AI in the Front Office

An AI Use Policy for Small Practices

A practical AI use policy for small medical practices: approved tools, patient information rules, review duties, disclosure, incident steps and records.

AssistCall research teamPublished Reviewed by Faruque Amin, VP of Sales and Marketing7 min read

On this page
  1. Why a small practice needs one now
  2. What the policy should cover
  3. The approved tools register
  4. Patient information rules
  5. Review duties
  6. Disclosure, fairness and access
  7. Incident steps
  8. Training, records and review
  9. Putting the policy in place in two weeks
  10. Example: a fictional three-physician practice
  11. Frequently asked questions
  12. References
AI in the Front Office: An AI Use Policy for Small PracticesArticle coverAI IN THE FRONT OFFICEAn AI Use Policy forSmall PracticesAva Medical IntakeInsightsAI in the Front Office: An AI Use Policy for Small PracticesArticle coverAI IN THE FRONT OFFICEAn AI Use Policy forSmall PracticesAva Medical Intake

Key takeaways

  • Most practices have no AI policy: 56% of medical groups have none and only 20% have one in place[1].
  • A useful policy for a small practice fits on a few pages: approved tools, patient information rules, review duties, disclosure, incidents, training and records.
  • The single most important rule: no patient information goes into any AI tool that is not on the approved list with a business associate agreement[2].
  • Keep the policy and its versions for six years, as HIPAA requires for compliance documentation[3].

Why a small practice needs one now

AI arrived in most practices without a decision. A physician tries a drafting tool, a vendor adds an AI feature to the phone system, a staff member uses a free chatbot to rewrite a patient letter. Each choice seems small. Together they mean patient information may be flowing to places nobody has approved, and AI output may be reaching patients without anyone checking it.

Governance has not kept up. In an MGMA poll from January 2026, only 20% of medical groups said they had AI governance or a formal AI-use policy, 22% were developing one and 56% had none[1].

AI governance in US medical groups, January 2026MGMA Stat poll, 328 responses
AI governance in US medical groups, January 2026Bar chart. No AI governance or policy: 56%; Developing one: 22%; Have AI governance or a formal policy: 20%.No AI governance or policyNo AI governance or policy: 56%56%Developing oneDeveloping one: 22%22%Have AI governance or a formal policyHave AI governance or a formal policy: 20%20%AI governance in US medical groups, January 2026Bar chart. No AI governance or policy: 56%; Developing one: 22%; Have AI governance or a formal policy: 20%.No AI governance or policyNo AI governance or policy: 56%56%Developing oneDeveloping one: 22%22%Have AI governance or a formal policyHave AI governance or a formal policy: 20%20%

Source: Medical Group Management Association (MGMA), 2026[1]

Show the data
ItemValue
No AI governance or policy56%
Developing one22%
Have AI governance or a formal policy20%

The risk from unapproved tools is real. In a 2026 KFF poll, 41% of adults who had used AI for physical or mental health said they had uploaded personal medical information into an AI tool or chatbot[4]. If patients do it with their own information, staff under time pressure may do it with patients'. A short written policy, explained to everyone, is the cheapest control a small practice has.

What the policy should cover

  1. Purpose and scope: which tools count as AI for this policy, and who it applies to, including clinicians, staff, contractors and students.
  2. Approved tools: a register of every AI tool the practice uses, what it may be used for and who owns it.
  3. Patient information rules: what may and may not be entered into any AI tool.
  4. Human review: who checks AI output before it reaches a patient or the medical record.
  5. Patient disclosure: how patients are told when they are interacting with AI.
  6. Fairness and access: testing for patients with limited English or disabilities, and avoiding biased decision support.
  7. Incidents: how errors and privacy problems are reported and handled.
  8. Training: what everyone must complete, and when.
  9. Records and review: where the policy and logs are kept, and how often the policy is updated.

The approved tools register

Fictional entries showing the structure. List every tool your practice actually uses.
Tool (Example)Approved usePatient information allowed?Review before useBAA signedOwner
Phone assistantOffice questions, capture and route requests, after-hours captureYes, limited to contact details and request typeWeekly call sample; every escalationYesPractice manager
Ambient scribeDraft visit notesYesSigning clinician reviews every noteYesMedical director
Portal reply draftingDraft replies to portal messagesYes, within the EHRClinician or nurse approves each replyYes, via EHR vendorMedical director
General chatbotDrafting policies, marketing copy, job adsNo, neverAuthor reviewsNoPractice manager

The register is the heart of the policy. Anything not on it is not approved. Adding a tool requires the owner to confirm the use, the data it touches, the review step and, where patient information is involved, a signed business associate agreement. HIPAA permits a covered entity to let a vendor handle protected health information only with satisfactory written assurance that the vendor will safeguard it[2].

Patient information rules

Sample policy wording: “Staff must not enter patient names, dates of birth, contact details, medical record numbers, images, clinical details or any other protected health information into any AI tool that is not listed in the approved tools register as permitted to receive patient information. When in doubt, do not enter it, and ask the practice manager.”

  • De-identifying text by deleting the name is not enough; dates, places and rare conditions can identify a patient.
  • Personal accounts on AI tools are never used for practice work, even for non-patient tasks, so that data does not end up outside practice control.
  • Screenshots and photos of the EHR are patient information.
  • Approved tools are used only for their approved purposes. A scribe approved for notes is not approved for drafting referral letters unless the register says so.

Review duties

Every approved use has a named reviewer type. Clinical content reaching the record or a patient is reviewed by a licensed clinician. Administrative content is reviewed by sample at a set frequency. The policy should say who, how often and what they record.

  • Notes drafted by AI are read and corrected by the signing clinician before signing; signing without reading is a policy breach.
  • AI-drafted messages about symptoms, results or medications are approved by a clinician or nurse before sending.
  • Phone and chat assistants are reviewed weekly by sample, plus every emergency escalation.
  • Reviewers log errors in the shared error log with date, tool, grade and fix.

Disclosure, fairness and access

State that patients are told when they are interacting with AI, in the first sentence or line of every automated interaction, with a route to a person. Some states have specific disclosure rules; see state laws on AI in patient communications.

Add a fairness clause. Under Section 1557 rules, covered entities must make reasonable efforts to identify and mitigate discrimination risks from patient care decision support tools, including AI tools that use factors such as race, national origin, sex, age or disability[5]. For a small practice, that means asking vendors about it, testing tools in the languages patients use, and keeping front office AI out of clinical decisions.

Incident steps

  1. Report

    Anyone who sees an AI error or a possible privacy problem reports it the same day using the error log or by telling the practice manager.

  2. Contain

    The tool owner fixes or pauses the affected use straight away.

  3. Assess

    The owner grades the incident. Safety issues go to the medical director; possible privacy breaches go to the privacy officer the same day.

  4. Notify where required

    If protected health information was disclosed improperly, follow the breach process. HIPAA requires notification to affected individuals without unreasonable delay and within 60 calendar days of discovery[6].

  5. Fix and record

    Correct the cause, retest, and record the incident and outcome in the log.

Training, records and review

  • Everyone reads the policy and signs that they understand it at hire and each year.
  • Staff using an approved tool complete its specific training before first use.
  • Keep the policy, each version, the register, signed acknowledgments and incident logs. HIPAA requires covered entities to retain required policies and compliance documentation for six years from creation or the date last in effect[3].
  • Review the policy at least once a year and whenever a new tool is added or an incident reveals a gap.

Putting the policy in place in two weeks

A policy that sits in a shared drive changes nothing. The rollout matters as much as the wording, and for a small practice it can be done in two weeks without outside help, with counsel reviewing the final draft.

  1. Days 1 to 3: ask every clinician and staff member, without blame, which AI tools they use for work, including free ones and features inside existing software.
  2. Days 4 to 6: build the register from the answers, and check each vendor for a business associate agreement where patient information is involved.
  3. Days 7 to 9: draft the policy using the sections above, keeping each section to a few short paragraphs.
  4. Day 10: have counsel or your compliance adviser review the draft, especially the patient information and incident sections.
  5. Days 11 to 12: present it at a staff meeting with real examples of what is and is not allowed, and collect signed acknowledgments.
  6. Days 13 to 14: remove or block tools that are not approved, update onboarding materials, and set the annual review date in the calendar.

Expect the first inventory to surprise you. Staff often use AI features they do not think of as AI, such as transcription in the phone system or writing suggestions in email. Include them; the register is only useful if it is complete.

Example: a fictional three-physician practice

Example: a fictional family practice discovers during a staff meeting that two employees use a free chatbot to rewrite patient letters, sometimes pasting in the original letter with the patient's name. The practice manager drafts a four-page policy using the structure above, with a register listing the phone assistant, the EHR's portal drafting feature and a general chatbot approved for non-patient writing only.

Example, continued: the privacy officer assesses the earlier letters with counsel to decide whether any notification is needed and documents the decision. Every staff member signs the policy, and the manager adds a ten-minute AI section to new-hire orientation. Six months later, when a physician asks to trial an ambient scribe, the request goes through the register process, and the vendor signs a business associate agreement before the pilot starts. For choosing between tools, see ambient scribes vs front office AI.

Frequently asked questions

What should an AI policy for a medical practice include?

Scope, an approved tools register, rules on patient information, human review duties, patient disclosure, fairness and access, incident steps, training and records. A small practice can cover all of it in a few pages.

Can staff use free AI chatbots at work?

Only for tasks the policy allows and never with patient information, unless the tool is approved for it and the vendor has signed a business associate agreement. Many practices allow general chatbots for non-patient writing only.

How long should we keep our AI policy records?

HIPAA requires covered entities to keep required policies and compliance documentation for six years from creation or the date last in effect, so keep every version, acknowledgments and incident logs at least that long.

References

  1. Medical Group Management Association (MGMA) (2026). AI governance in medical group practices: Rules for the humans in the loop (opens in a new tab) ↑

  2. Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (current). 45 CFR § 164.502 - Uses and disclosures of protected health information: General rules (opens in a new tab) ↑

  3. Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (current). 45 CFR § 164.530 - Administrative requirements (opens in a new tab) ↑

  4. KFF (2026). KFF Tracking Poll on Health Information and Trust: Use of AI For Health Information and Advice (fielded Feb 24 - Mar 2, 2026) (opens in a new tab) ↑

  5. Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (current). 45 CFR § 92.210 - Nondiscrimination in the use of patient care decision support tools (opens in a new tab) ↑

  6. Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (current). 45 CFR § 164.404 - Notification to individuals (opens in a new tab) ↑

Book a demo

See Ava handle your calls.

Book a short demo. We’ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo