How do we balance sounding like a serious medical provider with being an approachable consumer brand?
Texas HB 149 took effect on January 1, 2026, and made plain-language AI disclosure part of the healthcare experience. The answer is one stable brand voice with two operating modes: human guidance when the patient needs a clear path, and clinical precision when copy affects identity, eligibility, consent, privacy, limitations, or a care decision.
In 2026, “human” became a disclosure question
A telehealth brand can no longer treat an AI intake assistant as a tone experiment. Under the Texas Legislature's enrolled HB 149, a provider using an artificial intelligence system in relation to healthcare services or treatment must give the disclosure no later than the date the service or treatment is first provided, except in an emergency. The disclosure must meet three stated conditions: clear and conspicuous, written in plain language, and free of a dark pattern.
The enforcement numbers make this an operating issue. For violations not cured after notice, the law sets civil penalties of $10,000 to $12,000 for each curable violation, $80,000 to $200,000 for each uncurable violation, and $2,000 to $40,000 for each day a violation continues.
California draws a related line around identity. The California Legislature's AB 489 prohibits AI advertising or functionality from implying that its advice, care, reports, or assessments come from a licensed natural person. Each use of a prohibited term can count as a separate violation. The bill digest also describes existing California requirements for certain clinical communications generated by artificial intelligence: a disclosure that AI generated the message and clear instructions for reaching a human.
Those rules do not require a cold brand. They require the brand to stop using warmth to blur who or what is speaking. “Your care guide is here” is not harmless if the interface is automated and the label makes a reasonable person believe a licensed human is responding.
This is strategic guidance, not legal advice. Applicability, exceptions, required wording, and review responsibilities should be confirmed with qualified counsel for each state and use case.
Serious does not mean difficult to read
Dense medical language feels safe to the writer because it sounds formal. It can be unsafe for the reader because the point becomes harder to understand.
The U.S. Department of Health and Human Services says as many as half of U.S. adults have limited literacy skills and nearly 9 in 10 have limited health literacy skills in its Health Literacy Online guidance. Those are not edge cases. They are the audience.
Clinical precision means naming:
- what the service does and does not provide;
- which step is administrative and which is clinical;
- who makes a care decision;
- what depends on an evaluation;
- what information is being requested and why;
- when a human is available;
- what happens if virtual care is not the right setting.
None of those points requires a long sentence. “A licensed clinician will review your answers before deciding whether this service is appropriate” is both more clinical and more approachable than “Complete our eligibility process to get started.” The first sentence names the person, action, decision, and uncertainty. The second hides them.
Approachability is not a personality layer added after compliance review. It is the work of making the truth easy to find and understand.
Use two modes, not two brands
The brand's character should stay steady: direct, calm, candid, and precise. The mode changes with the patient's task and the cost of misunderstanding.
Human guidance mode
Use human guidance when uncertainty is about navigation. This mode belongs in scheduling, preparation, progress indicators, administrative reminders, and explanations of what happens next.
Good human guidance:
- uses familiar words;
- names the next action;
- says how long a process or response normally takes only when operations can support the claim;
- tells the patient where to get help;
- acknowledges friction without pretending to know the patient's feelings or diagnosis.
Example: “This form has four sections. You can save your progress after each one. A support team member can help with account or payment questions.”
The example does not say the process is effortless. It gives the patient information that reduces uncertainty.
Clinical precision mode
Use clinical precision when misunderstanding could affect consent, privacy, eligibility, expectations, safety, or the meaning of a care decision. This mode belongs in service scope, clinical intake, AI disclosures, limitations of virtual care, claims, pricing boundaries, and escalation instructions.
Good clinical precision:
- distinguishes a clinician from software and support staff;
- separates a process promise from a treatment outcome;
- preserves uncertainty until an appropriate evaluation occurs;
- states limits next to the promise they qualify;
- gives the patient a human route when the interaction is automated;
- avoids jokes, urgency, and social proof at the point of consent.
Example: “This assistant uses artificial intelligence and is not a clinician. It collects your answers for review. Contact the care team here if you want help from a person.”
That sentence is not a universal legal template. It shows the content architecture: identity, role, limit, and human route.
The mode switch belongs inside the journey
A homepage does not need one permanent mode. Neither does an intake flow. Mark the switch where the patient's task changes.
| Journey surface | Primary mode | The copy must make clear |
|---|---|---|
| Homepage opening | Human guidance | Who the service is for and the next practical step |
| Service scope | Clinical precision | What the service covers, what it does not, and who decides fit |
| Account setup | Human guidance | What information is needed and how long the administrative step takes |
| AI-assisted intake | Clinical precision | That AI is involved, what it does, what it does not do, and how to reach a human |
| Clinical questions | Clinical precision | Why the information matters and that a clinician reviews it |
| Scheduling and reminders | Human guidance | Time, preparation, rescheduling, and support route |
| Consent and privacy | Clinical precision | The decision, data use, limits, and available choices |
| Follow-up | Both | A warm path forward plus exact ownership and escalation instructions |
The switch should not depend on channel. An email can begin in human guidance mode and move to clinical precision when it explains a result or limitation. A chatbot can guide account setup in familiar language, then make its non-human identity unmistakable before collecting information for care.
The seam you should worry about is not the one above
From Pranay Parikh, MD, who disagrees with part of this page:
I would push back on the premise. The best experience does not have a seam. The language should be concordant from the ad to the exam, and if the tone lurches when the stakes rise, that is not a feature. It is a sign the earlier voice was never true.
The variance that actually exists is not consumer versus clinical. It is physician versus physician. Different doctors, different cultures, different backgrounds, different ways of saying the same clinically correct thing. Marketing wants to flatten that into a single brand voice. It would be cleaner. It would also erase the only part of the experience the patient can tell is human.
I would rather surface that than sand it down. Tell the patient who they are seeing and let that person sound like themselves. A brand voice that survives contact with forty different physicians was never a voice. It was a script.
The disclosure rules above are not optional, and the mode table is a reasonable way to keep a large team from drifting. But do not mistake the map for the goal. The goal is that a patient never notices a handoff, because there was never a moment where the company stopped meaning what it said.
Friendly copy cannot cover an operating contradiction
BetterHelp used an approachable consumer voice. The Federal Trade Commission later said the company promised to keep user information private but disclosed data to Facebook, Snapchat, Pinterest, and Criteo for advertising. The FTC's BetterHelp refund page says the company agreed to pay $7.8 million. A first refund round produced nearly $5.2 million, followed by more than $2.6 million in payments to over 534,000 people who accepted the first payment.
The lesson is not “sound more clinical.” It is that voice cannot repair a mismatch between a promise and the system behind it. “Your privacy matters” is weak if the data map, vendor settings, consent flow, and retention practice do not support it. The same rule applies to “continuous care,” “expert review,” and “always here.”
Every risk-bearing phrase needs an owner and evidence. The writer should be able to point from the sentence to a policy, workflow, configuration, staffing rule, or clinical standard. If that chain breaks, the sentence is removed or the operation changes.
Build the voice system in seven steps
-
Write a three-line voice charter. Complete: “We always sound…,” “We never sound…,” and “When the stakes rise, we….” Use behaviors. “Define medical terms on first use” is useful. “Warm and trustworthy” is not.
-
Inventory every speaker. List clinicians, support staff, automated reminders, AI assistants, educational content, and third-party services. Name each one accurately in the interface. Do not let all of them borrow the same “care team” label.
-
Mark risk-bearing moments. Flag eligibility, service scope, AI use, consent, privacy, price, claims, test results, treatment decisions, and escalation. Assign clinical precision mode before copy is written.
-
Create an identity-and-role pattern. For every automated surface, state what it is, what task it performs, what it cannot decide, and how the patient reaches a person. Apply the pattern to the first relevant touchpoint and repeat it where law, risk, or comprehension requires.
-
Translate without deleting meaning. Replace “contraindications” with “reasons a treatment may not be safe or appropriate,” then retain the precise term where clinical or legal review requires it. Plain language reduces decoding work. It does not remove limits.
-
Build a claim register. Record the exact claim, evidence, owner, location, state-specific variation, and next review date. Include operational claims such as response times and privacy claims such as who receives data.
-
Run paired review. A qualified clinical reviewer checks ambiguity, overstatement, and missing limits. A patient-facing operator checks jargon, cold phrasing, and unclear next steps. Privacy and legal reviewers handle the surfaces within their scope. No reviewer gets to flatten every sentence into their own mode.
A five-line test for any high-stakes sentence
Before approving copy, answer:
- Who or what is speaking?
- What decision is being made at this point?
- Which fact, policy, or workflow supports the sentence?
- What could a patient reasonably misunderstand?
- How can the patient reach a human or get the next appropriate step?
If the team cannot answer line three, the claim is unproven. If it cannot answer line four, the copy has not been tested. If it cannot answer line five on an automated clinical surface, the journey is incomplete.
Run the Marketing Triage diagnostic to map human guidance and clinical precision across the patient journey.
Voice cannot rescue a generic strategy, so make the how to differentiate when everyone claims convenience precise before polishing tone. Decide whether the differentiated position changes which language patients need explained, then work the differentiation worksheet before a rebrand with real decision language. The differentiation worksheet shows how those choices fit together.
One of seven patterns. This is one of the seven patterns regulators keep flagging in telehealth advertising. See all seven, with the enforcement language behind each one.
FAQ
Should a telehealth brand avoid humor completely?
No. Use humor only in low-risk moments, and never let it trivialize a concern, blur who is speaking, soften a clinical limit, or carry a claim that needs precision. Eligibility, consent, privacy, and escalation are poor places for a joke.
Does an AI disclosure have to sound robotic?
No. It has to be accurate, prominent, and easy to understand. State that artificial intelligence is involved, define its task and limits, and give the patient a clear route to a human. Required wording and timing vary by law and use case.
Who should own the telehealth brand voice?
Marketing can maintain the voice system. Risk-bearing language needs named clinical, privacy, or legal reviewers as appropriate, while patient-facing operations should confirm that every process promise matches what happens before, during, and after care.