Getting Patients Onto the Patient Portal in the First Place

Most health centers already own every patient portal feature they were sold. Messaging, results, forms, statements, self-scheduling. The features are on. The problem is that a large share of the patient panel has never signed in, so every workflow built on top of the portal quietly runs at a fraction of its intended volume.

Enrollment gets treated as a marketing task, which is why it never gets finished. Someone prints a flyer, the front desk mentions it at check-in when there is time, and the number climbs slowly enough that nobody can tell whether the effort did anything.

Meanwhile every downstream project assumes the portal is there. Digital forms, statement delivery, appointment confirmations, results notification. Each one gets designed for the enrolled share of the panel and then needs a full manual fallback for everyone else, which means the practice is running two processes and paying for both.

The gap is between offered and used, and it is large

National data has been clear about this for years, and it does not describe a technology shortage.

ONC reporting on 2022 survey data found that 73% of individuals were offered online access to their medical records by a provider or insurer, while 57% actually accessed a portal or online record at least once in the past year. Being offered access and using it are two different numbers, and the distance between them is where the operational work sits.

For a health center the stakes scale with the panel, because the portal is not an add-on channel. It is where forms, results, statements, and reminders are supposed to land. A gap of any size across a panel that size decides whether those workflows carry real volume or stay a pilot.

And the patients least likely to enroll are frequently the ones whose care is most disrupted by a missed message. That makes enrollment an access issue rather than a convenience feature.

Invitation sent is not a metric anybody should be reporting

The most common reporting mistake here is counting the outbound action instead of the outcome.

athenaOne separates these cleanly. POST /patients/{patientid}/portalinvitation sends the invitation. GET /patients/{patientid}/portalstatus tells you where that patient actually stands. Those are different facts, and only the second one describes the panel.

A practice that reports invitations sent will show a healthy number every month while the usable share of the panel barely moves. Invitations bounce off dead email addresses, land on a phone the patient no longer has, or arrive and are ignored because the patient does not recognize the sender. None of that is visible in a send count.

The cost of that gap has gone up as the channel has matured. ONC’s review of a decade of patient access data notes that patients used portals to message their providers at a rate of 64% in 2022, up from 53% in 2018, so an unenrolled patient is now excluded from a channel their peers use routinely.

Working from portal status instead changes what the outreach is for. The list stops being everyone and becomes the patients whose status says the invitation never converted, which is a much smaller list and a much more specific conversation. It also lets a practice stop re-inviting the same unreachable contact record every quarter and start fixing the contact record.

The complication: identity gets verified in a second language

This is where health center enrollment differs from enrollment at a suburban specialty group, and it is not solved by translating the invitation email.

Automated language detection works reasonably well until a bilingual caller intermixes languages inside a single sentence, at which point an agent that switches on every cue flips back and forth. Practices also want certainty before switching, because guessing wrong about a patient’s language in a health care setting is its own failure. One community health center’s Spanish-language line surfaced a separate problem underneath the language question entirely: patient identification friction in the non-English flow. Beyond English and Spanish, quality falls off and practices fall back to a human language line.

Enrollment sits directly on top of that friction, because a portal invitation is an identity event. Before an account can be created the patient has to be matched to the right chart, and the verification questions are exactly the ones that get hardest across a language boundary.

So the automation should be built to confirm rather than guess. Establish the language preference explicitly, hold it for the rest of the interaction, verify identity against the chart, then send the invitation to a contact method the patient confirms in the moment rather than the one already on file.

The handoff is any verification that does not resolve cleanly. A patient who cannot be matched with confidence, or who needs a language the automation does not carry well, goes to a person and to the practice’s language line. Nothing about that is a failure of the automation. Routing it to a human quickly is the correct behavior, and building the automation to fail loudly rather than proceed on a weak match is what keeps a wrong chart from being invited into someone else’s record.

Enrollment is a moment, and the moment is the visit

Campaign-style enrollment underperforms because it asks the patient to act at a time when they have no reason to.

The reliable moment is one where the patient already needs something from the practice. Confirming an upcoming appointment, delivering a form that has to be completed before a visit, telling someone their result is ready. In each case the portal is the path to a thing the patient wants right now, which is a different proposition from an email asking them to create an account for future convenience.

This is also why enrollment should run inside the existing outreach rather than beside it. An appointment reminder that ends by offering to send a portal invitation costs nothing extra, reaches the patient at a moment of engagement, and can capture a corrected phone number or email while the patient is on the line.

Adoption of digital tools generally is uneven enough to make this worth doing deliberately. A July 2025 MGMA Stat poll found 71% of practices have fewer than 25% of patients using digital tools to self-schedule, while only 3% reported more than 75%. The capability being present has never been the same thing as the panel using it.

Measure the panel, not the campaign

Four numbers describe whether enrollment is working, and none of them is invitations sent.

Active share of the panel is the headline. Of the patients seen in the last year, what percentage have a usable portal account. That number is what every downstream workflow is actually planning against, and it is the one to put on the wall.

Invitation-to-activation rate tells you whether the invitation itself works. A low rate points at the message, the sender name, or the timing. A high rate with a flat panel share points somewhere else entirely, which is usually contact data.

Contact reachability is the third, and it is the one most practices have never measured. What share of the panel has an email or mobile number that a message can actually reach. Enrollment cannot exceed reachability, so a practice with a reachability problem is running an enrollment campaign against a ceiling it has not identified.

The fourth is enrollment by language and by site. An aggregate number will look acceptable while one site or one language group sits far below it, and the aggregate is exactly what hides the disparity you would want to act on.

Key Takeaways

Report the share of your seen-in-the-last-year panel with a usable portal account. Invitations sent describes your effort, not your panel.
Read portal status per patient rather than assuming an invitation converted, and build outreach lists from the patients whose status shows it did not.
Treat a portal invitation as an identity event. The verification step, not the email copy, is what fails most often in a multilingual panel.
Establish language preference explicitly and hold it, rather than switching on every cue. Bilingual patients mix languages mid-sentence and a reactive agent flips back and forth.
Route weak identity matches to a person immediately. An automation that proceeds on a weak match can invite the wrong person into a chart.
Attach enrollment to a moment the patient already needs something: a reminder, a form, a result. Standalone enrollment campaigns ask for action with nothing attached.
Capture corrected contact details during outreach. Enrollment cannot exceed the share of the panel you can actually reach.
Break enrollment reporting out by site and by language. The aggregate will look fine while a specific group sits far below it.

A health center that fixes portal enrollment is making the capabilities it already pays for reach the whole panel instead of the share that happened to sign up. That work is unglamorous: verify identity carefully, fix contact records, ask at the moment the patient already wants something, and report on the panel rather than the campaign. An AI team working enrollment inside athenaOne can run that outreach at panel scale, confirm identity before it invites anyone, correct contact data while it has the patient, and hand every uncertain match to a person.

Sources