There is no badge for HIPAA compliance. No certificate, no seal, no government body that stamps a piece of software "HIPAA compliant" the way a rating agency grades a bond. When a vendor's homepage uses that phrase, it's describing something that isn't awarded — it's assembled.
What HIPAA compliance actually is: a set of safeguards, implemented and documented, backed by a signed Business Associate Agreement. The safeguards are architecture you can inspect. The BAA is a document you can read. If a vendor won't let you do either before you sign, that's the answer.
Skipping Insurance Claims Doesn't Skip HIPAA
The most common thing we hear from a solo injector or a single-location owner is some version of "I don't bill insurance, so HIPAA doesn't apply to me." That's half a story.
HIPAA draws a line between a Covered Entity — a provider, plan, or clearinghouse that transmits health information electronically in connection with certain transactions — and a Business Associate, which is anyone handling PHI on that entity's behalf. Insurance billing is one path into Covered Entity status. It isn't the only one. HSA and FSA payment rails and electronic transmission of client records both pull a practice into scope, insurance claim or not.
Several state attorneys general also enforce HIPAA-aligned obligations under their own state laws, sometimes with broader scope than federal HIPAA itself. So even in the edge case where a practice sits outside federal HIPAA's reach, a state overlay can still apply.
The real trigger isn't the payment method. It's the PHI itself: medical history, treatment notes, before/after photos. Collect that, and you're handling protected health information. Who bills whom is a separate conversation.
If the BAA Costs Extra, the Product Wasn't Built for This
Here's the opinion most vendors won't say out loud: if compliance is a paid tier, the base product doesn't have it. A BAA gated behind an upper plan isn't a neutral pricing decision. Tenant isolation and an audit log get built into a schema from the start, or they don't exist — you don't bolt them on later as a premium feature.
A vendor offering a cheap entry plan without HIPAA safeguards, then charging more for a version that has them, is telling you the entry tier was never meant to hold PHI. That's worth more of your attention than the price gap itself. The cheaper plan you're comparing against may be running on a different foundation entirely from the one being demoed to you.
Our own answer is to make compliance the floor: the BAA is in the standard contract at every tier, not an upgrade you chase once you've grown enough locations to justify asking for it.
The Four-Question Architecture Checklist
"Do you sign a BAA?" is where every ranked list stops. Necessary, not sufficient — a BAA is a legal commitment, and it says nothing about whether the system underneath was built to honor it. Ask these on the same call:
- Is the BAA in the standard contract, or is it an add-on? If it's an upsell, ask what the base tier is missing that makes it necessary.
- Is there an append-only audit log on every PHI read? Not just who edited a record — who viewed it, and when, in a log that can't be altered after the fact. That's the gap between an audit trail and an activity feed.
- Is tenant data isolated at the database level? Permission gates are software logic, and software logic can fail or get misconfigured. Database-level isolation means one tenant's data physically cannot be queried by another tenant's session.
- Is data encrypted in transit and at rest? Both. Transit protects data moving between systems. At rest protects it sitting on a server. A vendor doing only one is covering half the exposure.
A rep who answers specifically and immediately is describing something real. A rep who pivots to "we take security very seriously" is reading you a line. Our own answers sit at /security and /hipaa — not because we want your trust, but because we think you should be able to check.
Consent Isn't a PDF You Email Beforehand
A lot of practices treat consent as a document problem: email a PDF, get a signature back, file it somewhere. That's a side binder with legal exposure built in — there's no reliable record of which version the client actually signed, or when.
Real e-signed consent is versioned. The form itself carries a version number, and the signature record captures the client's name, a timestamp, the device or IP it was signed from, and the exact version they agreed to. Change your consent language — a new injectable protocol, an updated risk disclosure — and you need to know who signed the old version and who signed the new one. That's what makes the record defensible if it's ever questioned.
The form should auto-send on booking and land on the client's chart, not a separate folder disconnected from the rest of the record. One record, not two systems to cross-reference. More at /consent.
If the Vendor Sells You an AI Texting Agent, Ask These Five Things
An AI agent that reads and writes to a client record sits in PHI-adjacent territory by definition — it needs to know appointments, balances, provider schedules. That's a reason to ask specific questions before turning it on.
- Is the system prompt PHI-free? The instructions governing the AI's behavior shouldn't contain protected health information — it should retrieve what it needs through structured, audited tool calls (live availability, staff schedules, provider service eligibility, client account and package balances) rather than carrying PHI in its base instructions.
- Is there a pre-send pattern scan? A scan checking outbound messages for SSNs, dates of birth, and payment card numbers, blocking the send and escalating to a human on a match, catches the case where the instruction alone doesn't hold.
- Can staff pause a conversation? Per-conversation pause from the inbox, so one thread gets handed to a human without shutting off the whole system.
- Is there a daily send cap? A ceiling on outbound volume limits the damage if something misfires.
- Does it escalate clinical or payment matters to a human? The agent handles booking logistics. Anything clinical or financial gets handed off.
Underneath all five: is the LLM itself running on infrastructure eligible for a BAA. OpenAI, Anthropic, and Google all have BAA-eligible paths — not every deployment on top of them is configured to use one. Our own agent is built this way; details at /ai-sms-agent.
SMS Was Never Encrypted End-to-End — That's Why Scope Matters
SMS isn't encrypted end-to-end. That part is true and worth taking seriously. But the exposure is PHI content landing in an unencrypted channel — not SMS as a category.
That's why the operating rule matters more than the channel: AI-generated SMS should stay on booking logistics and escalate anything clinical or payment-related to a human. A text confirming a Tuesday appointment carries essentially no PHI risk. A text discussing a treatment complication does. Same channel, different exposure entirely.
No One Can Promise You'll Never Have an Incident
We frame this as defensible and addressable, not as a guarantee, because that's the honest version. No vendor — us included — can promise a practice will never have an incident. What a BAA, an append-only audit log, tenant isolation, and encryption actually buy is a position to stand on if you're ever investigated: a documented answer to "what did you have in place." That beats scrambling to explain after the fact with none of it.
Before You Sign
Ask in this order: BAA in the standard contract or an upsell. Audit log append-only or not. Tenant data isolated at the database level or permission-gated. Encryption in transit, at rest, or both. If AI texting is part of the deal, ask the five that go with it.
Don't take a homepage's word for it — ask a rep to walk through it live. Our own version of this is the Pro tier at $249 a month, bundling the AI SMS agent with the CRM as one working example of a BAA-inclusive, audited setup end to end. But the checklist matters more than which vendor you run it against. Pricing detail at /pricing.