The Insurance Agent AI Toolkit: 8 Tools That Work as a System

Most agents who try ChatGPT spend five minutes generating a decent renewal email, then eight minutes manually logging it, re-entering client details, and creating a follow-up task in their AMS. The AI saved them nothing — it just moved the friction.

The Big I Agents Council for Technology found in February 2026 that 45% of independent agencies now use a public AI tool, but only 8% say AI is embedded in their daily workflows. That gap exists because tools bought in isolation become another inbox to check.

The default starting stack is Google NotebookLM (free) for policy analysis, ChatGPT Free or Fathom Free for drafting and meeting capture, and Make's free tier as the connective layer. Upgrade to roughly $50/month — ChatGPT Plus plus Make Core plus Otter Pro — when one repeated task takes more than 30 minutes per week. And avoid any tool that creates a second copy of client data outside the AMS without a clear write-back path. That's where tool sprawl starts and E&O exposure grows.

The Free Stack: Three Tools That Connect

Google NotebookLM

This is the least obvious tool in the free tier and the one with the strongest practitioner evidence behind it. Upload a 60-page commercial policy PDF, ask "What are the liability limits, key exclusions, and three coverage gaps for a light manufacturing operation?" and you get a cited, source-linked answer in seconds instead of 20 minutes of manual scanning.

Dennis McCurdy, a veteran CIC and broker at McCurdy Group, uses NotebookLM for exactly this. He drops a commercial policy in, runs a saved prompt, and extracts the material for a detailed client renewal summary. His commercial lines staff now ask the tool questions instead of hunting through 40-page policies. "I find it a tremendous timesaver," he says.

One security caveat is non-negotiable: strip the client-identifying page before uploading, or use a redacted copy. Sensitive PII doesn't belong in an unvetted tool even if it appears private. The output is structured facts for the agent to verify — not a final recommendation. A licensed agent reads every extracted limit and exclusion before anything reaches a client.

Best for commercial-lines agents who regularly prepare renewal summaries or need fast answers from dense policy documents.

ChatGPT Free (or Claude Free)

Think of this as the drafting layer, not the analysis layer — NotebookLM handles documents; ChatGPT handles language. The most useful applications are turning rough call notes into a clean CRM entry, generating a first-draft renewal check-in email, or creating a list of missing-information questions for a new commercial submission.

One design point matters: the agent pastes the ChatGPT draft into their email client and reviews it before sending. The AI doesn't know the client's actual policy, can't access the CRM, and shouldn't write coverage language without a human reading it first. The free version may also use inputs for model training, so drafting a generic "hard market explanation" email is fine — pasting a client's claim history is not. If your agency runs on Google Workspace, Gemini is a natural alternative and integrates with tools already in use.

Fathom (Free Tier)

Fathom records, transcribes, and summarizes Zoom, Google Meet, and Microsoft Teams calls. The free tier produces a structured recap with action items that an agent can review, edit, and paste into the CRM within two minutes of hanging up. The consent requirement is not optional: disclose recording before the call starts, and agency policy should define which call types can be recorded.

The data flow is simple: Fathom recap → agent reviews and corrects → approved summary logged to CRM → follow-up tasks created. Otter's free Basic tier — 300 transcription minutes per month — is a direct alternative and worth naming if Fathom's Zoom integration doesn't fit your setup.

The free stack breaks down at volume. When you're running five client meetings a week and preparing ten renewals a month, the copy-paste steps accumulate. That's when $50/month earns its keep — and the upgrade is specific, not vague.

The $50/Month Stack: Where the System Starts to Run Itself

ChatGPT Plus or Claude Pro ($20/month)

This is the single most important upgrade, and it pays back faster than anything else at this price point. The key unlock isn't just better model quality — it's file uploads and persistent custom instructions.

With ChatGPT Plus, you can upload policy PDFs directly, create a Custom GPT trained on your agency's preferred tone and compliance language, and organize saved prompts by workflow lane inside Projects. Kimberly Fox of O'Connor Insurance Associates in North Carolina tested several AI platforms before selecting Claude for her whole team, specifically because it could generate actual documents in required output formats and required fewer revision cycles. Her team now uses it daily for process creation, document comparison, meeting summaries, and training guides.

We tested several different AIs before deciding. Some can't generate actual documents, which was a capability we knew we really wanted. Claude consistently had better results in most categories.
— Kimberly Fox, Commercial Operations Leader, O'Connor Insurance Associates

The upgrade trigger is precise: when you're copy-pasting the same prompt more than five times a week, or spending more than five minutes reformatting AI output to match agency templates, a Custom GPT or Claude Project eliminates that permanently. The practical security reason to pay: both ChatGPT Plus and Claude Pro explicitly prevent conversation data from being used for model training, which is the real reason to upgrade, not just the feature set.

For agencies already in the Microsoft 365 ecosystem, Copilot at a similar price point is worth considering as an alternative.

Make (Core Tier, $12/month)

This is where the stack becomes a system rather than a collection of tabs. Make connects events that already happen — a new form submission, a renewal date in the CRM, a service request changing status — to structured actions: creating a lead, drafting an acknowledgment, assigning a task to the right CSR.

The Core plan unlocks unlimited active scenarios and scheduling down to the minute, which is the minimum needed for a reliable intake or renewal trigger. Be honest about the setup cost: the first Make scenario takes two to three hours to build correctly, including field mapping, error handling, and a test run. That's a one-time cost, not a recurring burden.

The most valuable connection for insurance agents: a web form or email submission → Make creates a lead in AgencyZoom → Make sends that data to ChatGPT via the OpenAI module → ChatGPT returns a draft acknowledgment → the draft lands in the agent's email drafts folder for review before sending. That's the intake automation that moves a lead response from same-day to same-hour. Zapier's Professional plan at $19.99/month is a direct alternative — slightly more app connections, similar pricing, less visual workflow builder. The choice between Make and Zapier at this scale is mostly personal preference.

Start automating for free using Make's free tier as the entry point, then upgrade to Core once one scenario runs reliably.

Otter Pro ($17/month)

Worth adding briefly: Otter Pro delivers 1,200 recording minutes, advanced meeting templates, and Salesforce/HubSpot/Zapier integrations that let a call summary automatically create a CRM note. The upgrade over Fathom Free makes sense only when you're on more than ten client calls per month and the missing CRM automation is the actual bottleneck — not before.

Two Workflows That Show the System in Action

Renewal Prep (5 Steps)

A renewal date 90 days out triggers a Make automation from the AMS, creating a task for the account manager. The account manager uploads the expiring policy PDF to ChatGPT Plus or NotebookLM — redacted of identifying client details — with a saved prompt: "Extract building limits, liability limits, key exclusions, and identify three coverage gaps for this insured's industry." The AI returns a structured draft with limits listed and gaps flagged. The account manager reviews line by line, corrects anything that looks wrong against the actual policy, and approves the language. The approved summary becomes the basis for the renewal email, sent through existing Gmail or Outlook.

The human gate sits between steps three and four, and between four and five. Nothing reaches the client without the agent reading and approving it. Tanya Horst of Cedar Risk Management uploads multiple quotes and compares them against the bound policy before use, and she's explicit about why: "We always review everything to make sure we're not missing anything and nothing's going to hit our E&O."

While AI has been extremely helpful with drafting, polishing, coding, and content generation, I have not yet figured out how to make it actually "do" tasks for us in the way some agencies describe. True hands-off automation is still an area we are learning how to approach.
— Jack Tribble, Licensed Sales & Agency Admin, Jon Peters Insurance Agency

New Lead Intake (5 Steps)

A prospect submits a web form. Make detects the submission — Jotform, Typeform, and Google Forms all connect natively — and creates a lead in AgencyZoom or the agency CRM. Make sends the lead's stated information to ChatGPT via the OpenAI module, which generates a first-response email using a pre-written agency template. The draft arrives in the agent's drafts folder, not auto-sent. The agent reads it, adds one personal sentence, and sends within minutes of the form submission.

The draft should contain no coverage recommendations, no pricing, and no advice — only a warm acknowledgment and a next-step invitation. The agent adds the judgment layer before sending.

The Governance Checklist You Actually Need

Five non-negotiable rules before any workflow goes live:

1. Any AI output that explains, recommends, or limits coverage must be read and approved by a licensed agent before it reaches a client — no exceptions.
2. Real client names, policy numbers, and claim details belong only in vetted business-plan tools, not free tiers with undefined data-retention policies.
3. Coverage comparisons generated by AI must be checked field by field against the source document — AI can miss endorsements, sublimits, and conditional exclusions.
4. Vendor contracts often cap liability at six to twelve months of fees paid; an agency can face a significant E&O claim while recovering only what it paid in subscriptions.
5. Name an owner for every automated action: who gets the error alert, who corrects a bad output, who can pause the automation. Unnamed automations become mystery machines.

The 90-Day Path

Week 1: Pick one repeated task. Set up Fathom Free on your next Zoom call with client consent. Review the summary. Does it capture the material facts accurately? If yes, run it two weeks before touching anything else.

Month 1: Add ChatGPT Plus or Claude Pro. Create one saved prompt for your highest-frequency task — renewal summary or new-lead response. Run the AI output in parallel with your existing method for two weeks, counting corrections. If corrections stay under 10% and the workflow saves 30-plus minutes per week, keep it.

Month 3: Connect the workflow to your CRM using Make or Zapier. Build one trigger and one action — draft task or first-response email to the drafts folder. Measure weekly adoption. If two people are using it consistently and corrections stayed below 10%, expand to the next workflow lane.

Here's the simplest way to start today: open NotebookLM, upload a policy you know well, and ask three questions you'd normally have to hunt for by scrolling. Time how long the answers take versus your usual method. That gap is where the system starts.

Two things worth watching over the next year: AMS-native renewal intelligence tools are maturing, and the agency management vendors are actively building two-way write-back — ask your AMS vendor which AI partners update the record, not just read it. And the E&O landscape around AI is developing quickly; Swiss Re's published guidance recommends running AI and human processes in parallel before trusting AI output alone, which is also just good workflow design.

Recommended Tools & Resources