Responsible AI
AI proposes. People decide.
Accessibility work touches people at their most vulnerable moment — applying for help. So AccessForce is built so that AI changes how a question is asked, never who gets an answer. The person filling the form stays in control, and so does the agency behind it.
Six commitments
AI proposes, people decide
Every audit finding and every rewritten question is a recommendation. A human reviewer approves the redesigned experience before anyone uses it.
What that means hereIn this prototype, the audit and redesign screens show the AI's proposal, not a published change.
A person is always one tap away
The conversation never becomes the only door. 'Talk to a person' is visible at every step, and answers stay saved when someone switches to a caseworker.
What that means hereTry the 'Talk to a person' button on any question in the experience.
AI never decides eligibility
AccessForce changes how a question is asked, never what the rules are. Decisions about benefits stay with the agency and its existing systems.
What that means hereThe redesigned flow writes the same fields the legacy form always collected.
Ask for the least, keep it the shortest time
Questions that can be derived or reused are removed instead of repeated. Applicants' answers are not used to train models.
What that means hereHousehold size is asked once instead of four times.
Check who the AI could still miss
Plain-language and translation output is reviewed with disabled people and speakers of each language, because a model's idea of 'simple' is not the same as a person's.
What that means hereThe audit scores barriers by human impact, and each finding names who is affected.
Say what it cannot do
An AI score is evidence, not a certificate. It complements manual and assistive-technology testing; it does not replace it.
What that means hereScores on this site are labelled as AI assessments of a fictional sample form.
Human oversight
Where a person checks the AI's work
Checkpoint 1
Audit
An accessibility specialist confirms findings with a screen reader and keyboard before the report is trusted.
Checkpoint 2
Rewrite
Program staff confirm the plain-language question still asks for the same legal information.
Checkpoint 3
Publish
Nothing goes live automatically — a named owner approves the redesigned flow.
Checkpoint 4
In use
Drop-off per question is monitored, and any step people abandon is sent back for review.
Language and measurement
Translation with human review. Analytics without surveillance.
AI-assisted translation, agency-approved wording
- AI drafts a language experience; a reviewer approves it before anyone sees it.
- The agency's glossary, prohibited wordings and exact legal phrases always override what AI would choose.
- Certified translation review stays available where policy requires it — AI does not replace it.
- Offering a language is never an eligibility decision, and neither is helping someone navigate to the right process.
Privacy-conscious accessibility analytics
- Aggregate only: counts, times and drop-off by step, language and chosen preference.
- No individual answers, documents or identities are exposed in reporting.
- Accessibility preferences are used to find barriers, never to treat an applicant differently.
- Every prototype figure is labelled illustrative or projected, and is never presented as a measured outcome.
Accountability
Who audits the accessibility auditor?
A tool that scores other experiences has to be held to the same standard. Here is what AccessForce itself implements, and what it does not claim.
AXS — AccessForce Accessibility Experience Score
96 / 100
AccessForce's own illustrative AXS score, using the same dimensions.
Implemented in this prototype
- Keyboard navigation
- Visible focus
- Semantic structure
- Screen-reader labels
- High contrast
- Adjustable text
- Reduced motion
- Large touch targets
- Plain language
- Responsive design
- Accessibility preferences
Not claimed
- No assistive-technology user testing has been run on this prototype, so no credit is claimed for it.
- Machine translation and read-aloud are demonstrated as interactions, not shipped as working services.
- No formal WCAG conformance audit has been performed by a professional accessibility tester.
AXS is designed to identify improvement opportunities and compare experiences before and after redesign. It does not certify WCAG, ADA, Section 508 or other legal compliance.
What this prototype cannot do
- This is a prototype built for a hackathon. The audit runs on a fictional sample form, not a live agency system.
- The scores, barrier counts and impact figures are illustrative, not measured outcomes.
- No connection to a real system of record is made, and no personal data is collected by this site.
- Automated checks can only find part of what is wrong. Testing with disabled people remains essential.
Form Finder
Helping someone find a process is not deciding their case
The Form Finder suggests where a person may want to look. It labels every result a likely or possible match, links only to official government sources, and separates four different things that are easy to blur together.
Navigation assistance
AccessForce helps you find a likely agency, program and form.
Process explanation
AccessForce explains what a process asks for, in plain language.
Official agency requirements
The published requirements always come from the agency's own site.
Eligibility determinations
Only the agency decides. AccessForce never says “you qualify”.
- “This program may be relevant.”
- “This form appears likely to apply.”
- “Let's check a few details.”
- “You may want to confirm with the agency.”
AccessForce helps people navigate. Government agencies make final determinations.
Open the Form Finder