Using AI to Draft Client Communications
Consistent contact is what keeps clients. But it takes work. AI can help carry the load, if life and health insurance agencies use it correctly.
A NextAgency Resource
Last Updated: September 21, 2026
A practical look at what AI is actually good at, where it breaks down, and how life and health insurance agencies can use NextAgency’s NextAI writing assistance tool to draft communications faster, without letting a hallucinated detail reach a client.
A Human’s Note:
Many agents use AI to help craft messages of all types, including those to clients and carriers containing personal health information. The implications of this are rarely addressed, which is why we thought this to be a useful topic for the NextAgency Resources Directory. And who (or more accurately, what) better to write this article than an AI? So I worked with Claude to draft the following article. I reviewed it and offered suggestion. However, the final decision on the content of this article was Claude’s.
Synopsis:
Insurance agents know they should stay in touch with clients throughout the year, not just at renewal. It’s how successful life and health insurance agencies keep them as clients. What’s less clear is how to do that without the communications sounding like a form letter or consuming a producer’s entire afternoon. This article is a practical guide to using AI to lighten that drafting load — what it’s good at, where it breaks down, and how to instruct it well enough that the time it saves isn’t offset by a mistake it introduces.
The short version: large language models like Claude and ChatGPT predict likely text based on patterns in enormous amounts of training data. That’s exactly why AI is strong at rewriting, tightening, and adjusting the tone of a message someone has already drafted. It’s also exactly why the same AI will state an outdated fact with total confidence, and why it won’t go online to check a source unless it’s told to, even when the need to check is obvious. The article walks through NextAgency’s NextAI writing assistance tool as a working example — what it actually does, and how to instruct it well — before closing on why NextAgency requires a human to review every AI-drafted message before it reaches a client.
Why year-round communication is worth it — and why it’s hard to sustain
Clients rarely announce that they’re leaving their current insurance agency. They just stop hearing from their agent, then start listening to someone else. As Consistent Client Communication lays out in more depth, the insurance agencies with the highest retention aren’t working harder than everyone else. They’re systematic about staying visible: a welcome sequence when coverage starts, a regular newsletter, a structured check-in before renewal.
Knowing what’s required doesn’t make it easier to execute what’s required. Newsletters need to get researched and written without repeating themselves. A rate-increase conversation has to be drafted client by client, in a tone that doesn’t sound canned. NextAgency, an agency management and CRM system, already handles distribution — scheduling campaigns, automating workflows, sending the right message to the right segment on the right date. What those tools don’t do is write the words. That’s where AI enters the picture, and where it earns its keep or doesn’t.
Why AI is good at writing insurance agency communications
It helps to understand what a large language model is actually doing when it drafts or edits text, because that mechanism explains both its strength and its failure mode when drafting or editing communications.
Models like Claude and ChatGPT are trained on enormous volumes of text and learn to predict the most statistically likely next word, given everything that came before it. That’s the entire operation. There’s no separate “understanding” module sitting behind the prediction — the fluency flows from pattern-matching, not insights or knowledge.
That mechanism happens to be extremely well suited to a specific, bounded task: taking existing text and reshaping it. Rewriting a paragraph in a warmer tone, tightening a rambling email down to three sentences, fixing grammar without changing meaning — these are exactly the kind of pattern-transformation problems a language model is built to solve well. NextAgency’s own approach to AI, described in Deploying AI, reflects this: match the tool to the task, and the task where AI performs most reliably is editing text that already exists, not inventing facts that don’t.
Why AI struggles with facts
The trouble is that the exact process that makes AI good at rewriting also makes it a poor fact-checker. A model predicting the next likely word has no internal mechanism for distinguishing “this is something I know to be true” from “this is something that sounds right.” It generates the statistically probable continuation either way, and it does so with the same fluent, confident tone whether the underlying claim is accurate or fabricated.
AI’s Built-In Unreliability describes this as driving with a navigator who never says “I’m not sure.” A human editor who doesn’t know a detail will usually hedge, or say so. AI, as a rule, doesn’t. It states an invented plan detail, a wrong effective date, or an outdated regulation with the identical confidence it uses for something true.
The practical rule that falls out of this: trust AI with how something is phrased. Verify everything about what it says.
Brainstorming with AI — and its blind spot
AI is genuinely useful for a narrower, lower-stakes task: generating ideas. Asked for five newsletter topics relevant to a Medicare book in October, or three subject lines for a rate-increase email, AI will produce reasonable starting points quickly. This article’s own topic came out of exactly that kind of back-and-forth.
There’s a blind spot worth knowing about here, because it’s easy to miss. Left to its own devices, AI defaults to generating what’s plausible based on its training, not what’s current — and it generally won’t search the web to check, even when checking is the obvious next step, even when a link to the source is sitting right there in the conversation. It has to be told, explicitly, to search and verify before an agency treats anything it says as current. Asking “what’s changing with Medicare Advantage this year” without that instruction risks an answer that sounds specific and confident and is quietly out of date. Asking the same question with “search for the current CMS guidance before answering” attached gets a different quality of answer. Prompts matter.
Case Study: How NextAgency’s NextAI writing assistance tool works
As an insurance agency management system and insurance agency CRM, NextAgency helps you create and manage notes, tasks, and emails. NextAgency’s NextAI writing assistance tool, opened in the text editor of any of these items is a working example of AI used for exactly the task it’s suited to: reshaping text a person has already started.
NextAgency structures this text crafting tool by surfacing six actions. Proof catches spelling and grammar without altering tone. Improve Writing revises text based on custom instructions — “make this clearer,” “simplify this into three bullet points” — and accepts specific direction, not just a general request to make something better. Shorten trims length while preserving the point. Change Tone rewrites a message as Professional, Empathetic, Casual, Neutral, or Educational, which matters when the same underlying content needs to reach a client differently than it reaches a colleague. Translate converts text into several languages. Summarize, available for notes, tasks, and tickets, condenses a long call note into something scannable.
Note that instructions like these aren’t vague requests along the lines of “make the email better.” They’re describing specific ways to make the email better — and that specificity is what determines whether the draft that comes back is usable. “Make it more persuasive” is broad enough that the output could go several directions. “Rewrite this for a client who just received a rate increase, acknowledging the increase before explaining what’s included” gives the model a much narrower, better-defined target — and produces a draft closer to what an agent would actually send. Full mechanics, including every available tone and action, are documented on NextAgency’s help desk, published in a machine-readable format as NextAgency’s AI and Misc Tools articles.
Worth noting: the line between “editing” and “drafting” is thinner than it looks. A detailed custom instruction to Improve Writing — describing what the email needs to say, not just how it should sound — gets an agency most of the way to a first draft, not just a polish. The tool is framed as an editor. Used well, and provided with enough information as to what you want your text to say, it does more than that.
Why nothing reaches a client without a human’s approval
AI can’t be trusted to avoid inventing information, and that’s exactly why NextAgency believes a human has to check its work before anything goes out the door. The agent and support staff servicing a case know that client in ways the AI never will, because the AI isn’t drawing on the relationship at all — it’s predicting the next likely word based on patterns in its training data. When that prediction is wrong and an inaccurate email reaches a client, it isn’t the AI that answers for it. The agency is responsible, and liable, for what went out under its name. The right posture is the same one used for any capable but unsupervised new hire: review the draft before it leaves the building, every time. NextAgency designed its NextAI writing assistance tool to require that review. The AI cannot send out an email; only a human can.
There’s also a data-handling dimension. As AI (Un)Privacy explains, insurance agencies handle health and financial information that carries real compliance obligations. Personal and commercial subscriptions to AIs, the kind most agencies have, are not HIPAA compliant. There’s no BAA. There’s none of the structural safeguards HIPAA requires of technology tools. Only the AI company’s enterprise versions offer the opportunity to use AI in a HIPAA-compliant way. This is why NextAgency runs Claude through AWS Bedrock, Amazon Web Services’ enterprise platform, governed by the agreements and safeguards HIPAA requires — not the consumer version of the tool.
The bottom line for life and health insurance agencies
Consistent client communication is critical for retaining clients. AI can help carry that load. But insurance agents need to know where it helps and where it doesn’t. A wrong tone in a client email is an inconvenience. A wrong fact in a client email — sent because no one checked it — is a compliance problem and a trust problem, wrapped into one message that already went out. Using AI to review text in a non-HIPAA compliant way creates a different kind of liability.
The solution: Read every draft before it ships. Use an enterprise version of the AI through a trusted partner like NextAgency. These aren’t limitations of the tool. They’re the difference between using AI to write faster and using AI to write carelessly.
Key Takeaways:
- Consistent client contact drives retention, but writing that contact by hand, all year, without it sounding canned, is a real time burden — one AI can help carry.
- AI predicts likely text from patterns in its training data, which is why it’s strong at rewriting and tone adjustment and unreliable at generating facts it presents with the same confidence either way.
- Left unprompted, AI will not search the web to verify current information, even when the need is obvious — it must be told explicitly to search and confirm before its output is trusted.
- Specific instructions to an AI writing tool (“rewrite this for a client who just received a rate increase”) produce meaningfully better drafts than vague ones (“make this better”).
- Commercial AI subscriptions most accessible to insurance agencies are not HIPAA compliant. If you’re sharing PHI with an AI, that AI needs to be through an enterprise model that includes a BAA and other HIPAA-required safeguards.
- NextAI’s writing assistant edits and drafts, but it never sends — an agency management system like NextAgency keeps a human between every AI-drafted message and the client’s inbox, because the liability for what’s said stays with the agency, not the AI.
Learn More:
For more on why staying in touch matters and what a year-round outreach plan looks like in practice, see Consistent Client Communication Increases Renewals. For additional ways to grow your agency, browse the full NextAgency Resource Center.