Can an AI read my Annual Notice of Change for me? Five prompts
September 26, 2026
Listen Now
Subscribe
Ready to talk with an advisor?
The advisors at American Retirement Advisors have been having these conversations for over 25 years. A no-cost discovery session is the fastest way to find out where you actually stand.
Schedule a Discovery Session
Show Notes
Eddie and Betty's Conversation
Welcome to The American Retirement Advisor. I'm here with Eddie, as always, and today we are talking about something I did not expect to find genuinely useful: using an AI assistant to read your Medicare plan's Annual Notice of Change. That letter that shows up every fall, the one most people drop on the counter and ignore? Turns out there is a right way and a very wrong way to use a chatbot with it, and Ian Schaeffer laid the whole thing out in a piece that honestly surprised me.
That tracks for me, and I'm the one who reads everything. Ian is the company's COO, and he's also an engineer who uses AI tools all day. So when he writes a how-to on this, it's not a tech person guessing. It's someone who does it.
And he's upfront that he's not an advisor. Which I appreciated. He's showing you the workflow, not telling you what plan to pick.
Right. And it starts with that letter. Medicare's own guidance says your current plan has to send you the Annual Notice of Change by September 30, so you've got it in hand before the October open enrollment window opens.
Okay so you've got the letter. Most people glance at it, see a bunch of tables, maybe notice the premium went up, and that's about it.
And that's where the AI thing is interesting, because Ian's point is that an assistant with the letter in front of it can do things most of us just won't sit down and do. Find every line that changed. Explain what a section means. Do the arithmetic on your specific situation.
That's the part I want to talk about, because he has this step where you basically describe your own year to it. Like, how many specialist visits, what tier your medications moved to, and it runs the whole calculation for you.
The example he runs with is eight of those and one daily medication that went from tier two to tier three. Those are just numbers he pulled from an earlier piece in the series, not a recommendation for anyone. But the idea is you swap in your own numbers and it shows the arithmetic line by line.
Which is the thing I never do because I don't want to do the math.
Nobody does. That's the point.
So before we get into the five prompts, there's this other piece of the article I cannot stop thinking about. The AARP stat.
Go ahead.
AARP surveyed adults fifty and older in February 2026, and sixty-nine percent said they do not feel comfortable sharing their personal health information or medical documents with an AI tool. Nearly seven out of ten.
And Ian's response to that is basically: that instinct is correct, and also you don't have to share any of that to make this useful.
Wait, really? Because I assumed the whole point was to hand it your documents.
Here's the thing. The useful part of the letter, the Summary of Important Costs table, is the same for every single member of the plan. Your name isn't on it. Your Medicare number isn't on it. It's a table. So you cover the front page where your name and address are printed, and you just photograph or type the table.
So you're not giving it anything personal. You're giving it the same thing every other person on that plan got.
Exactly. And Ian is very clear about what you should never paste in. Your Medicare number, your Social Security number, date of birth, street address, the member ID on your card. He quotes Medicare's own fraud guidance, which says you share your number only with your doctor or people you know should have it. A chatbot is not on that list.
I love that he put that in writing, because I think a lot of us would just paste the whole letter and not think twice.
Cover page goes face down. Table gets typed in. That's the move.
Alright, walk me through the five prompts, because this is where he gets very specific and I want people to hear it.
Before any of them, there's one sentence you type first, every session. And this might be the most important sentence in the whole article. You type: Use only the text I paste. If something is not in the text, say so.
That one line.
Ian calls it the difference between an assistant that reads and an assistant that improvises. Without it, the AI might fill in gaps with things it learned from somewhere else, and you'd have no way to know.
It sounds like telling a contractor: only use the lumber that's on site. Don't go buy something and charge me for it.
That's a pretty good analogy. Then the five prompts, in order. First one: find what moved. You paste the table and ask it to list every line where next year's number differs from this year's. Show both numbers and the difference.
So you're not reading the whole table yourself hunting for changes. You're asking it to do the comparison for you.
Right. Second prompt: translate a section. You point it at a specific section, like Changes to Part D Drug Coverage, and you ask it to explain it in plain English. And here's the key part of that prompt: for each point, quote the exact sentence from the letter you're relying on.
So it has to show its work.
It has to show its work. Third prompt is the arithmetic one we mentioned. Fourth prompt is one I really liked: write five questions I should ask a Certified Medicare Planner about this letter, in order of how much money each could involve.
Oh that's clever. You're showing up to the meeting with an agenda instead of a sealed envelope.
That's close to how Ian puts it. A client who arrives with five ranked questions and a page of arithmetic gets a better meeting than one who walks in with nothing.
And the fifth prompt?
This is the one Ian says people skip, and the one he would not. Ask it to list everything in the letter it cannot evaluate without information it doesn't have, and tell you what that missing information is.
So you're asking it to tell you its own blind spots.
And he says a good assistant answers that with a real list. A poor one says nothing, and now you know which one you have.
I want to stay on that for a second because he spends a good chunk of the article on where it will go wrong. And some of these surprised me.
Three specific failure points.
The first one is about other plans. So you've been reading your current plan's letter and you ask the AI: are there cheaper plans in my county next year?
And the correct answer is that it cannot know. Because carriers aren't allowed to market next year's lineup before October 1, and Medicare's comparison tool switches over the same day. So any assistant that confidently names other options in September is describing this year's coverage.
Which is like asking someone what's on next week's dinner menu and they just read you what was on last week's.
Pretty much. Second failure point: ask it whether your doctor is in the plan next year. The letter itself says the network changed and tells you to check the 2027 directory. But the directory is a separate document. The AI doesn't have it.
So it can tell you the network section exists and that you should do that, but it can't actually do it for you.
Correct. Third one is about your prescription cost at your specific pharmacy. The letter gives you tiers, not what the actual counter will ring up. And Ian dug into why that difference matters quite a bit back when he first laid out the tier framework, so if that gap surprised you, that earlier installment is worth a second look.
So tiers tell you the category, not the dollar amount you'd actually pay at the counter.
Right. And a good AI assistant, if you've set it up with that opening sentence, will tell you it can't answer all three of those. If it answers them confidently, Ian says it just failed a test.
He frames it as a test. I thought that was a great way to put it. You're not just using the tool, you're evaluating it at the same time.
And you don't need any special knowledge to run the test. Just ask those three questions and see what happens.
Now there's something else in here that I think a lot of people might run into and not even realize it's different. What if the letter you get isn't an Annual Notice of Change at all? What if it says the plan is ending?
Ian addresses that. If the first page says your plan won't be offered next year, you're holding something categorically distinct, with its own set of rights. He covered the specifics of that in an earlier part of the series, what it triggers, what doors it opens. But his advice here is: paste that sentence in first, tell the assistant to read everything else in that light, and then call a human, because that document has a clock on it.
Meaning there's a window you have to act in.
Right. The exact deadlines and what options open up when a plan ends, that's a question for one of our advisors at American Retirement Advisors. That's the kind of thing where the specifics really matter and you don't want to guess.
Something I want to go back to, because it surprised me: the government's own take on this. Because I expected the article to say the government is skeptical of AI tools for Medicare.
The opposite. In February 2026, they posted a request for information on the federal contracting site specifically asking about conversational AI tools for plan selection.
Wait, Medicare was out there actively looking for this kind of tool?
And the reasoning they gave was pretty blunt. Current tools rely on static comparison tables that are difficult to navigate, particularly for people with limited health literacy, language barriers, or cognitive challenges.
The government basically described the very letter you're holding as hard to read.
Ian's line on that is: reading it with a tool that quotes its sources is not cheating. It's the intended use.
I want to put a bow on the section numbering piece because Eddie I think this is useful and people might not know it.
So every plan builds its letter from a government template, the Model Annual Notice of Change. The sections are numbered the same way across plans. On a Medicare Advantage plan with drug coverage, 1.1 is the premium, 1.2 is the out-of-pocket maximum, 1.3 is the provider network, and it goes from there through the drug list and drug costs.
So if you ask the AI a question and it can't point to a section number and quote a sentence, what does that mean?
It didn't come from your letter. That's Ian's line. The receipt is what you need. No receipt, no trust it.
That is such a clean rule. I feel like that applies to a lot more than just Medicare letters.
It really does. Show me where you got that. If you can't, we're not done.
And he mentions, if you don't want to use your actual letter, Medicare publishes the template that every plan builds from. So you can practice on that with no risk at all.
Which is a nice on-ramp if someone is new to this and just wants to get comfortable with the process before October.
Now the piece ends with what a Certified Medicare Planner does that the AI can't. And I think this is important to say out loud, because nothing in this article is suggesting you skip the human.
Not even a little bit. The way Ian describes it: the planner reads the same table you just had the assistant read, and then reads the two documents the assistant didn't have. Next year's provider directory against your specific doctors, and next year's full plan list for your county the day it goes live on October 1.
And they know your pharmacy.
And they've opened hundreds of these letters. So they know which section the surprise is usually hiding in before they even start.
So the AI does the prep. The human does the judgment.
That's a clean way to put it. And you show up to that meeting with the five questions already ranked, which means you spend the meeting on strategy instead of on reading the table out loud to each other.
There are workshops coming up for anyone who wants to do this with a real person in the room. Friday, October 2nd, two locations. Foothills Library in Glendale at 10:30 in the morning, and Mustang Library in Scottsdale at 2:30 in the afternoon. These are at no cost to you. You can register at 123easymedicare.com slash medicare-workshop, or just call 877-220-1089.
And if you want to talk to someone directly, the number for American Retirement Advisors is 602-281-3898. That's in the article too.
The big picture of Ian Schaeffer's piece is really this: the letter is hard, the AI is pretty good at the parts the letter is hard for, and there are three specific things it cannot do no matter how good it is. Know those three limits and you know exactly when to hand it off.
That little sentence we mentioned earlier -- the one you type before anything else -- it's pulling a lot more weight than it might look like.
Dry understatement from the numbers guy. Shocking.
I contain multitudes.
If this episode got you thinking about your own letter, or if you're not sure what letter you're even going to get this fall, that is the exact right moment to sit down with a Certified Medicare Planner. Don't wait until November when the decisions are already made. The advisors at American Retirement Advisors can look at your specific situation and your specific county and tell you what your options are. That's what they're there for. Thank you so much for listening today, and we'll see you in the next one.