Start here
In about thirty minutes, you will tighten a voice profile, test AI on work you already finished, watch it fail safely, and use it on one real task you still owe.
Most people in this field have already tried it. The Alliance's 2025 survey found that 95 percent of CME and CPD professionals use AI at work. What is missing is not access. It is time, a checking habit, and a way of working that still holds up when someone asks who wrote the objective and whether a learner saw promotional language.
This introduction is for that Tuesday: what AI does well, where it fails, and the workflow that keeps you author of record.
What these things actually are
A language model is trained on a huge amount of text. It then predicts what words come next, based on patterns in that training and whatever you just gave it. That is enough to understand the central risk.
Because it learned from how people write, it's very good at shape. The structure of a document, the tone, the rhythm of a professional email, turning notes into prose. And because it's predicting plausible text rather than looking up verified facts, it can be unreliable with facts. It may produce a polished citation that does not exist, and present incorrect and correct claims in the same fluent style.
That's the whole idea. There isn't a deeper layer you're missing, and you don't need to know how it works inside any more than you need to know how a search index works to search well.
If you tried it and quietly stopped
Most people in this field have already been through some version of this. Naming it helps you treat a disappointing result as a workflow problem, not a verdict on the technology.
The first tries feel uncanny. You ask it something and it answers fluently. It feels like a big deal. You tell a colleague.
Then real work is a letdown. You give it an actual file and get back something generic, faintly corporate, and wrong in one important place. You decide it doesn't understand your field and quietly stop. This is the point where a better workflow matters.
It becomes useful when you change the ask. Not because the tool changed. You stop asking it to know things and start asking it to do things with material you supplied. That's the entire shift, and it's what this handbook is for.
Your first twenty minutes
Do these four in order, right now, in whichever tool you have. The third one is designed to fail on purpose.
Make it sound like you. 5 minutes.
Open your sent folder. Choose three ordinary emails with no sensitive content, then remove names, dates, project details, signatures, and contact information before using them:
Here are three emails I've written. Describe my writing voice back to me: how I open and sign off, sentence length, how direct I am, how formal I am, and what I never do. Then turn that into a short set of instructions you could follow to write in my voice. [paste three emails]
Save what it gives you. That's your voice profile, and it's the most reusable thing you'll make today. It goes at the top of every writing request from now on.
Give it something you already finished. 5 minutes.
Take a document you completed and were reasonably happy with. Ask:
Name the three weakest things about this document. Don't tell me what works. Assume a doubtful reviewer who thinks it overstates its case. What do they go after first? [paste your document]
You're not hoping it is right. You are testing whether any criticism helps you see the work differently. Keep what is specific and defensible; discard what is generic or unsupported.
Notice also that you had to ask for criticism. If you'd asked "is this good?", you'd have been told it was good. That pull toward agreement never goes away, and you have to work around it.
Watch it fail. 5 minutes.
This is the important one. Pick a topic you know well and ask:
Give me three peer-reviewed references on [a topic you know thoroughly], with authors, journal and year.
Now go and look for them. Depending on the tool and whether it searched the web, the references may be real, may not support the claim, or may not exist. The point is to practice verifying, not to guarantee a failure on demand.
Do this once, deliberately, while nothing is riding on it. It's the most valuable five minutes on this site because it makes the need for source verification concrete. That reflex can prevent a fabricated or misapplied reference from reaching learners.
Do one real thing. 5 minutes.
Go to the faculty email recipe and use it on an email you genuinely need to send today. Paste your voice profile from step one at the top.
Not a practice exercise. A real email, actually sent. The value shows up in your actual work or it doesn't show up at all.
The habit that makes AI useful
Stop treating it as a search engine that talks. Start bringing the material.
People ask it what the requirements are, what the evidence says, what the prevalence is, what their accreditor expects. And it answers, fluently and plausibly, sometimes right and sometimes not, with nothing to tell the two apart. Then they either get burned once and drop the whole thing, or, worse, they don't get burned for a while.
The habit that replaces it: bring the material. Your gap statement, your evaluation data, your last report, your notes, the guideline you already found. Then ask it to reorganize, summarize, critique, give you options, or write a first draft. That's what every recipe here does, and it's the difference between the tool being a liability and being the most useful thing in your week.
Three things to go and find out
One email to your privacy officer, compliance lead or IT contact settles hundreds of future decisions:
- Which AI tools has my organization licensed and approved for work?
- Is any of them cleared for confidential or regulated information, and which kinds?
- Who do I ask when I'm not sure?
Ask by email so you have the answer in writing, then send the reply to your team. You won't be the only person who wanted to know and didn't want to ask.
Where to go next
If you're a CME coordinator or planner: start with faculty email for the volume work, then the weekly RSS packet if grand rounds is your Tuesday, then evaluation comments for the big file.
If you write the accreditation reports: go to the outcomes report and the disclosure log. The log is what makes "you are author of record" visible in the activity file. Year-end numbers start at the PARS close companion, not in a chatbot.
If you work in leadership development: rehearsing a hard conversation and the facilitator guide. The leadership desk lists the three files.
If you consult or write grant-funded education: start with turning needs into objectives, then the IME needs narrative so supporter language never becomes the gap. Open the activity-file audit before you draft a self-study, and use supporter outcomes when the 90-day report is due.
If a learner asked whether something counts: use the credit-count process and never answer yes or no for a board you do not speak for.
If you'd rather understand ACCME first: the plain-language guidebook covers the Standards, the activity file, Moore's levels, RSS, PARS, enduring materials, joint providership, and the 2026 AI rules.
If you just need the three files for this week: answer two questions or start from your desk. The five skill modules take a little over an hour.
Start here · bettermethod.ai/start-here/ · Last reviewed August 2026
These tools change often. Check the review date before you rely on anything
product-specific in here, and check the guidance against your own accreditor and
organization policies.