Some patients never finish a web form: no smartphone, low vision, limited reading or simply no patience for twenty screens. Agnotic Technologies builds chat bots and call bots that ask the same questionnaire as a conversation, confirm every answer and bring in a person the moment an answer needs one.
Trusted by global innovators
AI powered chat bot questionnaires deliver a patient questionnaire as a conversation, by chat or by phone, asking one question at a time, mapping each typed or spoken answer to the questionnaire's coded options and reading it back for confirmation. Agnotic Technologies builds them so validated wording stays exact and risk answers reach a person at once.
The questionnaire itself does not change. It is still a FHIR Questionnaire with fixed items and answer codes; the bot is another way to deliver it. Form building, scoring and EHR write-back are covered on our patient questionnaire software page.
Conversation helps patients who struggle with forms. It also brings failure modes a form does not have, and most of this page is about those.
One question at a time
The Questionnaire drives the conversation. The language model handles the talking, not the content.
The bot starts from the exact Questionnaire version your team published, so every answer can be traced to its item.
Branching follows the Questionnaire's enable rules, or the SDC $next-question operation for adaptive instruments.
A typed or spoken answer is matched to the item's coded options. Free text is kept only where the item allows it.
The bot states the matched option and asks the patient to confirm. A low-confidence match is asked again with the options listed.
Rules your clinicians set are checked before the next question, so a risk answer stops the flow at once.
Answers are saved as a QuestionnaireResponse with item links and a pointer to the transcript, ready for the write-back your EHR supports.
Original decision table
Most programs offer more than one. The choice depends on the patients and on what a wrong answer would cost.
| Question | Web form | Chat bot | Call bot (voice) |
|---|---|---|---|
| Who it suits | Patients comfortable with a screen who want to see every question | Patients who prefer messaging, or who need a question explained | Patients without a smartphone, with low vision or limited reading, or who prefer to talk |
| How answers are captured | Tap or type into the exact answer options | Buttons for coded options, or free text mapped to an option | Speech mapped to an option, with keypad digits as a fallback |
| How answers are confirmed | The patient reviews the whole form before submitting | The bot shows the matched option and asks the patient to confirm or change it | The bot reads the matched option back, with numbers and dates spoken digit by digit |
| Accessibility | Built to WCAG 2.2 for screen readers and keyboards | Built to WCAG 2.2, with time limits the patient can extend | No screen needed: slower pace, repeat on request, transfer to staff |
| Risk answers | Staff alert after submission | Safety message and staff alert during the chat, before the next question | Transfer to a person during the call, or an urgent callback after hours |
| What is stored | QuestionnaireResponse | QuestionnaireResponse and the chat transcript | QuestionnaireResponse, the transcript, and audio only if your policy keeps it |
Many clinics send a web form first and offer a call bot to patients who have not finished it. The same Questionnaire and the same answer codes sit behind all three.
Wording rules
A conversation tempts a language model to reword questions. For a validated instrument, a reworded question is a different question.
Example: Screening tools and patient-reported outcome measures scored by a published method.
Approach: Item text and answer options come from the Questionnaire exactly. The model may repeat them or list the options, never paraphrase. Licence terms are checked before a conversational version is used.
Example: Intake history, preferred pharmacy, transport needs, pre-procedure checks.
Approach: Clinical leads approve one or two alternative phrasings per item, and the bot chooses among them. It does not write new ones.
Example: Greeting, instructions, how to pause and come back, closing message.
Approach: The model phrases these within a style guide, and never asks a clinical question in its own words.
Call bot design
Voice removes the screen, so everything a form shows has to be said, heard and confirmed.
A form can wait until it is submitted. A conversation cannot, so escalation is built into the flow.
Which items and answers count as risk, and the response for each, are agreed with your clinical leads.
The bot stops asking questions and gives the message your clinicians approved, in the patient's language.
In hours, a staff member joins the chat or takes the call. After hours, the route your on-call policy sets.
The answer, the time, the message given and who responded, kept for review under a BAA.
Failure points
The failures we test for before the first patient sees the bot.
Challenge
The model paraphrases a validated item
Agnotic approach
Item text is read from the Questionnaire, and automated tests compare every spoken or shown question with the item text.
Challenge
"Sort of" is recorded as "yes"
Agnotic approach
Answers below a match threshold are asked again with the options listed, and the patient's own words are kept where the item allows free text.
Challenge
The bot keeps going after a risk answer
Agnotic approach
Risk rules run before the next question is chosen, and tests cover every item and answer marked as risk.
Challenge
A patient answers two questions at once
Agnotic approach
The bot confirms each answer separately, so one reply never fills two items without the patient agreeing.
Challenge
Transcripts leak PHI into logs
Agnotic approach
Logs are redacted, transcripts are encrypted at rest and in transit, and access is limited to named roles.
Challenge
Patients drop off halfway
Agnotic approach
Partial responses are saved as in progress, and the patient can resume by chat link or a later call.
Where it applies
Short, structured questionnaires with clear answer options suit chat and voice best.
History and medication questions by chat before the first visit.
Patient-reported outcome measures by phone for patients who do not use the portal.
Fasting, medication and transport questions asked by call bot days before a procedure.
Food, housing and transport questions asked in plain language, with referrals routed to staff.
Short symptom questionnaires by call, with risk answers sent to a nurse.
Conversational questionnaire products that must file answers into a partner's EHR.
Health Insurance Portability and Accountability Act
Protect PHI with privacy-first architecture, encrypted storage and transmission, strict access controls, and traceable audit logs.
General Data Protection Regulation
Implement lawful consent flows, data minimization, retention controls, and secure processing for sensitive health data.
Fast Healthcare Interoperability Resources
Enable standardized health data exchange across apps, care teams, and systems through standards-based FHIR APIs.
Health Level Seven International
Support enterprise-grade interoperability with HL7-based integrations for records, events, and clinical messaging workflows.
Health Information Trust Alliance
Align security programs to healthcare-specific control and risk management practices trusted by providers and ecosystem partners.
Health Information Technology for Economic and Clinical Health Act
Design with breach notification readiness, digital record safeguards, and operational controls that support regulated care programs.
FDA Software as a Medical Device
Plan software quality, traceability, and documentation pathways for products that may require SaMD review and submission.
Medical Device Regulation (European Union)
Prepare EU market-ready processes for risk classification, evidence tracking, and lifecycle governance under MDR expectations.
Substance Abuse and Mental Health Services Administration
Apply confidentiality controls and consent-aware sharing models for behavioral and mental health data experiences.
Short facts we design around, each linked to its primary source.
The HL7 Structured Data Capture implementation guide (STU 4) describes adaptive questionnaires that rely on the $next-question operation to choose and return each next question based on the answers already given.
Source: HL7 FHIR SDC IG, Adaptive QuestionnairesIn FHIR R4, a choice item is answered with a Coding drawn from its answer options or value set, while an open-choice item also accepts a free-text string.
Source: HL7 FHIR R4 QuestionnaireItemType value setEvery artificial or prerecorded voice telephone message must state clearly, at the beginning, the identity of the business or other entity responsible for the call, and must state its telephone number during or after the message.
Source: 47 CFR 64.1200(b) (eCFR)The HIPAA Security Rule requires technical measures that guard against unauthorized access to electronic protected health information sent over an electronic communications network, with integrity controls and encryption as addressable specifications.
Source: 45 CFR 164.312(e) (eCFR)WCAG 2.2 success criterion 2.2.1 Timing Adjustable (Level A) requires that, for each time limit set by the content, the user can turn it off, adjust it, or be warned and given at least 20 seconds to extend it.
Source: W3C WCAG 2.2Scope drivers
We size the work on a scoping call. These factors move effort most.
Chat alone ships sooner than chat plus an outbound and inbound call bot.
Each validated instrument needs a licence check and exact-wording tests.
Each language needs approved item text, speech tuning and its own review.
Live transfer to staff needs telephony and staffing rules; a callback queue is simpler.
Filing a QuestionnaireResponse over FHIR is simpler than mapping answers into EHR flowsheets.
Three stages, starting with one questionnaire in one channel.
We pick one questionnaire and one channel, list its risk answers and agree who responds to them.
The bot, answer mapping, read-back, escalation and response filing, tested with scripted patients first.
More questionnaires, the second channel and more languages, with transcripts sampled and reviewed each month.
