Muistiinpanojen poiminnan opas

Note extraction turns a session note you have written into bullets, tags and a suggested next action. This page explains what it does, what it deliberately does not do, and what to check before you save. Please read it once before relying on the feature.

What it does

You write a session note and press Extract. It reads what you wrote and returns:

• Bullets — 3 to 5 short factual points drawn from your note.
• Tags — 2 to 3 topic labels.
• A next action — one concrete follow-up, if your note implies one.

On some plans it also suggests task drafts — to-dos it spotted in the note. Nothing is saved until you say so, and you can edit or discard any of it.

What it does not do

This matters more than the feature list.

• It does not diagnose anything. It does not assess, score or predict anything about your client. If your note says a client mentioned feeling low, the extraction should say the client mentioned feeling low — not that they have a condition. Output that reads like a diagnosis is a bug; please report it.
• It does not give professional advice. Nothing it returns is medical, dietary, psychological, legal or financial advice. It is a drafting aid.
• It reads only the note you are extracting. Not your other notes, not the client's history, not their file.
• It never sends anything to your client.
• It does not add facts. It is only meant to restate what you wrote. When it invents a detail, that is the failure mode we watch most closely — because it is the one most likely to slip past a quick read.

Your part: check it before you save

The extraction is shown for review and marked as AI-generated. Read it before saving.

The specific thing to check is numbers and names. Fluent but wrong text is much easier to skim past than obviously broken text, and the errors that matter look like this:

• a weight, dosage, rep count or price that is not in your note
• a date that has shifted
• a name that is not your client's

If your note said 78 kg and the bullet says 87 kg, that is the kind of mistake that survives a glance and ends up in a record. Skimming the specifics against what you wrote takes a few seconds, and it is the whole point of the review step.

Then edit anything that is off, delete anything wrong, and save what is left. Editing is normal. It is not a sign the feature failed.

Where your note goes

When you press Extract, the text of that note is sent to an external AI provider for processing, and the result comes back to you.

Before it is sent, we automatically remove phone numbers, email addresses, card numbers, national ID numbers and dates of birth.

We do not remove names, and we do not remove health or personal content — because those are usually the substance of what you asked us to summarise, and stripping them would leave nothing to work with.

So please assume that whatever you type into a note is what gets sent. If there is something you would not want leaving your systems, do not put it in a note you are going to extract. Which providers we use, where they are, and the safeguards that apply are listed in our privacy policy and on our sub-processors page.

We do not use your notes to train AI models.

If you work in a health-adjacent field

Therapists, counsellors, dieticians, nutritionists and trainers: your notes routinely contain health information about another person, which most privacy law treats as a special category needing a stronger legal basis than ordinary data. Using this feature means that content is processed by an external provider. Two things follow.

1. Make sure your own consent covers it. Your intake and consent paperwork is yours — we do not set it. If it does not mention that you use software with AI features to help write up sessions, it probably should.
2. Check your professional body's position. Some have specific rules about AI tools and clinical records. ClientFlow is a general-purpose business tool, not a clinical system, and we make no claim that it meets any clinical standard.

If either gives you pause, the honest answer is to write the note without extracting it. The feature is optional and nothing else in ClientFlow depends on it.

Limits and quirks

• Minimum 40 characters. Shorter notes are rejected — there is not enough to work with.
• Maximum 8,000 characters. Longer notes are cut off at that point and you will see a truncation warning. Only the first part was read.
• Occasionally the format comes back wrong. You will get bullets only, with an option to retry. Retrying usually works.
• Availability depends on a third party. It can be slow or briefly unavailable.
• It works best on notes written in prose. Very terse shorthand still works, but expect to edit more.

If something goes wrong

Please report it — especially any of these:

• output that reads like a diagnosis, a prescription or a treatment recommendation
• a detail in the output that was never in your note
• anything that looks like another client's information
• an extraction that saved without you confirming it

Contact us at the address in our privacy policy. The first three are treated as incidents rather than feature requests, and you will get an acknowledgement within five working days.

The short version

It rewrites your own notes into bullets. It does not judge your clients, it does not advise you, and it does not act on its own. Read the numbers and names before you save. If a note contains something that should not leave your practice, do not extract it.