Your AI sounds generic because nobody trained it, and that’s the step almost every business owner skips. They open ChatGPT, ask for a social post, get something that could belong to any company in America, and decide the tool isn’t ready for them. It’s not a broken tool. It’s an untrained teammate.
On September 30, I ran a working workshop for insurance agents at Boomers, a follow-up to a presentation I gave their team the Friday before. Most of the room was heading into AEP, the busiest stretch of the year for Medicare agents, and they still showed up to build, which puts them in the 1% of people actually doing this work. The goal was simple: everyone walks out with at least one trained AI teammate, no matter which tool they use.
Here’s the exact process we worked through, plus the parts most people skip.
Your AI Isn’t Broken. It’s Untrained.
Training is the least exciting part of AI, and it’s the part that decides whether your output sounds like you or like a stock template. What makes your content different is what you put into the tool, so a teammate that knows your role, your clients, and your goals will outperform a smarter model that knows nothing about you.
I used to say the goal was AI that sounds like you. Now the bar is AI that thinks like you, which means it needs your decisions, your examples, and your standards, not just your vocabulary.
Stop Chasing Tools. Pick One and Go Deep.
We started with a show of hands, and the room split across ChatGPT, Claude, Gemini, Copilot, and Grok. Every one of those can do what we built that day, so the tool matters a lot less than the training.
Unless you test AI platforms for a living, pick the one you like and stop worrying about what everyone else uses. Copilot gets a bad rap, and most of the time that’s because company IT has locked down its features, not because the tool can’t do the job. At REBL Labs, we run Claude every day alongside Google Workspace, and that’s the setup I demoed.
Step 1: Build One Project Before You Build Anything Else
Projects are where your training lives, and ChatGPT and Claude both use the same name for them. Think of each project as a department, like strategy, marketing, sales, or social, and each one keeps its chats focused on that department’s work.
If you’re just getting started, build one project and name it whatever you want (one agent in the room calls hers her business brain). Put your brand guide, voice guide, and key documents in it, and run all your training chats inside it, because the fastest way to lose good work is scattering it across 40 random chats.
Get on a paid plan and change one setting
You need a paid plan for this, and $20 a month is the level I recommend so you aren’t constantly hitting usage limits. Paid plans give you the features we used in the workshop, plus more control over your data.
Then go into settings and turn off model training on your data:
- ChatGPT: Settings, then Data controls, then switch off “Improve the model for everyone.”
- Claude: Settings, then Privacy, then switch off the option to help improve the models.
- Team or company accounts: check with your admin, since most team plans already have this turned off.
While you’re in settings, make sure memory is turned on, because that’s what lets the tool hold on to what you teach it. I had to hunt for the Claude privacy toggle live on screen (I switch between client accounts all day, and they don’t all look the same), so if it takes you a minute, you’re fine.
Step 2: Answer Five Questions About Your Business
This is the prompt we ran live. Paste it into your project and answer one question at a time:
Ask me five questions about my business, one at a time. Wait for my answer before asking the next. When I’ve answered all five, summarize everything you’ve learned about me without missing a single detail and save it to memory.
1. What’s my exact role, and what does success look like a year from now?
2. Who do I serve, and what’s the problem I solve for them?
3. What does my income look like today, and what number would change everything in the next 12 months?
4. What’s the one thing (time, money, people, or skill) most limiting my growth right now?
5. Who am I becoming, not just professionally, but as a person?
Working on a project instead of a business? Swap the word “business” for “project” and run the same prompt.
Talk your answers instead of typing them, using the voice button in the app, because people give far richer detail out loud than they do in a text box. The more stories, examples, and numbers you give it, the better the output, and vague answers get vague results.
Here’s the level of detail to aim for on question one: “I’m a Medicare broker helping people compare plans during AEP. Success in a year looks like 30% more enrollments without working every weekend in October.”
When you’re done, test it. Ask it to write a social post or answer a real client email, tell it what’s off, and keep going. If something seems wrong, ask the tool directly, something like “I just turned on memory. Do I need to answer these again?” It’ll tell you.
Five questions is a starting point, not the finish line. The full question set I share after workshops goes much deeper, and you’ll still want to feed it your best newsletters, your top-performing social posts, and your client stories.
Step 3: Turn Your Best Prompt Into a Skill
A skill is a saved playbook, basically an SOP your AI follows every time it does a specific task. Once a prompt is producing what you want, type “Please turn this into a skill for me,” and ChatGPT or Claude will package it and save it.
That’s the shift from repeating yourself to building a system. Skills work across all your projects and chats; you can call one in by typing a forward slash, and the tool will often pull in the right one on its own when it recognizes the task.
We run skills for building keynotes and workshops, for giving first outreach campaigns, for our SLAC social post format, and for my voice, which covers how I talk, the words I use, and the words I won’t touch, so my whole team gets the same Reb every time. When your process improves, you update the skill, because training never really ends, and a skill you wrote last month should get sharper this month the same way a good junior associate does.
If you’ve been using ChatGPT or Claude for a year or two, you have more playbooks buried in your chat history than you realize. Ask your AI to pull them out and organize them, and you’ll have your first few skills by the end of the week.
Stay ahead of the AI curve: get practical insights delivered to your inbox →
Step 4: Connect Your Tools and Let Your Teammate Work
This is where AI stops acting like a search bar and starts working like a teammate. Connectors let your AI talk directly to the tools you already use, and most of them install from settings in a few clicks, with no developer and no code.
Our Claude is connected to Fathom for meeting notes, Asana for tasks, Canva for design, Gmail, Google Calendar, Google Drive, Slack, and our CRM. The Boomers agents already have a CRM built for them, and that’s the first thing worth connecting.
If you don’t use an AI note taker yet, start there. We use Fathom for video and voice calls and Granola for voice notes on the go, and you can get a lot done on the free versions.
What this looks like in real life: our newsletter
Our newsletter used to take weeks of back and forth, with my coordinator chasing ideas, running them by me, and the team debating topics. Now our newsletter teammate reads our Fathom call transcripts and my voice notes, pulls the best client questions and stories, and drafts four weekly issues in our voice.
It asks me questions before it writes, like which stories to feature and whether to anonymize a client, so a human still makes the calls. Then it pushes the drafts straight into our CRM and messages the coordinator, who does final edits and schedules the sends.
The whole thing now runs on the 15th of every month, and our part takes about 10 minutes. I’m not a tech person, and I built it by talking to Claude in one long chat, which is exactly the point.
One rule doesn’t change: keep a human in the loop. Never let AI send anything you haven’t reviewed, the same way you’d check a new coordinator’s work before it goes to a client.
The Mistakes That Keep AI Sounding Generic
- Tool hopping. Switching platforms every month resets your training to zero, and the agent who goes deep on one tool beats the one who dabbles in five.
- Skipping organization. Training scattered across random chats gets lost, and one project fixes that in about two minutes.
- Thin answers. “I help people with insurance” teaches it nothing, while “I help retirees in my county compare Medicare plans during AEP” teaches it your client, your season, and your stakes.
- Treating training as one and done. Your process changes, so your skills should change with it.
- Automating before training. A connected tool that doesn’t know your voice just produces generic content faster.
- Leaving default settings on a personal plan. Turn off model training on your data before you upload anything about clients.
Your AI Teammate Checklist
- ☐ Pick one AI tool and commit to it
- ☐ Upgrade to a paid plan ($20 a month is the sweet spot)
- ☐ Turn off model training on your data and turn on memory
- ☐ Create one project for all your training
- ☐ Upload your brand guide, voice guide, and three of your best pieces of writing
- ☐ Run the five-question prompt and answer by voice
- ☐ Test it on three real tasks: a social post, a client email, and a meeting recap
- ☐ Tell it what’s off and keep refining
- ☐ Type “Please turn this into a skill for me”
- ☐ Connect an AI note taker like Fathom or Granola
- ☐ Connect one more tool you live in, like your CRM, calendar, or project manager
- ☐ Pick one manual workflow, like your newsletter, social posts, or email triage, and hand the first draft to your teammate
- ☐ Review everything before it goes out
Frequently Asked Questions
How do I train ChatGPT or Claude on my business?
Create a project, upload your brand and voice documents, and run a prompt that interviews you one question at a time about your role, your clients, your income goals, and your biggest growth constraint. Save the summary to memory, test it on real tasks, give it feedback, and turn the version that works into a skill so it follows the same process every time.
Do I need a paid AI plan to do this?
Yes. Paid plans give you projects, memory, skills, and connectors, plus more control over whether your data trains the model. Around $20 a month is the level that keeps you from constantly hitting usage limits.
What’s the difference between a project and a skill?
A project is a workspace that holds your documents and chats for one area of your business, like a department. A skill is a saved playbook for one specific task, like writing your newsletter, and it travels with you across every project and chat.
Is it safe to put my business information into AI?
It’s safer when you set it up correctly. On a personal paid plan, turn off the setting that lets your data train the model, and on a team plan, confirm with your admin that it’s already off. Keep a human reviewing anything before it reaches a client.
Which AI tool is best for a small business or insurance agent?
The one you’ll actually use every day. ChatGPT, Claude, Gemini, Copilot, and Grok can all handle projects, prompts, and training, and the results depend far more on how well you train it than on which logo is at the top of the screen.
Want your whole team building trained AI teammates in one session, the same way the Boomers agents did? Bring REBL Labs in for a working workshop →
