AI Charting for Dietitians: Drafting ADIME Notes With a Review Step

Dictate or record your findings with the client's consent, ask an AI tool covered by the right data agreement to draft the Assessment, Diagnosis, Intervention and Monitoring sections, then check every line against your own measurements before it enters the record. The AI drafts; you diagnose, decide and sign.

The ADIME format suits AI well because it is structured. A tool that knows the four headings, and that a diagnosis is written as a problem linked to a cause and evidenced by signs, can turn a rambling dictation into something tidy. The risk is that it fills gaps with plausible content, and a nutrition note with one invented intake figure is a record with a false fact in it.

Follow me on Instagram@sagnikteaches

What ADIME asks for, and where AI can and cannot help

ADIME comes from the Nutrition Care Process: Assessment, Diagnosis, Intervention, and Monitoring and Evaluation. Each section has a job and each has a different AI risk.

Connect on LinkedInSagnik Bhattacharya

Notice the pattern. AI is good at wording and structure and weak at facts it was not given. That tells you how to use it: supply the facts, let it do the form, and review the facts again.

Subscribe on YouTube@sagnikteaches

A scribe or a chatbot: two routes to a draft

An AI scribe is a tool that listens to the consultation and drafts a note into a template. A chatbot is a general assistant you type or paste into. They suit different practices.

• Scribe. Best for live consultations where you want to keep eye contact and not type. It needs recording consent, runs through a vendor that usually has health-data terms, and may write directly into your records system. Check that it supports an ADIME or similar template, and that you can edit the template.
• Chatbot with your dictation. Best when you record your own summary after the session. No client conversation is recorded, which makes consent simpler, but your summary still contains client details. It requires a plan with appropriate terms, and the discipline to keep every identifier out unless the agreement covers them.
• Your practice platform's built-in feature. Some practice-management platforms for dietitians now include note drafting. If yours does, it may be the simplest route, since the agreement already exists. The tutorial on dietitian practice platforms and their AI features compares them.

For a solo practitioner with a handful of clients a day, dictated summaries into a suitable chatbot are often the cheapest and least disruptive start. For a clinic with several dietitians, a scribe connected to the records system usually saves more time. Either way, the review step below is the same.

Choosing a tool under HIPAA or UK GDPR

This is where most dietitians get stuck, so keep it practical. The tool matters less than the paperwork behind it.

If health-privacy rules such as HIPAA apply to you, you need the vendor to sign a business associate agreement (BAA) before client health information goes in. Not every plan or product is covered. As an example of how specific it can be, Anthropic's help page says its BAA covers its first-party API and Claude Enterprise, and that Enterprise features are covered only after the organisation's Primary Owner turns on HIPAA compliance in the settings and accepts the agreement. Even then it excludes features such as Cowork in the cloud and beta products such as Claude Docs and Claude Slides, and the Team plan is not among the services it names. Other vendors have their own lists. Read the vendor's page, not a third-party summary, and get the agreement in writing before the first real note.

If UK GDPR applies to you, the questions are different: a lawful basis for processing health data, a data processing agreement with the vendor, where data is processed, retention and deletion, and whether the tool uses your data for training. You will probably need a data protection impact assessment, and the regulator's guidance on AI and health data is the place to check. The tutorial on implied or explicit consent for AI scribes covers consent in more depth.

In either case, ask the same five questions of any vendor.

1. Will you sign the agreement my rules require, for this exact plan?
2. Is client content used to train your models?
3. How long is audio, text and output retained, and can I delete it?
4. Where is it processed and stored?
5. Which features, apps or integrations are outside the agreement?

If the answer to the first is no, the tool is not for identifiable client notes. It might still be useful with generic content, like turning a handout into plain language, but that is a different job. For a fuller checklist, the tutorial on whether an AI scribe is HIPAA compliant sets out BAA and consent steps before go-live.

Consent: what to tell clients and when

Tell clients you use an AI tool, what it does, what it does not do, and that you review every note. If a session is recorded or transcribed, get explicit permission before you start, and record that permission. Recording laws vary, and some places need every party's consent, so check the rule where you and the client are.

Illustrative wording for an intake form or a verbal script:

To save time on notes, I use an AI tool that can turn my summary of our session into a draft record. I always read and correct the draft, and I never let it make decisions about your care. Your information is handled under an agreement that covers health data. You can say no at any time and I will write notes the usual way. Are you comfortable with that?

Make opting out easy and free of consequences. Note the answer in the record. If a client declines, they should receive the same care and the same quality of notes.

A worked example: one follow-up visit, from dictation to signed note

Illustrative. Say a dietitian sees an adult client for a follow-up about type 2 diabetes and weight. After the visit, she dictates this summary into a chatbot covered by her practice's agreement, using an ID number rather than a name:

Client 0417, follow-up four. Weight 94.2 kg, was 96.0 four weeks ago. HbA1c 7.4 percent from last week, was 8.1. Food diary shows breakfast skipped three days a week, sugary drinks down from five to two a day. Client says evenings are the problem, takeaway after late shifts. Goal this month: breakfast five days, no sugary drinks, one batch-cooked meal on days off. Gave handout on portion sizes. Review in four weeks, repeat weight, ask for diary again.

She pastes this with the following prompt:

Draft an ADIME note from my dictation. Use only the information given. Do not add values, advice, diagnoses or targets that I did not state. For Diagnosis, offer two PES statement options and mark each "needs clinician decision". If anything is missing for a section, write "Not stated" instead of filling it in. At the end, list every number you used so I can check it.

A good draft looks like this:

A: Weight 94.2 kg (previous 96.0 kg, change -1.8 kg over 4 weeks). HbA1c 7.4% (previous 8.1%). Food diary: breakfast skipped 3 days/week; sugary drinks reduced from 5 to 2 per day. Evening eating reported as difficult; takeaway after late shifts.
D: Option 1: Inadequate food and beverage intake pattern related to late-shift schedule as evidenced by skipped breakfast 3 days/week and evening takeaway. Option 2: Food- and nutrition-related knowledge deficit related to ... (Not stated). Both need clinician decision.
I: Goals for the month: breakfast 5 days/week; no sugary drinks; one batch-cooked meal on days off. Portion-size handout given.
M: Review in 4 weeks. Repeat weight; request food diary.
Numbers used: 94.2; 96.0; 1.8; 7.4; 8.1; 3; 5; 2.

Now do the review. Check each number against your own record; the draft is right on all of them, but check anyway because that is the habit. The tool has done something sensible by leaving the second diagnosis incomplete: aetiology not stated. A weaker draft would have written "related to lack of education", which you never assessed. Choose option 1, or write your own, and delete option 2. Before you keep option 1, check its first words: "inadequate food and beverage intake pattern" is the tool's own phrase, not a term from the standardised Nutrition Care Process terminology. Swap in the exact diagnosis term from your terminology reference, because a home-made label makes the note harder to audit and harder for the next dietitian to search. The change in weight is arithmetic the tool did itself: 96.0 minus 94.2 is 1.8, so it passes. The Intervention section contains nothing you did not say, which is exactly what you asked for.

The review step, line by line

Treat review as a procedure, not a glance. A checklist you can run in two or three minutes catches most errors.

1. Numbers. Compare every value and unit with your source: weight, height, lab results, dates, doses of supplements.
2. Attribution. Everything "reported" by the client must be marked as reported. Everything "observed" must be something you saw.
3. Diagnosis. Each PES statement has a problem you identified, an aetiology you can support and signs you measured or recorded. Delete anything else.
4. Intervention. Only what you did or advised. If the draft adds "encouraged regular exercise" and you did not say it, remove it.
5. Monitoring. Targets and review dates match what you agreed with the client.
6. Safety flags. Allergies, interacting medicines, eating disorder history and pregnancy are easy to lose. Check they are still there if relevant.
7. Tone and wording. Remove blame or judgement that crept in ("failed to", "non-compliant").
8. Sign-off. Only when you have done steps 1 to 7, sign. If you have not read it, do not sign it.

A realistic target is two to four minutes of review for a follow-up note. If you find yourself skimming, slow down: the most dangerous errors look like normal text.

Three more examples of drafts to check

A before and after: from shorthand to a clear Assessment

Before, in your own notes: "Hx IBS, low FODMAP 6 wks, bloating better, wt stable, Hb 128, ferritin low-ish." After, a good draft: "History of irritable bowel syndrome. Low-FODMAP diet followed for 6 weeks; bloating improved per client report. Weight stable. Haemoglobin 128 g/L; ferritin recorded as low (value not stated)." The tool kept "low-ish" honest by flagging the missing value. A poor draft would have invented "ferritin 12 micrograms per litre". Your line-by-line check catches it, and you add the real value.

A mistake that looked fine: the misplaced goal

A dietitian dictated two goals, one for the client and one for a carer who helps with shopping. The tool merged them into one list under the client's name, which read as though the client had agreed to all of them. Nothing in the wording was false, yet the record was misleading. The fix is to name who owns each goal when you dictate: "client goal" and "carer goal".

A quick sum: is it worth it?

Say you see eight clients a day and each note takes 12 minutes by hand, about 96 minutes. With a drafted note and a three-minute review plus two minutes of dictation, you spend about 40 minutes. The saving is nearly an hour a day. If the tool costs, say, $25 to $60 a month (team chat plans list at $20 to $25 a seat; scribe prices vary, so check the vendor's own page), the time pays for it quickly, but only if the review is real. A saving that comes from skipping the review is not a saving; it is a risk. Check against your own diary for a fortnight before you commit.

A first assessment is a different job from a follow-up

Follow-ups are the safest place to start because most facts already sit in the record. A first assessment is longer, richer and riskier, since the tool sees more history at once and has more room to blur it. If you use AI there, split the job in two. Dictate the Assessment facts first and ask only for a structured list, with no diagnosis. Review it. Then, in a second step, ask for PES options based on the list you have just approved.

Illustrative: a first visit with an adult who follows a vegetarian diet and has low iron. Your dictation gives age band, diet pattern, symptoms (tiredness, no other findings noted), a ferritin result, a food diary summary and a goal of more energy. The first prompt returns the headed Assessment. You correct the line where the tool wrote "iron deficiency anaemia" when you only recorded low ferritin and a normal haemoglobin. That one word changes the clinical picture, so the staged approach pays for itself. The second prompt, with the corrected Assessment, then drafts two PES options. This sequence also keeps the model from diagnosing before the data is right.

A reusable prompt you can keep and refine

Once you have a prompt that behaves, save it and treat it like a template you version. A good one carries the rules about not inventing, the "Not stated" convention, the numbers list and your preferred wording. Add a line for your own style, such as "use plain language a referrer would understand" or "use metric units". Keep it in a text file, note the date you last changed it, and change one thing at a time so you can tell which edit helped.

You are helping a registered dietitian draft an ADIME note. Rules:
1. Use only what is in my dictation. Never add values, advice, diagnoses, allergies or targets.
2. If a section lacks information, write "Not stated".
3. For Diagnosis, offer up to two PES statements and label each "needs clinician decision". Do not write an aetiology I did not state.
4. Copy numbers and units exactly. Finish with a list of every number used.
5. Use neutral, non-judgemental wording. Say "reported" for what the client told me.
6. Do not give general nutrition advice.

Run the same dictation through it twice. If the two drafts differ on a fact, not just on wording, that is a sign your dictation was ambiguous; tighten it. Variation in wording is normal and harmless. Variation in facts means the source was unclear.

Telehealth and group sessions

If you see clients by video, the platform may offer its own recording or transcription, separate from any scribe you use. Be clear which tool records what, and do not run two at once without the client knowing. For group sessions, a transcript contains several people, and a note drafted from it risks mixing one participant's information into another's record. Draft notes per client from your own dictation after the session rather than from a group transcript.

The same caution applies when a carer or parent attends. Say in the dictation who said what, and make sure the record separates a client's own report from a carer's report. A tool will happily merge the two into one voice.

Keeping a short audit trail of your own

If a record is ever questioned, you want to show that a clinician wrote and approved it. Add a line to your notes procedure: "Drafted with AI assistance, reviewed and edited by [name]". Keep it in your practice policy and, if your records system allows, in the note footer. Keep the prompt you used and the date of your last test, and every few months run a handful of old, de-identified examples through to check the tool has not changed its behaviour. Providers update their models regularly, not always with notice, so a prompt that behaved in spring may behave differently in autumn. This takes ten minutes and gives you something concrete to point to.

Where AI charting goes wrong for dietitians

The confident diagnosis. The tool offers "excessive carbohydrate intake related to emotional eating" because the transcript mentioned a stressful week. The client never said emotional eating. Your review rule: aetiology needs your evidence, not the tool's inference.

Unit and rounding drift. mmol/L becomes mg/dL, or 94.2 becomes 94. Ask for numbers to be copied exactly and listed at the end, as in the prompt above.

Template creep. A tool trained on medical notes adds "Plan: follow up with the doctor" because that is common elsewhere. Ask for "Not stated" rather than a guess, and delete extras.

Sensitive content in the wrong place. Eating disorder history, pregnancy and religious or cultural food rules are sensitive. Decide in advance whether they go through the tool at all, and do not include them if the tool's agreement is uncertain.

The copy that outlives the session. Chat histories can hold client details for a long time. Know where your chatbot keeps conversations and how to delete them, and delete after you have pasted the final note into your record.

A two-week trial that keeps you honest

Start with a small, low-risk slice: follow-up visits, not complex first assessments. For ten visits, write the note yourself first, then compare it with the AI draft. Record the time each way and note every difference that mattered. If the draft misses or invents something clinically relevant more than once in ten notes, change the prompt or the tool, or stop. Once you are confident, move to drafts first and review second, but keep a monthly spot-check of five notes against what you remember of the session.

For ideas beyond charting, see AI tools for dietitians by job. If you would like help choosing a tool and setting up the review checkpoint, that is something my AI implementation consultation can cover.

Questions dietitians ask about AI charting

Can AI write the PES statement for me?

It can draft one from your findings, but the diagnosis is your clinical judgement. Give it the assessment data, ask for two options, and choose or rewrite. Never accept a PES statement whose aetiology or signs you did not observe or record.

Is an AI scribe or a chatbot better for ADIME notes?

A scribe suits live sessions because it captures the conversation and drafts into a template. A chatbot suits dictated summaries after the visit. The deciding factor is not quality but whether the vendor will sign the data agreement your jurisdiction needs.

Do I need client consent before using AI on notes?

Tell clients, and get explicit consent whenever a session is recorded. Even for dictated notes with no client identifiers, say that you use an AI tool and give them a way to opt out. Ask your professional body for its current guidance.

Can I use a free chatbot if I remove the client's name?

Removing a name is not enough. Age, conditions, medications and rare details can still identify someone. Use a tool covered by the agreement your rules require, or keep the input genuinely de-identified and general.

Further reads

• ChatGPT or Claude for Freelance Dietitians? Pick by Job and Data — Pick between ChatGPT and Claude for freelance dietitian work.
• AI Scribe for Counsellors: A Safe Trial Plan With Consent and Opt-Out — A safe trial plan with consent and opt-out, transferable to dietetics.
• Can Dietitians Use AI for Admin and Content? What Stays Human — What stays human in a dietitian's AI use.
• AI Training for Dietitians: A Practice Owner's Team Plan — A team plan if your practice has several dietitians.
• AI Scribe Consent Form: US Wording and State Recording Laws — Consent form wording and recording-law checks for scribes.
• AI Scribe for Psychologists: Notes, Reports and Consent Checks — Another clinical note workflow with consent checks.
• AI Tools for Nutritionist Practices: Client Check-Ins and Handouts — How a solo nutritionist can use AI for check-in messages, diary summaries and handouts, with review rules so no unsafe advice reaches a client.
• AI for Naturopaths: Intake Summaries and Follow-Up Emails — A naturopath's guide to AI intake summaries and follow-up emails: what to redact, what to keep out, and prompts to copy.
• Nutritionist AI Prompt: Write Consent and Cancellation Wording — A fill-in-the-blanks prompt that drafts consent and cancellation terms for a nutrition practice, plus the checks before a solicitor sees it.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Nutrition Care Process summaries (Academy of Nutrition and Dietetics framework); Anthropic's business associate agreement help page; Claude and ChatGPT team plan pricing pages, checked 4 November 2026.