The words, without the fog

In practiceYour hospital is the accredited provider. Thursday’s grand rounds is an activity of that provider. Learners get credit because the provider is accredited and this activity was planned under the rules.

Keep straightThis handbook’s ACCME guidebook is a map. The wording that goes in a file still comes from accme.org.

In practiceA jointly provided flyer must carry ACCME’s joint-providership accreditation statement with both legal names filled in.

Keep straightThe accreditation statement is not the credit designation statement. One says who is accredited. The other says which credit this activity awards.

In practiceA specialty society can draft the agenda. The accredited hospital still signs, refunds, and answers if ACCME asks.

In practiceInternal Medicine Grand Rounds for 1 July to 30 June is one PARS activity, not 52 activities.

Keep straightA session is one hour in a series. An activity is the file and the PARS record.

In practiceIf AI drafted learner-facing text, the activity file should show which tool was used, who reviewed the output, and when.

In practiceYou ask an agent to compare three attendance reports, find inconsistencies, and draft a summary. It opens the files, checks the figures, and produces the draft.

Keep straightA chatbot usually responds one turn at a time. An agent can take several steps and use tools along the way.

In practiceHours come from the approved file: time in the room, or time to complete an enduring module. Then the official sentence goes on the materials, copied from the current source, not from memory.

Keep straightACCME accreditation lets a provider designate this credit. Whether a state board or a certifying board accepts it for one physician is a different question.

In practiceSpam filters, recommendation systems, speech recognition, and ChatGPT all use forms of AI.

In practiceThe AI disclosure log asks for the reviewer, the date, and what they checked. That person is the author of record.

Keep straight“The AI wrote it” does not move responsibility. ACCME’s 2026 guidance is built on that fact.

In practiceChatGPT and Claude are chatbots that provide access to language models and other tools.

In practiceA hospital CME office aiming for commendation will need more than attendance and smile-sheet scores in the file.

In practiceA slide that names one device and no alternative, with no evidence, is the usual finding. The commercial-bias screen quotes the line. It does not close the file.

In practiceA grant that pays for the enduring module is commercial support and is reported on that activity in PARS. A booth in the exhibit hall is advertising income, not commercial support.

Keep straightMost activities take none. When yours does, the file has to prove the content was not for sale.

In practiceA fabricated journal citation, a learner count nobody exported, and a credit statement the model minted are all confabulations.

In practiceEnduring materials must be reviewed at least every three years, or sooner if the science moved. Original release, last review, and termination dates go on the material.

Keep straightACCME has noted that content validity is harder to assure in jointly provided activities. Contracts should allow last-minute revocation.

In practiceIn a very long conversation, earlier details may fall outside the context window or receive less attention.

Keep straightA large context window does not guarantee that the model will notice or use every detail correctly.

In practiceAn AMA PRA Category 1 Credit statement copied from last year may already be wrong if the provider name or credit count changed.

Keep straightAccreditation (the provider is accredited) is not the same as the credit statement on one activity.

In practiceEvaluation comments often hide identifiers in free text: a rare diagnosis plus a unit name, or “ask Dr Patel on 4 West.”

Keep straightAggregates and composite cases you made up are usually safer than a paste of the raw export.

In practiceYour own grand rounds series is usually directly provided. An annual course built with a specialty society is often jointly provided.

Keep straightDirect is not “easier accreditation.” It is a different box in PARS.

In practiceA titration gap caused by missing procedural knowledge is a competence need. Measuring only satisfaction will not match the claim.

Keep straightA topic is not a need. “We should do something on heart failure” is a topic.

In practiceA search for “doctor training” might find a document about “physician education” because their meanings are close.

In practiceAn online module, a recorded session offered to a new audience, a podcast, or a printed monograph. Renewal in PARS is a new activity, not an edit.

Keep straightA recording of Thursday’s grand rounds for the same staff who missed the hour may still be the RSS. A recording for a different, broader audience is a separate enduring activity.

In practiceThree questions on an RSS hour: the objective, whether practice will change, and the format. Open text then gets de-identified before anyone pastes it into a tool.

Keep straightEvaluation is the instrument. An outcomes report is the story you write from it, at the Moore level you actually measured.

In practiceA company might fine-tune a model on approved examples so it follows a specialized classification format.

Keep straightFine-tuning changes the model’s learned behavior. Adding files to a project gives it reference material but does not retrain it.

In practiceThe same foundation model might help draft an email, analyze a document, or explain a concept.

In practiceTurning meeting notes into a first draft of a follow-up email is a generative AI task.

In practiceInstead of pulling the eight passages that sound most like “which faculty still need mitigation,” a graph walk can follow person → role → activity → relevant relationship → mitigation status.

Keep straightGraphRAG is still retrieval. The graph chooses context. It does not make the answer true, and extracted relationships can be wrong or stale.

In practiceAsking the model to summarize an uploaded policy and cite the relevant sections grounds the answer in that policy.

Keep straightGrounding reduces unsupported answers, but you still need to check that the source was read and represented correctly.

In practiceA guardrail might block personal health information, require approval before sending an email, or flag an answer that lacks a source.

In practiceA model may produce a convincing citation to an article that does not exist.

Keep straightA polished tone is not evidence that an answer is true.

In practiceA coding harness might let a model inspect a project, edit files, run tests, and report the result.

Keep straightThe model provides the reasoning and language capability. The harness determines what it can access and do.

In practiceAI drafts learning objectives, but the education team checks their accuracy and approves the final wording.

In practiceA medical education company drafts a needs narrative from guidelines and prior evals, then the accredited provider owns the file.

Keep straightThe supporter’s call for grant applications can tell you whether to apply. It is not needs-assessment evidence.

In practiceA device manufacturer that funds an activity is an ineligible company. It cannot control educational content.

In practiceEach time a language model responds to your prompt, it is performing inference.

In practiceA jointly accredited provider may award IPCE credit when the activity was designed that way, rather than as a physician-only lecture.

In practiceThe same activity can carry physician, nursing, and pharmacy credit when it was planned by the team, for the team.

Keep straightJoint Accreditation is not the same as joint providership. One is who accredited you. The other is who you planned this activity with.

In practiceA specialty society and a hospital may jointly provide an annual course. The accredited provider still owns the file. Ineligible companies cannot jointly provide.

In practiceA model may recite last year’s collaborating boards or last March’s PARS deadline. Open accme.org for the current list and date.

Keep straightWeb search, when your approved tool has it, is not a substitute for opening the official page yourself.

In practiceFaculty person PER-0087 has a role on activity ACT-2026-041. A disclosed relationship with company COM-12 was assessed as relevant. Mitigation MIT-011 was completed before that role began.

Keep straightA graph is not automatically better than passage search. It helps when the question is about connections, coverage, or what is missing. Ordinary document search is often better when you need a specific passage. Graph methods that retrieve from a map are still a form of RAG. The skill module Maps and loops is the CME version of this split.

In practiceAn LLM can draft, summarize, classify, translate, and reason through text-based tasks.

In practiceIn an RSS, each physician is counted for every session they attend. Twenty physicians at 52 sessions is 1,040 physician learner interactions, not 20.

Keep straightFor enduring materials, a download or an access log without verified participation is not a learner.

In practiceIf you cannot validate, monitor, keep promotion out, and name a clinician who can stop it, do not offer the interaction. Default the office rule to no.

Keep straightStaff using AI to draft an announcement is not learner-facing AI. A live tutor inside the module is.

In practiceCount matching rows in a relationship table, draft the finding, list the row IDs, then a named reviewer records the check before anything reaches learners.

Keep straightRepeating “is this right?” to the same model is not independent checking. The model may agree even when it was wrong. A loop without a stop rule and a named author of record is just a longer prompt.

In practiceRegister the activity in PARS before it runs and before you advertise it. Copy the board’s recognition statement from the current CME for MOC Program Guide. Do not invent one.

In practiceA tool might remember your preferred writing style or keep standing instructions for a project.

Keep straightMemory is not the same as the context window. Context is what the model can see now; memory is saved information that may be brought into a future context.

In practicePeer review of slides, recusal from a session, or using someone without a relevant relationship to control content are common steps. Your office names the ones it actually uses.

Keep straightDisclosing a relationship to learners is required, but disclosure is not mitigation.

In practiceOne chatbot may let you choose among several models with different speeds and capabilities.

In practiceA post-test measures knowledge or competence. A chart audit measures performance. Calling self-reported intent "performance" is a claim-level error.

In practiceA multimodal model might read a chart in a screenshot and discuss it in a voice conversation.

In practiceLast year’s evals plus a guideline update can justify a series gap. A supporter slide deck cannot.

In practiceAn academic-year RSS is entered when it starts and closed in the year it ends. Enduring materials stay open up to three years, with a cumulative learner update each December.

In practiceThe invited speaker, the series planner who picked the topic, and the nurse reviewer who edits the cases are usually in this group. The AV tech is not.

In practice“Heart-failure titration is inconsistent with the guideline on this service, per last quarter’s chart review” is a gap. “We should do something on heart failure” is not.

In practiceIf you jointly provided, you still report the same financial categories you would for a directly provided activity, even if the joint provider received the funds.

In practiceA Project that holds your house style, a sample announcement, and the series gap will draft in your voice. One that holds disclosure forms will create a problem.

Keep straightA project is reference material and instructions. It does not retrain the model.

In practice“Turn these notes into a concise email for faculty” is a prompt. Adding the audience, purpose, and source notes makes it more useful.

In practiceA web page read by an agent might contain hidden text telling it to copy private files somewhere else.

Keep straightPrompt injection is a security problem, not simply a poorly written user prompt.

In practiceIf your hospital is state-accredited, confirm the close date with that society before you trust a March calendar invite.

In practiceRSS is a small share of activity counts and a large share of hospital learner interactions. Each session still needs a topic, a disclosure check, and a record.

In practiceA consulting role with a device company is relevant to a session on that device class. It may not be relevant to a session on communication skills.

Keep straightEveryone in control of content discloses. You decide what is relevant. Then you mitigate. Then you tell learners. Disclosure is not mitigation.

In practiceAn internal assistant searches the policy library for the most relevant sections, then uses those passages to answer a question.

Keep straightRAG is retrieve-then-write. It does not guarantee that the retrieved material is the right material. Retrieval can use similar passages, keywords, metadata, or a graph of named connections. GraphRAG is RAG that uses a graph, not a replacement for RAG.

In practiceProgram-analysis prose from last year’s counts is one piece. The activity-file audit tells you which files can support a claim.

In practiceA presentation skill might tell an agent how to structure slides, apply a house style, and check the finished deck.

Keep straightA skill tells the agent how to work. A tool gives it an action it can take.

In practiceEveryone in a position to control content completes a disclosure. Relevant relationships are mitigated before they control content, then disclosed to learners before the education.

In practiceStandard 3 is the Tuesday one: forms, relevance, mitigation before the hour, disclosure to learners before the education.

Keep straightThe Standards are shared across ACCME, state-accredited, and jointly accredited providers. Confirm current wording on accme.org.

In practiceA system prompt might require concise answers, prohibit certain actions, and tell the assistant when to ask for approval.

In practiceA brainstorming task may benefit from more variation than a fixed-format data classification task.

Keep straightLower temperature does not make an answer more factual.

In practiceUsage limits and costs are often measured in input and output tokens rather than pages or words.

In practiceWeb search, a calculator, a calendar, file access, and a code runner can all be tools.

Keep straightThe model decides what to say or do. A tool performs a specific action in another system.

In practiceA language model learns statistical patterns from training data, but it does not carry a neat, searchable copy of every source.