Test a draft on simulated readers first
Before a letter goes to real members, try it on simulated readers. They are built from your own personas. See where they get lost, which numbers trip them up, and what they think they should do. Then fix it and test again.
Plan for the people you serve
What it does
- Your readers, not generic ones. Personas come from your own data and research.
- Real comprehension probes. Cloze, teach-back, action and numeracy questions.
- Honest about limits. Confidence bands and a limitations statement on every export.
- Decisions on record. Each test ends in a change plan your team signs off.
Your reader panel
Readers built from your evidence, and honest about it
Each simulated reader carries a grounding level from 1 to 4. The level shows how much real evidence stands behind it, and every persona lists its sources.
Grace, 52
Grounding level 1
Your own description
Marta, 71
Grounding level 2
Published population statistics
Devon, 34
Grounding level 3
Your organization's data
James, 64
Grounding level 4
Voice-of-customer research
Illustrative · synthetic personas. AI fills empty fields only after you approve.
Comprehension probes
Four ways to ask, "Did they get it?"
Each reader answers the same questions about your draft. Every probe checks a different part of understanding.
Cloze
Fill in the missing word.
Shows whether the words make sense to this reader in context.
Teach-back
Explain it back in your own words.
Shows whether the reader can say the key point, not just repeat it.
Action
What should you do next?
Shows whether the reader can find the step you want them to take.
Numeracy
What will you pay?
Shows whether doses, dates and dollar amounts come through right.
Every test ends with
- Scores with confidence bands
- A limitations statement
- A change plan your team signs off
- A decision record
- HTML, PDF, CSV or JSON export
How it works
Step 1: Pick personas
Built from your data or research, each with a grounding level of 1 to 4.
Step 2: Run probes
Cloze, teach-back, action extraction and numeracy.
Step 3: Read results
Scores with confidence bands, plus a virtual focus group.
Step 4: Compare
Test version A against version B.
Step 5: Decide
A change plan and a decision record; export to HTML, PDF, CSV or JSON.
An illustrative result
Five simulated readers answer the same three probes on version A, then on the revised version B.
Action probe
"What should you do next?"
Version A2 of 5 named the step
Version B5 of 5 named the step
Version B states the action first.
Numeracy probe
"What will you pay?"
Version AMixed answers
Version BConsistent answers
Version B shows one number, not a range.
Cloze probe
Fill in the missing word
Version ALow band
Version BHigher band
Version B uses familiar words. The shaded band is the confidence band.
Synthetic example
One notice, three simulated readers
A health plan tests a drug-list change notice before it goes out. Goal: the reader can say what they will pay and what to do if they cannot pay it. Pick a simulated reader to see their profile and what the test found.
Version A · tested text · Flesch-Kincaid 8.6
Starting January 1, your plan's drug list will change. Your cholesterol medicine, atorvastatin 40 mg, will move from Tier 1 to Tier 2. Depending on your pharmacy, your copay may range from $10 to $47 for a 30-day supply after you meet your deductible. To ask for an exception, you can submit a coverage determination request.
Marta, 71
MemberMedicare Advantage member. Reads her plan mail with her daughter.
L2 · Population-grounded · 40%
40% of the core attribute groups have current evidence.
- Primary Language
- Spanish · Requires Language Assistance
- Health Literacy
- Marginal
- Financial Literacy
- Marginal
- Reading Level
- 5th grade
- Preferred Channels
- Phone call, mail
- Health Locus of Control
- Powerful others
- Support System
- Daughter helps with mail
How Marta responded
- Q · What will you pay for a 30-day supply?Fail
- Q · What can you do if you cannot pay that amount?Partial
- Stated uncertainty:
- "I don't know what Tier 2 means for me."
- Missing from this answer:
- The $10 price at a preferred pharmacy
Potential communication friction · Priority 1 · HIGH
"your copay may range from $10 to $47 for a 30-day supply after you meet your deductible"
Reduce the work required to interpret numbers
- Numbers
- Plain language
Editable replacement: You will pay $10 for a 30-day supply at a preferred pharmacy. At other pharmacies, you will pay $47.
Devon, 34
MemberWorks night shifts. Manages his family's plan on his phone.
L3 · Org-data grounded · 55%
55% of the core attribute groups have current evidence.
- Primary Language
- English
- Health Literacy
- Adequate
- Financial Literacy
- Marginal
- Reading Level
- 8th grade
- Preferred Channels
- Text message
- Technology Access
- Smartphone only
- Perceived Barriers
- High: cost and time
How Devon responded
- Q · What will you pay for a 30-day supply?Partial
- Q · What can you do if you cannot pay that amount?Fail
- Stated uncertainty:
- "Where do I send a coverage determination request?"
- Missing from this answer:
- A phone number to call
Potential communication friction · Priority 1 · HIGH
"you can submit a coverage determination request"
Add a usable callback path
- Action clarity
Editable replacement: Call us at 1-800-555-0142 (TTY 711). We can help you ask for an exception.
Grace, 52
CaregiverCares for her father. Compares plan details online.
L1 · Built from your description · 10%
10% of the core attribute groups have current evidence.
- Primary Language
- English
- Health Literacy
- Proficient
- Digital Literacy
- Proficient
- Reading Level
- 10th grade
- Preferred Channels
- Caregivers
- Is the caregiver for her father
- Self-Efficacy
- High
How Grace responded
- Q · What will you pay for a 30-day supply?Pass
- Q · What can you do if you cannot pay that amount?Partial
- Stated uncertainty:
- "I would have to read it twice to find what to do."
- Missing from this answer:
- What to do first
Potential communication friction · Priority 1 · MEDIUM
"To ask for an exception, you can submit a coverage determination request."
Put the next action where readers can find it
- Action clarity
- Plain language
Editable replacement: Can you not pay this amount? Call us at 1-800-555-0142 (TTY 711).
Then: accept the changes and retest
AI drafts · people decide · the model never applies an edit. Your team accepts each replacement, then runs the same test on version B.
Version B · Flesch-Kincaid 5.0
Starting January 1, your cholesterol medicine will cost more. Your medicine is atorvastatin 40 mg. You will pay $10 for a 30-day supply at a preferred pharmacy. At other pharmacies, you will pay $47. Can you not pay this amount? Call us at 1-800-555-0142 (TTY 711). We can help you ask for an exception.
Comparison · matched conditions
Improved under matched conditions
- Marta, 71 Q1: Fail to Pass Q2: Partial to Pass
- Devon, 34 Q1: Partial to Pass Q2: Fail to Pass
- Grace, 52 Q1: Pass to Pass Q2: Partial to Pass
Stability: single sample per profile; not measured.
Synthetic example · made-up personas and notice. Simulation evidence—not real-patient validation, predicted behavior or clinical outcome. Grades are the Flesch-Kincaid formula on each version.
Behavior change, built in
The behavior science behind every persona and journey
Clear words are where change starts, not where it ends. Personas record what helps and what blocks each reader, and journeys use established behavior-change techniques at each touchpoint. Every message is built to help people understand, decide and act.
- Capability: Can they understand it and do it?Plain language at the persona's grade, one action per message, teach-back to confirm.
- Opportunity: Does their world make it possible?The right channel, language and time; a ride, a phone number, a caregiver who can help.
- Motivation: Do they want to, right now?Messages that answer the persona's own barriers and fit their stage of change.
A different message for each stage of change
- Stage 1: Not yet thinking about itWhy it matters to you
- Stage 2: Thinking about itWhat gets in the way, answered
- Stage 3: Getting readyHow to start, step by step
- Stage 4: Doing itFeedback on progress
- Stage 5: Keeping it upReminders tied to routines
Stages from the Transtheoretical Model (Prochaska and DiClemente).
Techniques at work in personas and journeys
- Stages of changeTranstheoretical ModelEach persona records where the reader is, from not yet thinking about it to keeping it up. Journeys send the message that fits the stage.PersonasJourneys
- Barriers and benefitsHealth Belief ModelPersonas capture what readers believe about cost, risk and benefit. Each message answers the barrier the persona names.PersonasJourneys
- Who they trustHealth locus of control, self-efficacyPersonas note whether readers decide for themselves or lean on a doctor or family, and how confident they feel. Tone and messenger follow.Personas
- One clear actionCognitive load, plain languageEvery touchpoint asks for one thing, at the persona's reading level, and puts it first.PersonasJourneys
- When, where, howImplementation intentionsReminders turn "should" into a plan: the day, the place, who comes along, what to bring.Journeys
- Cues and routinesHabit formationMessages arrive at the moment the action happens, tied to something the reader already does.Journeys
- Make it easyChoice architectureReply 1 to book, a ride offered, a number to call. Help is the default, and nothing is pressured or hidden.Journeys
- Teach-backAHRQ Health Literacy Universal PrecautionsStaff confirm understanding by asking people to say it back, and practice first in the Teach-Back simulator.PersonasJourneysTeach-Back
- Their own reasonsMotivational interviewingCall guides and simulator scenarios ask for the person's own goals and worries, instead of lecturing.JourneysTeach-Back
What we never do. No fear appeals, false urgency or hidden choices. Behavior change supports your quality and adherence measures; the result is measured against your own baseline, not promised.
Built for healthcare, compared with general-purpose AI
| What you need | General-purpose AI out of the box | HealthLiteracyCopilot | Rung |
|---|---|---|---|
| Reviewer workflow and approvals | Not provided | Approve, edit or skip each change; a decision record for each document | L2 |
| Version history and audit trail | Chat history only | Locked versions and an append-only audit trail you can export | L2 |
| Testing with simulated readers | Not provided as a method | Persona testing: cloze, teach-back, action and numeracy probes, with confidence bands | L3 |
Reviewer workflow and approvalsRung: L2
- General-purpose AI out of the box
- Not provided
- HealthLiteracyCopilot
- Approve, edit or skip each change; a decision record for each document
Version history and audit trailRung: L2
- General-purpose AI out of the box
- Chat history only
- HealthLiteracyCopilot
- Locked versions and an append-only audit trail you can export
Testing with simulated readersRung: L3
- General-purpose AI out of the box
- Not provided as a method
- HealthLiteracyCopilot
- Persona testing: cloze, teach-back, action and numeracy probes, with confidence bands
See all 16 differencesHide the full comparison
| What you need | General-purpose AI out of the box | HealthLiteracyCopilot | Rung |
|---|---|---|---|
| Scoring against health-literacy standards | Can simplify on request; no built-in N-PLAT or PEMAT scoring | N-PLAT and PEMAT scores, item by item | L2 |
| Spanish measured as Spanish | No built-in Spanish readability measures | SOL, Fernández-Huerta and INFLESZ, with targets per document type | L1 |
| Meaning kept after a rewrite | Depends on the prompt; no automatic check | Meaning check and clinical integrity check on every revision | L2 |
| Your brand voice, glossary and content types | Re-entered in each prompt | Set once in profiles; every team inherits them | L2 |
| Answers grounded in your sources | General knowledge unless you add files | Your knowledge base, with citations; strict mode refuses rather than guess | L2 |
| Reviewer workflow and approvals | Not provided | Approve, edit or skip each change; a decision record for each document | L2 |
| Version history and audit trail | Chat history only | Locked versions and an append-only audit trail you can export | L2 |
| Accessibility review with evidence | Not provided | WCAG 2.1 AA review; a qualified reviewer decides; an attested record is kept | L2 |
| Accreditation evidence | Not provided | Evidence organized against the requirements you track, with binders and share links | L2 |
| Testing with simulated readers | Not provided as a method | Persona testing: cloze, teach-back, action and numeracy probes, with confidence bands | L3 |
| Journeys and channel previews | Not provided | Touchpoint matrix and channel previews for email, text, letter, web and more | L4 |
| Channel limits checked | Not checked | Text message segments and opt-out; email subject and preview text lengths | L2 |
| Numbers flagged for checking | Not flagged | A "Numbers to verify" card for every dose, date and dollar amount | L2 |
| Workforce training | Not provided | ClearComm Academy and the Teach-Back simulator | L3–L4 |
| Security and access | Varies by vendor and plan | HealthcareGPS is SOC 2 Type II audited and HIPAA compliant; separate tenant per organization; role-based access | Enable |
| An independent grade | Grades its own draft if you ask | The scorer never grades its own revisions; what it can't score with confidence goes to a person | L2 |
Scoring against health-literacy standardsRung: L2
- General-purpose AI out of the box
- Can simplify on request; no built-in N-PLAT or PEMAT scoring
- HealthLiteracyCopilot
- N-PLAT and PEMAT scores, item by item
Spanish measured as SpanishRung: L1
- General-purpose AI out of the box
- No built-in Spanish readability measures
- HealthLiteracyCopilot
- SOL, Fernández-Huerta and INFLESZ, with targets per document type
Meaning kept after a rewriteRung: L2
- General-purpose AI out of the box
- Depends on the prompt; no automatic check
- HealthLiteracyCopilot
- Meaning check and clinical integrity check on every revision
Your brand voice, glossary and content typesRung: L2
- General-purpose AI out of the box
- Re-entered in each prompt
- HealthLiteracyCopilot
- Set once in profiles; every team inherits them
Answers grounded in your sourcesRung: L2
- General-purpose AI out of the box
- General knowledge unless you add files
- HealthLiteracyCopilot
- Your knowledge base, with citations; strict mode refuses rather than guess
Reviewer workflow and approvalsRung: L2
- General-purpose AI out of the box
- Not provided
- HealthLiteracyCopilot
- Approve, edit or skip each change; a decision record for each document
Version history and audit trailRung: L2
- General-purpose AI out of the box
- Chat history only
- HealthLiteracyCopilot
- Locked versions and an append-only audit trail you can export
Accessibility review with evidenceRung: L2
- General-purpose AI out of the box
- Not provided
- HealthLiteracyCopilot
- WCAG 2.1 AA review; a qualified reviewer decides; an attested record is kept
Accreditation evidenceRung: L2
- General-purpose AI out of the box
- Not provided
- HealthLiteracyCopilot
- Evidence organized against the requirements you track, with binders and share links
Testing with simulated readersRung: L3
- General-purpose AI out of the box
- Not provided as a method
- HealthLiteracyCopilot
- Persona testing: cloze, teach-back, action and numeracy probes, with confidence bands
Journeys and channel previewsRung: L4
- General-purpose AI out of the box
- Not provided
- HealthLiteracyCopilot
- Touchpoint matrix and channel previews for email, text, letter, web and more
Channel limits checkedRung: L2
- General-purpose AI out of the box
- Not checked
- HealthLiteracyCopilot
- Text message segments and opt-out; email subject and preview text lengths
Numbers flagged for checkingRung: L2
- General-purpose AI out of the box
- Not flagged
- HealthLiteracyCopilot
- A "Numbers to verify" card for every dose, date and dollar amount
Workforce trainingRung: L3–L4
- General-purpose AI out of the box
- Not provided
- HealthLiteracyCopilot
- ClearComm Academy and the Teach-Back simulator
Security and accessRung: Enable
- General-purpose AI out of the box
- Varies by vendor and plan
- HealthLiteracyCopilot
- HealthcareGPS is SOC 2 Type II audited and HIPAA compliant; separate tenant per organization; role-based access
An independent gradeRung: L2
- General-purpose AI out of the box
- Grades its own draft if you ask
- HealthLiteracyCopilot
- The scorer never grades its own revisions; what it can't score with confidence goes to a person
Who uses it
Next step
HealthcareGPS is SOC 2 Type II audited and HIPAA compliant. How we protect your documents
Questions
Does this replace testing with real members?
No. It gives you directional evidence before human review. Use it to fix the obvious problems first, so real testing goes further.
Where do the personas come from?
From your own data or research. AI fills empty fields only after you approve. Each persona shows its sources.
Is any member data sent to the test?
Personas are built from data your organization approves. Your data policy controls what goes to AI.