Key takeaways
- AI is good at turning unstructured input, such as photos of cards, faxed referrals and phone requests, into draft fields. It is not reliable enough to file them without review.
- Keep a person on identity matching, insurance exceptions, consent forms and anything clinical that reaches the chart.
- Patients increasingly use digital tools, and caregivers do too: proxy portal access more than doubled from 2020 to 2024[1].
- Measure intake by errors caught at check-in and time to a complete record, not by how many forms were sent.
What intake really involves
New patients made up 16.8% of US physician office visits in 2019, about 173.9 million visits[2]. Each one starts with an intake process that touches several systems and several people. Before deciding what to automate, list the parts. Most practices find that intake is not one task but six or seven, and they vary widely in how much judgment they need.
- Demographics: name, date of birth, address, phone, email, emergency contact, preferred language.
- Insurance: plan, member ID, group number, subscriber details, and a photo of the card front and back.
- Identity: photo ID and matching the patient to any existing record to avoid duplicates.
- Consents and acknowledgments: financial policy, privacy notice acknowledgment, communication preferences, telehealth consent where relevant.
- Clinical history: medications, allergies, problems, surgical and family history, screening questionnaires.
- Referral and records: referral letter, prior records, imaging reports, sometimes still by fax.
- Scheduling details: reason for visit, visit type and the provider the patient should see.
Each part has its own risk if it goes wrong. A misspelled street name causes a returned statement. A wrong member ID causes a denial. A merged record with another patient's allergies can cause harm. The right level of automation is different for each.
What is realistic today
| Intake task | What AI can do well | Human review needed | Why |
|---|---|---|---|
| Insurance card capture | Read fields from a photo into draft fields | Yes, before eligibility runs | Card layouts vary; one wrong digit fails the check |
| Demographics by phone | Collect and read back structured details | Spot check | Spelling of names and addresses |
| Referral faxes | Extract patient, referring provider, reason and urgency into a summary | Yes, every referral | Urgency and clinical detail must be judged by staff |
| Duplicate record check | Flag likely matches | Yes, always | Merging records wrongly is hard to undo |
| Consent forms | Send, track and remind | Confirm signed version is filed | Legal record |
| History questionnaires | Pre-fill from patient answers and prior records | Clinician reviews at the visit | Clinical content |
| Visit type and routing | Suggest based on reason and rules | Yes for new patients | Wrong visit type wastes the slot |
The pattern is consistent. AI performs well at the first draft: converting a photo, a fax or a phone conversation into fields. It performs poorly as the final authority on anything that is hard to undo or clinically meaningful. A realistic plan uses AI to draft and people to approve, then moves specific fields to automatic filing only after weeks of measured accuracy.
Where AI saves the most typing
In MGMA's 2026 polling, practice leaders ranked intake well below eligibility, prior authorization and scheduling as a time-consuming phone task, at 9%[3]. The savings from intake automation mostly show up away from the phone: fewer minutes re-keying data at check-in, fewer callbacks to fix missing information, and fewer claims rejected for demographic errors.
- Card and ID capture: patients photograph cards before the visit, the tool drafts fields, and staff confirm in seconds rather than typing from scratch.
- Referral intake: faxed or emailed referrals are summarized into a standard form with the key fields highlighted, so the referral coordinator reads one screen instead of eight pages.
- Phone-to-form: when a new patient calls, the details collected on the call pre-fill the registration record and the digital forms link sent afterwards.
- Form follow-up: the system tracks which forms are incomplete and sends reminders at set intervals, escalating to a staff call two days before the visit.
Where staff review stays essential
- Identity and duplicates. A person decides whether a new registration matches an existing patient.
- Insurance exceptions. Secondary coverage, coordination of benefits, workers' compensation and out-of-network situations.
- Minors and proxies. Confirming who can consent and who can access the record.
- Clinical content entering the chart. Medication and allergy lists from forms are reconciled by clinical staff at the visit.
- Urgent referrals. A summary can flag the word urgent; only staff can decide the appointment needs to happen this week.
- Anything the patient flagged as wrong. Corrections from patients always go to a person.
Design for how patients actually behave
Patients are more digital than many intake processes assume. In 2024, 65% of US individuals were offered and accessed their online records or portal, up from 25% in 2014[1]. App-based access rose sharply, and caregiver access more than doubled.
- 2020
- 2024
Source: Office of the National Coordinator for Health IT (ASTP/ONC), HHS, 2024[1]
Show the data
| Item | 2020 | 2024 |
|---|---|---|
| Access through an app | 38% | 57% |
| Proxy or caregiver portal access | 24% | 51% |
Three design points follow. First, send intake links by text, because about nine in 10 adults own a smartphone[4]. Second, allow a caregiver to complete forms on behalf of a patient, with the relationship recorded. Third, tell patients personally that the forms matter: ONC found people encouraged by their provider to use the portal used it at higher rates[1]. A line from the scheduler such as please complete these before you come so we can spend your visit on you outperforms an automated email alone.
Keep a paper and phone route for patients who cannot or will not use digital forms. Automation should shorten the queue for everyone, not create a second-class process for older or less digital patients.
Accessibility and language
Digital intake must work for patients with disabilities and limited English. Under the amended Section 504 rule, recipients of HHS federal financial assistance with 15 or more employees must make web content and mobile apps meet WCAG 2.1 Level AA from May 11, 2027, and smaller recipients by May 10, 2028[5]. Check your form vendor's accessibility conformance report now.
If you use machine translation for forms, Section 1557 rules require that critical or technical patient content translated by machine be reviewed by a qualified human translator[6]. Consent forms and financial policies clearly qualify.
A rollout plan
Measure the baseline
For two weeks, count incomplete registrations at check-in, demographic claim rejections and minutes per new-patient registration.
Start with drafts
Turn on card capture and referral summaries in draft-only mode. Staff approve every field.
Track field accuracy
Log every correction by field. Member ID, date of birth and name spelling usually need the longest review period.
Promote fields one at a time
Only fields with consistently low correction rates move to automatic filing, and only with a daily exceptions report.
Keep the human gates
Identity, duplicates, consents and clinical content stay reviewed permanently.
Questions to ask an intake vendor
Demonstrations of intake tools tend to show clean, typed documents. Your reality is crooked card photos, handwritten referral notes and patients who abandon a form halfway through. Bring your own fictional test documents to every demo and ask:
- Which fields does the tool draft, and which does it file without review? Can I control that per field?
- How does it show confidence for each extracted field, and can low-confidence fields be forced to review?
- Where do card images and referral documents go, how long are they kept, and who can see them?
- Does it write into my practice management system directly, or does someone copy and paste?
- How does it handle a patient who already exists in my system under a slightly different name?
- What happens when a patient starts a form on their phone and finishes it at the desk?
- Can a caregiver complete forms for a patient, and how is that relationship recorded?
- Will you sign a business associate agreement, and what does it cover?
Phone script for new patients: “I'll text you a secure link to complete your registration and take a photo of your insurance card. It takes about ten minutes. If anything's tricky, bring your card and ID and we'll finish it together at the desk.”
A vendor that cannot answer the first three questions clearly is selling a demonstration, not an intake process. The last question is not optional: any vendor that handles patient information on your behalf needs a signed agreement before go-live.
Example: a referral-heavy specialty office
Example: a fictional cardiology office receives most new patients by referral fax. Before automation, a coordinator reads each fax, types the patient into the system and calls to book. The office turns on referral summaries and phone-to-form capture in draft mode. The coordinator now reviews a one-page summary, corrects fields and approves.
Example, continued: after a month, the correction log shows the tool often confuses the referring provider with the practice that sent the fax, and misreads handwritten dates. The office keeps those fields under review, promotes phone number and insurance plan name to automatic filing, and adds a rule that any referral containing the words urgent or ASAP goes to the coordinator's priority queue within an hour.
Measuring intake automation
- Share of new patients arriving with a complete registration.
- Minutes per registration at the desk, before and after.
- Demographic and eligibility rejections per hundred claims.
- Corrections per field per week during review.
- Duplicate records created per month.
Intake rarely stands alone. Eligibility checks are the next step, covered in AI for insurance verification, and the after-hours side is in AI for after-hours practice calls.
Frequently asked questions
Can AI fill in patient registration automatically?
It can draft fields from card photos, phone conversations and referral documents. Most practices should keep staff review on every field at first, then move individual fields to automatic filing once correction rates stay low.
Should AI review a patient's medical history forms?
It can organize answers and pre-fill from prior records, but clinical content entering the chart should be reconciled by clinical staff at the visit.
Do digital intake forms need to be accessible?
Yes. HHS-funded entities face Section 504 deadlines to meet WCAG 2.1 Level AA for web content and apps, and machine-translated critical content must be reviewed by a qualified human translator under Section 1557.
References
Office of the National Coordinator for Health IT (ASTP/ONC), HHS (2024). Individuals' Access and Use of Patient Portals and Smartphone Health Apps, 2024 (Data Brief No. 77) (opens in a new tab) ↑
CDC National Center for Health Statistics (2019). National Ambulatory Medical Care Survey: 2019 National Summary Tables (opens in a new tab) ↑
Medical Group Management Association (MGMA) (2026). Phones are still a bottleneck costing medical practices time they can't afford (opens in a new tab) ↑
Pew Research Center (2026). Americans' Use of Mobile Devices in 2026 (Mobile Fact Sheet) (opens in a new tab) ↑
Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (2026). 45 CFR § 84.84 - Requirements for web and mobile accessibility (opens in a new tab) ↑
Legal Information Institute, Cornell Law School (Electronic Code of Federal Regulations text) (current). 45 CFR § 92.201 - Meaningful access for individuals with limited English proficiency (opens in a new tab) ↑