Name the action in plain language
Start by saying exactly what the assistant is about to do.
"Help with reminders" is too broad. "Turn on text reminders for future appointments" is specific. "Take care of this invoice" is vague. "Email this finished invoice to this client now" tells you what the click will cause.
A useful action description should answer three questions:
- What will change?
- Which client, appointment, invoice, or setting will change?
- Will it happen now, or only prepare something for later?
If you cannot answer those questions from the confirmation in front of you, do not approve it yet. Open the relevant record or setting and check there first.
This is especially important when the assistant's original message sounds conversational. Friendly language can make a suggestion feel lower-risk than the action behind it. Read the action as if it were a direct instruction from a team member.
Separate a draft from a delivery
Drafting and sending are two different jobs.
An assistant may produce a polished message in seconds, but that does not mean the message is ready to leave your practice. Before anything is sent, check:
- The recipient
- The reason for contacting them
- The tone
- The next step you are asking them to take
- Any personal or appointment details included in the message
Read the draft from the client's point of view. Does it sound like your practice? Does it make an assumption about why they have not returned? Does it reveal more than the message needs to carry? Is there a clear way for the client to decline or ask a question?
A strong workflow keeps the draft editable and makes sending a separate approval. If one button creates and sends the message at the same time, slow down and make sure the review happens before you press it.
Check the affected record
The assistant may have understood your request correctly and still be working from the wrong context.
Two clients can have similar names. An appointment may have been moved since the suggestion was created. A draft invoice may already have been edited. A reminder setting may have changed in another browser tab.
Open the record that will be affected and look for one or two identifying details that are appropriate for the task. For an invoice, that might be its status, amount, and client. For an appointment, it might be the date, time, and practitioner. For a practice setting, it might be the current value and the part of the workflow it controls.
You do not need to reread the entire client history. You need enough context to know that this is the right record and that the proposed action still makes sense.
Ask what happens immediately
Some changes are quiet. Others start a chain of events.
Turning on a reminder setting may affect future messages. Sending an invoice creates a client email and a payment request. Publishing a booking change can alter what new clients see. Moving an appointment can affect availability and notifications.
Before approving an action, identify the immediate result and the next likely result.
Try finishing this sentence:
"When I approve this, the system will ___, and then ___."
If the second blank is unclear, that is your cue to investigate. Look for a confirmation message, help text, or the related settings page. Do not assume that a change is retroactive, that a message will remain a draft, or that another team member will be notified.
Prefer one clear action over a batch
AI can make a list of tasks feel easy to clear all at once. That is useful for low-risk organization. It is less useful when every item touches a different client or financial record.
Start with one action when:
- A message will go to a client
- Money is being requested, refunded, adjusted, or recorded
- Availability or booking rules will change
- A client record will be created or updated
- You have not used this action before
Review the result before repeating it. Did the setting change where you expected? Did the message arrive with the right subject and sender? Did the invoice remain connected to the right appointment?
One successful result gives you evidence. A successful-looking button gives you only reassurance.
Look for a safe stopping point
A good assistant should make it easy to stop between suggestion and action.
That stopping point may be a confirmation dialog, an editable draft, a link to the affected record, or a review screen. Use it. It is there to let you bring in the context the assistant does not have, including a recent phone call, a private agreement, or a change another staff member made.
Be more cautious when the action has no clear review step. Ask the assistant to explain the proposed change without doing it, or navigate to the normal practice screen and complete the task there.
Also look for the recovery path. Can you turn the setting back off? Can you edit an unsent draft? Can you correct the record? A reversible action is not risk-free, but knowing how to recover helps you decide how much review it needs.
Keep the final decision human
An assistant can summarize what it sees. It cannot know every promise you made in the room, every nuance of a client relationship, or every reason your practice handles an exception differently.
Keep human judgment closest to actions involving:
- Clinical decisions
- Sensitive client communication
- Money
- Consent
- Access to records
- Changes that affect many appointments or clients
The goal is not to distrust every suggestion. It is to use the assistant for speed without handing over the part of the work that depends on context and responsibility.
A practical rule for your team is simple: let AI notice and draft quickly, then let a person approve deliberately.
If you use Stillpoint, Clio, the AI practice assistant, can answer product questions and help with practice setup. When a Clio suggestion would change a setting or send a finished draft invoice, Stillpoint shows a confirmation first. Client check-ins are generated as editable drafts and are sent only after you approve the final text.
