Your AI clinical note and AI letter templates are only as good as the instructions inside them. The square bracket instructions in your template are what tell PracSuite AI what to write, how much detail to include, and how to format it. Vague instructions produce vague output. Specific instructions produce output that actually sounds like you.
This applies equally to AI Clinical Note Templates and AI Letter Templates, since both use the same square bracket instruction and merge field system.
Why specificity matters
Compare these two instructions:
[Summarise what was said]
[Summarise the patient's reported pain level, location, and any change since the last session, in one short paragraph]
Both are valid square bracket instructions. Only one gives PracSuite AI enough to work with. The first can produce almost anything: a single vague sentence, a long paragraph, or content that misses what you actually wanted recorded. The second tells the AI exactly what to look for, how much of it to include, and how to present it.
If your templates are producing generic or inconsistent output, the instruction text is almost always the first place to look.
Start with the actual problem
Before changing anything, get specific about what's actually wrong. "It's not very good" isn't something you, or an AI tool, can act on, and it usually means one of several quite different things:
The output is too generic or vague
It's missing details you know were in the recording
It's using the wrong tone for the section or the audience
It's too long, or too short
It's structured in the wrong order, or mixing up sections
Each of those points to a different fix. "The Plan section is too short, it doesn't mention treatment frequency even though I said it in the recording" tells you exactly which instruction to rewrite and what to add. "The Plan section isn't good" doesn't. Naming the problem precisely is often most of the work, the fix tends to follow once you know exactly what's wrong, whether you're fixing it yourself or asking an AI tool to help.
Self-help checklist
Once you know what's actually wrong, most templates can be improved by working through these on your own:
Say what to include, not just the topic. "[Summarise the exam findings]" leaves the AI guessing. "[Summarise objective exam findings, including range of motion, strength grading, and any positive special tests, as short bullet points]" doesn't.
Say what to leave out, where it matters. If you don't want the AI speculating on a diagnosis or including information beyond what's in the transcript, say so directly in the instruction.
Plan for missing information. Add "if available" or "if applicable" to instructions covering things that won't come up in every consult, so those sections stay blank instead of being padded out.
Structure the template the way you think. If you work in SOAP, DAP, or another standard structure, lay your headings out in that order. A logical structure makes it easier to write clear instructions section by section.
Use merge fields for anything already on file. Patient name, date of birth, referring practitioner, and similar details should be inserted with merge fields like <<FirstName>>, not requested through a square bracket instruction. If you find yourself writing "[insert patient's date of birth]", check whether a merge field already covers it.
Test before you rely on it. Run a real or realistic transcript through the template's test option before using it in a live consult, so you can see exactly what it produces.
A less obvious technique: context-only instructions
Not every square bracket instruction needs to produce a visible block of text in the finished note or letter. All of the instructions in a template are considered together when your document is generated, which means you can write an instruction purely to give the AI background information, information that shapes how it handles the other instructions in the template, without producing any visible output of its own.
For example:
[Context only, do not output any text for this instruction: this is a follow-up appointment, not an initial consultation. Write the rest of the note accordingly.]
This is useful whenever there's something true about every note or letter this template produces that the AI should factor in, but that isn't itself something you want written out as a sentence somewhere in the document. Without knowing this technique exists, it's easy to assume every square bracket has to generate a paragraph, and end up either leaving useful context out entirely or awkwardly folding it into a section where it doesn't belong.
Working with Claude or ChatGPT to refine a template
If you've worked through the checklist above and your template still isn't producing what you want, a general AI tool like Claude or ChatGPT can help rewrite the instructions for you. The trick that makes this work well is giving it something concrete to aim for, not just a description of what you want, but an actual example of what "good" looks like.
Give it four things:
The specific problem you're having, using the thinking from "Start with the actual problem" above. "The Assessment section is too generic and doesn't reference the outcome measures I dictate" gives the AI something to actually target.
Your profession. This shapes the terminology, tone, and clinical focus the rewritten instructions should aim for.
Your current template, exactly as it is in PracSuite, square brackets and all.
A complete, de-identified example of a note or letter you've written that reflects the level of detail and style you want. Remove the patient's name, date of birth, and any other identifying details before pasting it in.
Items 1 and 4 are the ones that make the biggest difference. A vague problem statement or a description of what you want in the abstract ("make it more detailed") is far weaker guidance than a named problem and a real example the AI can compare your template's output against directly.
Sample prompt
Copy this into Claude or ChatGPT, filling in the four bracketed sections:
I use PracSuite, a clinical practice management platform. It generates clinical notes and letters using templates written in plain text, where AI instructions sit inside square brackets, like [summarise the patient's presenting complaint], and known patient details are inserted using merge fields like <<FirstName>> and <<DateOfBirth>>, which should be left exactly as they are.
I'm a [YOUR PROFESSION].
The specific problem I'm having is: [DESCRIBE THE PROBLEM, e.g. "the Assessment section comes out too generic and doesn't reference the outcome measures I mention in my dictation"]
Here's my current template:
[PASTE YOUR CURRENT TEMPLATE]
And here's a de-identified example of a note/letter I've written that reflects the level of detail and style I want the AI to produce (I've removed all identifying information):
[PASTE DE-IDENTIFIED EXAMPLE]
Please review my template against the example, with the specific problem above in mind, and rewrite the square bracket instructions so they're specific enough to reliably produce output like the example rather than something generic. For each instruction, make sure it specifies what to include, how much detail to give, and the format (paragraph or bullet points). If there's background information that should shape multiple sections without appearing as its own visible text, you can add a context-only instruction for that, one that explicitly says not to output any text. Keep everything outside the square brackets, and all merge fields, exactly as they are. Give me the full revised template back, ready to paste into PracSuite.
Once you have the result, paste it back into your template in PracSuite and test it with a real transcript before relying on it.
Where to go next
For the full walkthrough of setting up templates from scratch, including built-in template options and formatting rules, see: