This is a question my wife asks on a somewhat regular basis. And she is justified in feeling like we’ve been misled, especially with Uber’s announcement a few years ago that they’d be introducing flying cars in Dallas. It still hasn’t happened.

Source: Fortune

So am I. Are you???

A lot of agents think they’re just now dipping their toes into artificial intelligence—but the truth is, you’ve probably been using it for years. If you’ve ever asked Alexa for the weather or used Siri to text a client or provide directions, you’ve already experienced AI in action.

What’s available now, though, is something very different—and much more powerful.

Tools like ChatGPT, Claude, and Microsoft Copilot are large language models that can generate content, answer questions, summarize regulations, and organize information at lightning speed. And the most important thing for agents to know? These tools are already built into the systems you’re using—from your browser to your CRM. The only question is whether you’re making the most of them.

So, what makes me qualified to talk about AI?

• In February, 2024, I completed an 11-month Artificial Intelligence and Machine Learning program through UT Austin in partnership with Great Learning, finishing 5th out of more than 100 participants.

• I also completed a three-day AI Immersion Program at UT Austin in August, 2024, learning directly from some of the top AI instructors in the country.

• These courses gave me a valuable “look under the hood” of how AI models work and helped me build a stronger technical foundation.
• But more importantly:
  I’ve been using AI almost daily for more than two years — applying it to real-world projects in both work and personal life.
• The best way to learn AI is to use it.
  Experiment with it for anything and everything—brainstorming, summarizing, rewriting. You’ll quickly see what works and what doesn’t.
• You don’t need a formal class to start.
  We’ll spend some time today discussing prompt engineering, but the best advice is simple:
  Talk to AI like it’s a person.
  Be clear. Be specific. If you don’t like the results, tell it — and it will adjust.
  (Sometimes better than an actual human would!)

AI can be definted as “machines doing smart things.”

It’s an accurate definition, but today’s AI models are much more advanced than what we thought of  as AI just a few short years ago.

Let’s take a step back and define what we’re talking about. The most common type of AI agents are interacting with today is something called a Large Language Model (LLM)—a tool trained on massive amounts of text that can generate human-like responses. Tools like ChatGPT, Claude, and Microsoft Copilot use these models to answer questions, write content, summarize documents, and more.

These tools aren’t pulling answers from a search engine. They’re predicting words based on patterns they’ve seen before. That means they’re fast, smart, and surprisingly helpful—but not always perfect.

📄 Summarizing long or complex information (like plan documents or regulations)

✏️ Rewriting content in different tones or reading levels (e.g., “explain this at a 5th-grade level”)

🗂️ Organizing messy content into outlines, bullet points, or FAQs

📨 Drafting first versions of emails, social media posts, or benefit overviews

💡 Brainstorming ideas, names, or alternate phrasing

🤯 It sometimes “hallucinates”—making up facts or misquoting regulations.

🕰️ It doesn’t know recent news or plan updates unless you feed it that info.

⚖️ It has no built-in compliance filter—you’re still responsible for what it generates.

---

 

That said, it knows a LOT of stuff, and I generally trust its answers.

• Hallucination: There’s not that much hallucination with ChatGPT, at least not right now.
• Recent Info: Even though there’s a cutoff point for its training, the paid version of ChatGPT is able to search the internet and can provide up-to-date info.
• Compliance: And, while double-checking is important, it often knows the info better than we do. In fact, I run a lot of compliance questions by ChatGPT to make sure that whatever source I’ve found is actually correct.

If you’ve ever typed something into ChatGPT and gotten a vague or unhelpful answer, you’re not alone. The output you get depends almost entirely on the prompt you give it.

Think of prompting as giving directions. If you’re too vague, AI will guess. If you’re specific, structured, and clear, you’ll get much better results.

Weak Prompt:

“Explain Medicare.”

Better Prompt:

“Explain the difference between Medicare Advantage and Original Medicare to a 65-year-old client who’s new to the program. Use plain language and keep it under 200 words.”

You don’t need to learn some new coding language, but you do have to tell the AI model what you want.

(That said, you can give it some rope. These models are very creative, and sometimes letting them run with an idea will yield better results. Despite the advice below, give it a try sometime.)

• Give context – Who is this for? What are they trying to understand?

• Specify the format – Bullet points, paragraph, comparison table?

• Set tone or reading level – Professional, conversational, 5th-grade reading level?

• Add constraints – Length limits, word choice preferences, compliance warnings

You can also give feedback like:

“That’s too technical. Can you simplify it?”
or
“Make this sound more like something I’d say to a client.”

Prompting is a skill. The more you practice, the more you’ll get AI to sound like you—and work like an assistant who gets your business.

What I do:

• I will often ask ChatGPT to write an initial draft of an article.
• Then I will ask it to find a couple of reputable sources and add a couple of quotes with citations to the article.
• Then I will ask it how the article can be improved. Even though it wrote it, it will look at it with “fresh eyes” and provide an honest assessment of how it could be better.

The very BEST ADVICE I can give you

Once you get the hang of prompting, AI becomes a powerful tool for handling everyday insurance tasks. Think of it as a smart assistant that never sleeps, never complains, and can generate five variations of the same message in under 10 seconds.

Here are a few ways agents are already using AI in the field:

AI can help you turn complicated topics into easy-to-read explanations.
You can say:

“Write an email to a client turning 65 that explains their Medicare options, including Original Medicare, Advantage, and Part D. Use plain language.”

Then review and tweak the response. It gives you a head start—especially helpful during AEP season.

Ask AI to create side-by-side plan comparisons using data you provide.
For example:

“Compare Plan A and Plan B using these notes: Plan A has a $2,000 deductible, $30 PCP copay, $5,000 max out-of-pocket. Plan B has a $1,500 deductible, $25 PCP copay, and a $2,500 max out-of-pocket.”

AI can turn that into a clean summary or even a chart for your proposal or email.

You can use AI to draft answers to common client questions:

• What’s a deductible?

• How does a telehealth visit work?

• Is my child still covered under my plan after college?

Let AI write a draft, and you can fact-check and personalize it before sending.

If you’re stuck on what to post or how to explain a benefit update, ask AI:

“Write a LinkedIn post reminding employers about the deadline to submit 1095 forms.”
or
“Create a short client-friendly blurb about the new $2,000 Medicare Part D cap in 2025.”

It helps you stay visible without starting from scratch.

---

These are just a few examples, but they all show one thing: you don’t have to use AI for everything—but it can help with almost anything.

I use ChatGPT A LOT, so this isn’t a comprehensive list.

I originally taught this class May 1 for the Austin chapter, and I told them that I was not going to file it for CE credit because the Texas DOI was unlikely to approve it.

I decided to ask ChatGPT if it thought I could get it approved for CE credit. I typed in some of the topics I wanted to cover and told it we needed to give it an insurance slant. It created the outline and overview, and after asking for a couple edits, it was ready to file with the Department of Insurance. (This is not the first course I’ve done this with). I created the basic structure of the course from the outline it created. And it created many of the images.

For weeks, I’ve been certain I’m dying. So certain that I went to a primary doctor (a rarity for me) to get bloodwork done and a back doctor for an MRI. In the follow-up appointment after the MRI, with the doctor’s permission, I recorded the nearly 20 minute visit. Afterwards, I transferred the recording to my computer, hit play, and hit record on the ChatGPT app on my phone. In five-minute segments, I had it summarize that section of the visit, then it wrote a condensed and organized summary afterwards that I could share with my wife. Spoiler alert: It’s possible I might live.

In 2019, I recorded 12 HSA Video FAQs. Each was about 1 to 3 minutes long. We had them embedded on our agency website. They were pretty good, but the numbers were outdated. So I played the videos on my computer and recorded using the Chat GPT app on my phone. I then asked it to clean it up so we could have a written version on our website. I also asked it to update the text with 2025 numbers. I pasted that text on our website, and, for the videos that need to be re-recorded, it will serve as my script.

It helped with the content for multiple pages and helped with some formatting. For example:

• It generated a list of FAQs for the individual and small group pages of the website, then provided answers to those questions.
• It updated the answers for our Medicare and HSA FAQs.
• It put the questions in a logical order.
• It also provided short answers for the FAQ toggle – we then link to the longer answers.

I needed to write an article about final expense insurance. I don’t know much about final expense. So I asked ChatGPT. Here’s the prompt I used (slightly modified so I’d get a new article):

My Prompt:

I need to do a blog post about final expense policies, but I don’t know anything about them. This would be targeted at health insurance agents who have the necessary license to sell these products but may not currently be doing so. Maybe agents who sell Medicare supplements or advantage plans, since they already have clients who might benefit from these products. We could even frame it as a bit of an ethical obligation. Most people don’t have a different agent for each type of policy they might need. And they also may not realize what they need. Basically, if you as an agent don’t bring it up, they very well might go without. Obviously, brokers get paid for the products they sell. That’s why we work. At the same time, we do have a bit of an obligation to make sure we’re asking the right questions and selling products our clients need. A lot of agents sell products clients want, like dental or vision coverage, but we also need to initiate the conversation sometimes. If we don’t speak up, our clients could be uncovered. Talk a little about the repercussions of not having a final expense plan. What might it mean for their loved ones when they’re gone? Can you think of anything else that needs to be in the article? In the last paragraph of the article, please say that ABC Marketing can help agents get appointed with carriers that offer final expense policies and get them up to speed so they’ll be comfortable selling them.

Here’s the first draft of the article that ChatGPT provided:

Why Final Expense Policies Deserve a Place in Every Agent’s Toolkit

Because if we don’t bring it up, who will?

If you’re already helping clients enroll in Medicare Advantage or Medicare Supplement plans, then you’re in the perfect position to offer final expense insurance. These policies—also known as burial or funeral insurance—are simple, affordable whole life plans designed to cover end-of-life expenses. But here’s the thing: most clients won’t ask about them unless you do.

That puts the responsibility squarely on us as agents.

And frankly, it’s not just a missed business opportunity—it’s an ethical issue. Many of our clients think they’ve “taken care of everything” just because they signed up for Medicare. But Medicare doesn’t cover funeral costs. And Social Security only pays a one-time $255 death benefit. Without final expense coverage, a client’s loved ones could be left scrambling to cover $8,000 to $15,000 in funeral and burial costs—not to mention any lingering medical bills.

If You Don’t Mention It, They Might Go Without

Most people don’t have one agent for Medicare, another for life insurance, and another for long-term care. They have you. That’s why it’s so important to be proactive. If we only sell what clients ask for, we’re not really doing our job. We’re consultants, not cashiers.

You already ask questions about prescriptions and doctors. Final expense should be part of the same conversation.
“Have you made any plans for how those costs would be covered?” is a simple, respectful way to bring it up. You’ll be surprised how many clients want to have that conversation—once you open the door.

What Happens If No One Talks About It?

Loved ones may need to dip into savings, max out credit cards, or even start a crowdfunding campaign to cover final expenses. It’s heartbreaking—and preventable. Offering final expense coverage means you’re helping your clients protect their family from stress and financial strain at one of the most emotional times in their lives.

You Don’t Have to Go It Alone

ABC Marketing can help you get appointed with trusted final expense carriers and get you up to speed so you feel confident selling these important policies. If you’re already serving the senior market, adding final expense is a natural next step—and one your clients will thank you for.

My wife recently had an appointment. I took her and sat in the waiting room for about 30 minutes. I saw someone with a shirt that said something about “Bubba”, so I asked ChatGPT to write a short 200 word story about Bubba, Dominoes, and Dynamite. It did so immediately. I decided to keep going, and soon I had a 10-chapter book with illustrations. It was dumb, but I didn’t give it much to work with. The pictures are inconsistent from one chapter to the next. It couldn’t remember what Bubba looked like. Most image creators have that problem. Here’s the first chapter:

Bubba had a strange hobby: setting up elaborate domino trails across his junkyard. Each weekend, he’d design patterns winding over rusted cars, under busted lawn chairs, and through hollowed-out refrigerators. But Bubba wasn’t content with just dominoes. No, sir. He added dynamite. “Adds excitement,” he’d grin.

One Saturday, Bubba invited his cousins to watch the latest masterpiece. “Bubba,” they warned, “you sure mixin’ dynamite with dominoes is a good idea?” Bubba just laughed, lighting the fuse with a shaky hand. The dominoes clicked beautifully, arcing through the yard like magic.

Then—BOOM!

A fridge door landed in a neighbor’s pool. Bubba, singed but grinning, held up a single domino. “Perfection,” he said.

Next day, Bubba was on the news. “Local man Bubba turns junkyard into art—and crater.” Asked what’s next, Bubba winked. “Thinking bottle rockets and mousetraps.”

Because of course Bubba is.

---

This is not the first time I’ve used ChatGPT to write stories.

• Early on, I used it to write 15 children’s books. I haven’t done anything with them yet, but I captured them, and the newer versions will be able to make them even better.
• Two Christmases ago, I chatted with two of my wife’s younger cousins and, based on their input, wrote a multi-chapter, illustrated e-book for them. They loved it. This past Christmas, we wrote the sequel.

I’ve also used ChatGPT to write more than 1,000 articles. No, I’m not kidding. I’ve captured them all on a non-public website. Eventually, I’ll do something with them – perhaps write a book, perhaps fill a public website. We’ll see.

I currently have about 100 hours approved for CE credit.

Here’s my plan:

• Select courses that are somewhat timeless (where the numbers don’t have to be updated every year)
• Play the recording five minutes at a time – have ChatGPT transcribe it.
• Ask ChatGPT to clean it up without changing the meaning or flow.
• Have it then format the entire course and suggest ways to improve it.
• Ask ChatGPT to create multiple-choice questions based on the course content.
• Post the recording and text so people have a choice about how they would like to learn.
• Eventually… go back and re-record the courses in modules so people can watch five or ten minutes at a time rather than watching a one-hour videos. For new courses, I’ll begin by recording this way.

Once you start using AI regularly, you’ll notice something:
It doesn’t always remember your style or preferences from one conversation to the next.

That’s where Custom GPTs come in.

A Custom GPT lets you set instructions once—things like your tone of voice, your preferred formatting, and the type of answers you want. Then, every time you use it, the AI responds according to those saved preferences without you having to repeat yourself.

You could create a Custom GPT that always:

• Writes emails in a friendly but professional tone

• Summarizes plan differences in 3–5 bullet points

• Uses simple, client-friendly language unless you ask otherwise

• Consistency: Your AI output will match your voice and expectations every time.

• Efficiency: You won’t have to waste time “retraining” it for every new task.

• Better client experience: Materials will feel like they came from you, not a robot.

---

You don’t have to create a custom GPT to use AI effectively—but if you start leaning on it for client-facing work, it’s a powerful shortcut that saves time and makes the experience more seamless for you and your clients.

I personally created a  Custom GPT when I needed to move 20 years of client notes from one database to another. When pulling them out of ACT, they were ugly and contained a lot of extraneous html. ChatGPT was able to remove the unneeded coding, format all notes the same, and even correct grammatical errors. This would have been impossible for a human to do.

As powerful as AI can be, it’s important to understand its limitations—and the risks that come with using it without careful oversight.

One major lesson: AI can change without warning.

In early 2025, for example, OpenAI made updates to ChatGPT’s behavior that significantly altered the way it responded.

• Outputs were less detailed.

• Responses felt more vague and less consistent.

• For some users, the tools they had built workflows around suddenly weren’t reliable anymore.

This change lasted for six weeks, disrupting many businesses (including yours) that had integrated AI into daily operations.

My cousin HATES Siri and Alexa. He says they can’t get anything right. He has a point…

 

It reminds me of the Jack in the Box commercial where a phone someone has handed Jack says “I think you said… [something totally incorrect]. He hands it back and says “I like things that work.”

Well, starting around February 1 and continuing to mid-March, ChatGPT was WORSE than Siri and Alexa. Way worse.

For example, here’s a back and forth conversation between me and ChatGPT about Gene Hackman’s death.

And while the issue is fixed now, there are still problems. I recently asked about the Pope’s death. ChatGPT provided two answers for me to choose from. Side by side, one confirmed the Pope’s death, the other denied it.

 

---

I wasn’t the only one who noticed. There was a very lively discussion on the OpenAI website about it. Some comments:

• I don’t get it. This latest update has completely ruined the way ChatGPT interacts, and I’m shocked that more people aren’t complaining. Custom instructions are now basically ignored. I had a specific way I wanted responses to be structured, and now ChatGPT just does whatever it wants. It used to follow my preferences consistently, but now it feels like a completely different (and much worse) model.
• I am genuinely shocked at what they’ve done. All it’s writing looks like some health guru blog post trying to convince you to buy probiotics. There’s random bolding and text changes literally everywhere and I’m finding it generally impossible to read and it’s made it less accessible for my brain to use. I tell it not to do it, but it keeps doing it, it can’t help it.
• OMG whatever this latest update is, its absolutely terrible. Before when asking it to find errors in code it would always spot the issue for me. Now it wont find the most basic problem. Its slower and all the formatting and bold font and icons it spits out are just … awful. How do we roll back to the previous version?
• Its the worst I spend hours creating overviews and now the format is all messed up. Bold in the middle of sentences new Icons that are hard on the eyes. Plus doesn’t remember how we did stuff last week. WTF. I am a paying subscriber as well. I don’t want to spend hours and hours to get it to try to do what was seamless last week.
• So I’m not alone in this! It’s horrible and totally robotic!
• I am shock truly. Today February 3rd, i wanted to continue my project, I am a paying subscriber GTP plus. And the update is like going back to GTP 3,5. I am totally in shock, truly. All the work done these few weeks: gone. The system is not stable. I am in shock. I feel terrible and do not know how to proceed.
• It got worse with each successive ‘update’ it’s practically unusable now, considering the money they’ve got off me I expect better!
• I use it for work, and l’ve noticed a frustrating decline in its performance. It used to pick up on nuances, responding with sharpness and precision. Now, it feels like it’s just generating random nonsense. It repeats itself, doesn’t check my screenshots by default (even though we’re constantly working on my report), and ignores my requests-l asked it to merge slides, and it didn’t. It used to help me manage burnout at work, but now it’s just one more idiot in the office.

---

Here’s an example of what it kept doing when I would ask it to perform a task.

Basically, I would ask it to write something, and it’s response would be “Got it! If there’s anything specific you’d like to adjust or elaborate on, just let me know!”

It reminded me of the mad tea party Alice in Wonderland.

“Have some wine.”

“I don’t see any wine.”

“There isn’t any!”

 

---

So, why did I stick with ChatGPT instead of switching to another AI model? Fair question. I guess because I trusted them. I figured this was simply an update gone bad and that they would fix it. But it took a full six weeks. By the end, I was becoming a doubting Didymus.

I’m happy to report that it’s mostly back to normal, and there is a new 4.5 version that’s slightly better than the 4.0 version, though there is a limit to the number of responses it will provide before making you wait ’til the next day.

They’ve also seemingly solved a HUGE problem with AI – problems with text as images. Many of the text images in this presentation were created with Dall-E 3 (from OpenAI). Midjourney is still much better at creating images, but the ability to use ChatGPT for presentation images with words actually spelled correctly is really nice.

• Model changes: AI companies update their systems regularly, sometimes without announcing changes in behavior.

• Inaccurate content: AI can “hallucinate” facts, misinterpret prompts, or create plausible-sounding but wrong answers.

• Overreliance: If you depend too heavily on AI for client communication or compliance-sensitive tasks, you could expose yourself to regulatory or reputational risk.

• Data concerns: Be careful not to input sensitive client information into public AI platforms unless you are confident in their privacy protections.

• Always verify important outputs. Double-check plan details, compliance explanations, and regulatory references.

• Save key prompts outside the platform. Maintain your own prompt library in case you need to switch tools.

• Experiment with multiple tools. Don’t put all your eggs in one basket—diversifying keeps you agile if a model changes.

• Remember: You’re still responsible. AI can be an amazing helper, but the agent must always review and approve final content.

---

Bottom line: AI is a tool—but it needs human supervision.
If you understand the risks and build good habits, you can use AI safely, efficiently, and confidently.

AI isn’t just a cool add-on anymore—it’s going to be deeply woven into the way health plans operate, the way agencies serve clients, and the way consumers access care.

Here are three major changes we expect to see over the next 3–5 years:

• Members will enter their symptoms into an AI tool that will suggest the next steps:

  • Home care or OTC medication

  • Telehealth visit

  • In-person primary care

  • Urgent care or ER

• Why it matters:
  AI could help reduce overutilization and underutilization, guiding people to the right level of care—saving money for both the insurer and the member.

• Agencies will build internal AI systems that know their products, their procedures, and their carrier partners.
• These bots could:
  • Answer common client questions instantly
  • Help staff draft plan summaries or client communications
  • Support new agent training and onboarding
• Why it matters: Having a bot that knows your book of business will save time, improve accuracy, and help deliver better service at scale.

• AI will help brokers and employers:

  • Analyze claims trends

  • Identify gaps in care

  • Recommend plan design changes

• Instead of just looking at last year’s renewal spreadsheet, agents will bring smarter insights to the table.

• Agents who avoid AI may find themselves at a serious disadvantage:

  • Clients will expect faster answers and more personalized service.

  • Agencies using AI will operate more efficiently and with better insights.

  • Falling behind could mean losing business to competitors who embraced these new tools.

The future is not about AI replacing agents—
It’s about agents using AI to deliver better advice, faster service, and greater value.

The only way to get comfortable with AI is to use it.
Reading about it is helpful. Watching a demo is helpful.
But nothing replaces hands-on practice.

Here’s your homework:

 

It’s like $20 per month plus tax.

No, I do not think you should get the free version. The model isn’t as good and it does not have all the bells and whistles.

If you won’t pay for the good version, you’re not taking this seriously and will conclude that AI isn’t all that useful.

• Draft an email explaining a complex insurance topic (like HSA eligibility rules).

• Create a client FAQ for common questions (like “What is a deductible?”).

• Summarize a benefit comparison between two plans.

• Is the information accurate?

• Is the tone appropriate for your clients?

• What would you edit or clarify before using it?

• Ask for different tones: formal, conversational, 5th-grade reading level.
• Try asking for bullet points, paragraphs, or side-by-side tables.
• See how the AI responds when you’re more (or less) specific.

• Save prompts that work well.

• Tweak them for different clients or scenarios.

• Over time, you’ll develop go-to templates that save you hours.

Create some fun images. Have it write a story. Play with it with your kids. Or do like I did and show it to your 97-year-old Aunt. She’ll love it.

AI isn’t perfect—and it never will be.
But neither are humans.
The agents who learn to work with AI—guiding it, checking it, improving it—will be the ones who succeed in the next phase of our industry.

Don’t be afraid of it.
Practice with it.
Get better at it.
And most importantly—use it to serve your clients better.

---

View this course at: https://comedyce.com/is-ai-useful-that-depends-on-the-user/