The client email prompt every CPA firm needs

Using LLMs like Gemini, Copilot or ChatGPT can be an affordable starting point before investing in expensive tools. What kinds of tasks are best suited for these generalist models versus specialized accounting software?

Here’s how I think about tools like Google Gemini, Microsoft Copilot and ChatGPT: they’re great “first-draft” helpers. If the job is mostly words (not math) and you don’t need to be inside your tax or accounting system, they can save a ton of time.

Where general models work well is for drafting or polishing client emails and memos; summarizing long documents (engagement letters, IRS notices, meeting notes); turning a rough process into a checklist; brainstorming questions before a client call; and getting a quick, plain-English explanation of something you haven’t seen in a while. My rule of thumb is simple: let AI help you write, summarize and think.

One of the biggest concerns we hear is around data security and client confidentiality. What are the guardrails firms should put in place before using AI tools with real client work?

This is the right concern to have. I usually start with one simple line in the sand: don’t include client confidential information in your prompts until you understand where that data goes, who can access it and whether it can be used to train the model. Once that’s clear, a short set of firm guardrails gets you 80% of the way there.

First, pick your approved tools. Decide what’s OK for firm work (and what isn’t). When in doubt, favor tools with business/enterprise privacy controls.

Next, set a client-data rule. Use anonymized/redacted examples unless you’re confident that the AI application’s settings are properly set to protect your data privacy.

Finally, require a human review. Treat AI output like a draft from a smart intern—helpful, but not ready to send until a CPA reviews it for accuracy, tone and compliance.

As a practical example: Microsoft Copilot for Microsoft 365 is meant to operate inside Microsoft 365’s security and compliance boundaries. They refer to it as “Enterprise Data Protection,” which ensures your prompt data will never leave the four virtual walls of your organization. This provides peace of mind.

What’s one workflow you’ve seen firms improve with AI that doesn’t require any integrations or new software?

One of my favorite “no-integration” wins is better meeting notes. If your meeting platform can record and create a transcript, you can turn that into 100% accurate notes in minutes. 

Then you can ask the transcript things like: “What action items did we agree on, and who owns them?” “Did we talk about next year’s estimates?” “What did the client say about that IRS notice?” 

What’s a simple way to use AI to draft client communications that still sound like the firm?

The easiest way I’ve found is to standardize how you ask for the draft. If everyone uses the same “recipe,” the emails and memos come out sounding like your firm—not like a generic template. There are two simple ways to do that.

The first is a set of custom instructions (if the tool supports it). Save a short description of your firm (services, tone, typical clients) plus your writing preferences (plain language, short paragraphs, bullets for steps, end with a clear call to action). That becomes the default background for future drafts.

Another option is creating prompt templates. Create a few reusable prompts for your “repeat offender” communications (IRS notice response, missing-docs reminder, extension explanation, bookkeeping clean-up summary). Your team fills in the variables, and the output stays consistent.

The most important step for CPA firms to take is to establish a culture that encourages exploration and application of generative AI within the framework of your firms’ responsible AI usage policy.