An AI scribe can look impressive in a demonstration. It listens to a conversation, produces a polished note, and appears to remove a large part of the physician's documentation burden.
Then the clinic tries to use it every day.
The note may be too long. Important details may be missing. The physician may spend more time correcting the output than expected. Staff may not know what to do when the recording fails. The finished note may still need to be copied into the EMR, reformatted, reviewed, and signed.
The problem is not always the tool. It is often the absence of a clear operating process around the tool.
An AI scribe is not only a documentation feature. It changes how a clinic captures a conversation, informs the patient, reviews the note, handles sensitive information, manages exceptions, and decides whether the result is ready for the medical record.
That is why a successful demonstration is not the same as a successful implementation.
This guide presents a practical seven-step method for testing an AI scribe in a Quebec clinic. It is designed to help a clinic decide whether to keep, adjust, expand, or stop a pilot based on evidence rather than enthusiasm.
Quebec readiness check, reviewed August 15, 2026
Before using an AI scribe, verify its current status in the Government of Quebec's AI transcription program and certification table. The table was last updated July 9, 2026 when this guide was prepared.
The Quebec government page states that an AI transcription tool requires Trousse globale de vérification certification from the supplier and an evaluation des facteurs relatifs à la vie privée, or EFVP, for the acquiring health organization. An EFVP completed for Santé Québec does not automatically cover another organization.
For physicians, the Collège des médecins du Québec states that explicit patient consent is required before using an AI scribe and that a waiting-room sign alone is insufficient. The physician remains responsible for reviewing, correcting, completing, and signing the note.
This is practical operational information, not legal, privacy, medical, security, or procurement advice. Requirements and product status can change. Confirm the current official guidance for your clinic before use.
The Teché controlled pilot framework
- Check the operating boundary: Confirm certification, privacy assessment, consent, data handling, and physician responsibility.
- Measure the baseline: Record how documentation works before the pilot.
- Limit the pilot: Start with a small group, a defined visit type, and a clear review period.
- Define an acceptable note: Agree on required sections, length, style, and unacceptable errors.
- Map the real workflow: Follow the note from encounter to reviewed and signed chart entry.
- Train and measure: Teach one consistent process and compare it with the baseline.
- Decide: Keep, adjust, expand, or stop based on evidence.
1. Check the operating boundary before the pilot
A clinic should not begin by asking which tool has the most features. It should begin by defining the conditions under which any tool may be tested.
In Quebec, this includes at least four separate questions.
Is the product currently certified for use in Quebec?
Canada Health Infoway prequalification and Quebec certification are not the same thing. Infoway's national program provided eligible primary care clinicians with funded one-year licences for pre-qualified products. Quebec maintains its own dated certification and EFVP status tables.
A clinic should check the current official Quebec table rather than relying on an old article, vendor page, sales presentation, or remembered product list.
Has the clinic addressed its own privacy assessment responsibility?
Section 106 of Quebec's Act respecting health and social services information requires an organization to conduct an EFVP for certain technology projects involving health information. The assessment must be proportionate to the sensitivity, purpose, quantity, distribution, and format of the information involved.
The Commission d'accès à l'information's EFVP guide describes a process that includes identifying the personal information involved, tracing its path from collection to destruction, evaluating risks, and establishing protective measures and an action plan.
This is specific to the organization and its actual environment. A generic checklist or a supplier's assurance is not a substitute for the clinic's own review.
Is the consent process clear?
The CMQ states that a physician must obtain the patient's free and informed consent before the first use of an AI scribe. The discussion should explain why the tool is being used, what it captures, and the relevant benefits, risks, and limits.
At later visits, the physician may remind the patient that consent was previously given and that the scribe is active, giving the patient an opportunity to withdraw consent. The CMQ recommends obtaining new consent when a visit becomes more sensitive or when a major technology change increases privacy or security risk.
If the patient refuses, the scribe should not be used. The clinic therefore needs a normal documentation process that remains available without friction or pressure on the patient.
Does the clinic understand the data path?
Before a pilot, the clinic should know:
- what is recorded;
- what is transmitted;
- where information is processed and stored;
- who can access it;
- whether information is used to train or improve the product;
- how long recordings and transcripts are retained;
- how and when they are destroyed;
- what happens when the service fails.
Canadian hosting alone does not answer all of these questions. The clinic needs to understand the complete information path and the controls that apply to it.
2. Measure the current documentation workflow
Without a baseline, almost any pilot can feel successful or unsuccessful depending on the day.
A clinic needs a simple picture of how documentation works before introducing the scribe. The goal is not to launch a research study. It is to gather enough evidence to make a responsible operational decision.
For a defined visit type, record:
- the time from the end of the encounter to the final signed note;
- the physician's active editing time;
- whether notes are usually completed the same day;
- how often substantial corrections are required;
- common reasons a note remains unfinished;
- manual transfer or formatting steps;
- recurring exceptions and rework.
The baseline should also capture quality expectations. A shorter note is not necessarily better. A faster note is not useful if the physician must correct missing, invented, or misleading information later.
Canada Health Infoway's national evaluation reports promising program-level findings, including reduced administrative burden for many participating clinicians. Those findings explain why AI scribes deserve serious evaluation. They do not guarantee that a particular clinic, specialty, workflow, or product will achieve the same result.
The clinic's own baseline is what makes its pilot interpretable.
3. Limit the pilot so that you can learn from it
One of the fastest ways to lose control of a pilot is to test too many situations at once.
A useful first pilot might include:
- one or two willing physicians;
- one repeatable visit type;
- a defined start and end date;
- a named operational owner;
- a small set of success and stop criteria;
- a scheduled review before expansion.
This is not about delaying adoption. It is about making the result understandable.
If a clinic tests several specialties, multiple note formats, complex consultations, virtual visits, walk-ins, and every exception at the same time, it becomes difficult to know why the tool worked in one situation and failed in another.
A narrow pilot makes it easier to improve the configuration, training, and workflow without exposing the whole organization to an untested process.
4. Define what an acceptable note looks like
An AI-generated note can be polished and still be unsuitable.
Before the pilot begins, the physician should define what the note needs to contain and what would make it unacceptable. This may include:
- required sections;
- preferred length and level of detail;
- specialty-specific terminology;
- relevant negative findings;
- information that must come from the physical examination;
- information the AI cannot observe, such as non-verbal behaviour;
- wording that requires particular caution;
- unacceptable omissions, additions, or assumptions.
The clinic should separate stylistic preferences from clinically important problems. A heading the physician dislikes may be a configuration issue. An invented medication, missing assessment, or inaccurate clinical statement is a different category of concern.
The CMQ advises physicians to pay particular attention to hallucinations, misinterpretations, bias, and missing information. It also states that the note is not official until the physician has reviewed and signed it.
The scribe may prepare a draft. It does not transfer documentation responsibility away from the physician.
5. Map the note into the real EMR workflow
The relevant question is not simply, "How fast does the AI produce a note?"
The better question is, "How does the encounter become a reviewed and signed entry in the medical record?"
Map the actual path:
- The physician explains the scribe and confirms consent.
- The tool is started for the appropriate visit.
- The conversation is captured and processed.
- A draft note is generated.
- The physician reviews and corrects it.
- Required information from the examination or clinical judgment is added.
- The note is transferred or imported into the EMR.
- The physician performs a final review and signs it.
- Recordings and transcripts are handled according to the approved process.
- Any failure or exception is documented and resolved.
Every manual step matters. If the physician saves ten minutes in drafting but loses the same ten minutes copying, reformatting, correcting, or searching for the note, the clinic has moved work rather than removed it.
Integration claims also require care. A vendor may use the word integration for anything from a browser extension to a structured data exchange. The clinic should test the exact workflow in its own EMR environment instead of assuming that a marketing label describes the finished process.
6. Assign responsibility before something goes wrong
An implementation becomes fragile when responsibility is implied rather than assigned.
Before the first pilot visit, the clinic should be able to answer:
- Who confirms that the tool is approved for the pilot?
- Who maintains the consent process and patient explanation?
- Who starts and stops the tool?
- Who reviews, corrects, completes, and signs the note?
- What happens if the recording fails?
- What happens if the note is incomplete or contains a serious error?
- Who monitors recurring exceptions?
- Who contacts the supplier when an incident occurs?
- Who decides whether the pilot continues?
The physician remains responsible for the clinical note. Other responsibilities, such as staff support, privacy review, technical escalation, and measurement, should also have named owners.
This prevents a common failure pattern: everyone assumes someone else is monitoring the problem until the pilot has already expanded.
7. Train the clinic on the process, not only the product
Product training usually explains where to click. Operational training explains when the tool should be used, how the patient is informed, what a good note looks like, how the physician reviews it, and what happens when the normal process fails.
A practical training session should cover:
- the approved visit types and exclusions;
- the patient explanation and consent process;
- how to start and stop the tool correctly;
- how to review for hallucinations, omissions, and misinterpretations;
- how to add information that was not spoken aloud;
- the path into the EMR;
- handling a refusal or failed recording;
- the escalation path for privacy, technical, or documentation concerns;
- uses that are outside the pilot boundary.
Training examples should be fictional and contain no patient information.
Consistency matters because a pilot cannot be evaluated if every participant uses a different process. The goal is not to remove physician judgment. It is to make the surrounding operating steps clear enough that professional judgment can be applied reliably.
8. Measure what changed
At the end of the pilot, compare the results with the baseline.
| Measure | Baseline | Pilot | How to interpret it |
|---|---|---|---|
| Time to final signed note | Clinic records | Clinic records | Did documentation finish sooner? |
| Active editing time | Clinic records | Clinic records | Consider speed and note quality together |
| Substantial corrections | Clinic defines | Clinic records | Look for repeated error patterns |
| Workflow exceptions | Clinic records | Clinic records | Identify unsafe or costly failure modes |
| Same-day completion | Clinic records | Clinic records | Did the tool reduce unfinished work? |
| Consistent use | Not applicable | Clinic records | Adoption alone does not prove value |
| Clinician confidence | Clinic baseline | Clinic follow-up | Compare perception with operational evidence |
The clinic should define its own measures and thresholds. A high adoption rate does not prove that the workflow is safe or valuable. A fast note does not compensate for substantial corrections. A physician's positive impression matters, but it should be considered alongside the work required to produce the final signed note.
A fictional example
Imagine a clinic testing an AI scribe for routine follow-up visits.
During the first week, physicians like the draft speed, but the notes are longer than expected and include repeated background details. Editing time remains high. Staff also discover that the final note still requires several manual formatting steps before it can be signed in the EMR.
The clinic could conclude that the product failed. A better response is to separate the problems:
- note length and repetition may require a clearer template or configuration;
- inconsistent review may require training;
- manual transfer may be an EMR workflow limitation;
- recurring clinical omissions may represent a more serious suitability problem.
The clinic narrows the pilot, defines a shorter note standard, retrains the physicians, and tests the real EMR path again. If the same high-risk problems continue, stopping remains a valid outcome.
The purpose of the pilot is not to prove that the purchase was correct. It is to learn whether the complete workflow is worth keeping.
9. Decide: keep, adjust, expand, or stop
Every pilot should end with an explicit decision.
Keep
The tool meets the clinic's defined operating, review, and documentation requirements for the tested use. Continue monitoring the workflow and the current Quebec requirements.
Adjust
The tool may be useful, but the configuration, note standard, training, consent workflow, EMR process, or exception handling needs another limited test.
Expand
The evidence supports a controlled extension to another clinician or visit type. Expansion should repeat the same baseline, training, review, and measurement discipline.
Stop
The tool does not fit the clinic's needs, creates unacceptable effort or risk, cannot meet the required operating boundary, or repeatedly produces problems that the clinic cannot safely control.
Stopping is not a failed pilot. A pilot that prevents a poor organization-wide rollout has done useful work.
Common reasons AI scribe pilots disappoint
- The clinic treats a demonstration as proof of implementation.
- There is no baseline, so improvement cannot be measured.
- A polished note is accepted without checking accuracy and usefulness.
- Review and escalation responsibilities remain unclear.
- Manual transfer and formatting erase the apparent time saving.
- Clinicians learn product features but not the clinic's approved process.
- The pilot expands before recurring exceptions are understood.
- National prequalification is confused with current Quebec certification.
- The privacy assessment is treated as a supplier document rather than an organization-specific process.
- The clinic has no simple alternative when a patient refuses or the tool fails.
Start with one controlled workflow
AI scribes can be valuable. Canada Health Infoway's national program results give clinics good reason to explore them. The safest way to do that is not to assume the technology will fix documentation by itself.
Start with one visit type. Define the baseline. Confirm the operating boundary. Set the note standard. Map the path into the EMR. Train the people involved. Measure the finished workflow, not only the speed of the generated draft.
Then decide whether to keep, adjust, expand, or stop.
If your clinic is planning an AI-scribe pilot or deciding whether to expand one, Teché can help map the workflow, clarify roles, define practical measures, and create a prioritized implementation plan.
Request a Clinical Operations Assessment
If the operating process is already defined and your team needs help using it consistently, explore Practical AI and Systems Training.
Sources reviewed
The regulatory and program sources below were reviewed on August 15, 2026. Product status and requirements can change, so readers should consult the current official pages.
- Government of Quebec: AI transcription projects for health professionals
- Collège des médecins du Québec: Scribe et intelligence artificielle
- Collège des médecins du Québec: Health care and artificial intelligence
- LégisQuébec: Act respecting health and social services information, section 106
- Commission d'accès à l'information: EFVP guide
- Canada Health Infoway: AI Scribe Program
- Canada Health Infoway: 2026 AI Scribe Program evaluation
