step and drafts an email needs Zapier's paid Professional tier. API usage adds only cents per checklist. If your logic grows complex or you want it hosted on your own infrastructure, compare it against n8n before committing to a long-term platform.

Will ChatGPT give correct Australian tax advice in the checklist?

It should not give tax advice at all, and your prompt must explicitly forbid it. The tool's job is to list documents and note what substantiation the ATO expects — not to assess eligibility, estimate refunds or interpret the law. Eligibility judgements and advice remain your responsibility as a registered tax agent.

How often should I update the saved prompt?

Review it at least once a year, ideally before tax season starts in June, and whenever your standard document requests or ATO substantiation expectations change. Store the prompt in a shared firm location so it survives staff turnover, and assign one person to own the annual update so it never quietly drifts out of date.

Want these ideas working in your firm? We build controlled AI workflows for Australian professional services firms — starting with a free automation audit.