AI interfaces · chat, copilot, voice
Accessibility testing for your AI assistant.
For assistants already live, preparing to launch, or undergoing a review. Find where users get stuck, from the first prompt to replies, errors and action confirmation. Get reproducible findings, practical fix guidance and evidence of what was tested.
Where AI interfaces break
Three kinds of interface, each with its own failures.
-
Support and sales chat
- The reply streams in and the screen reader says nothing, or reads every token. The panel opens and focus stays on the page behind it.
- Ask, wait, read the reply, follow a citation, close the panel and return to the page. Browser and screen reader combinations are agreed before testing.
-
Copilots inside your product
- Suggestion chips with no names, a stop button the keyboard can’t reach, a transcript you can’t move through, and generated content in colours nobody checked.
- Keyboard operation, focus and control names in the agreed journeys, navigation through the transcript, and the structure and contrast of sampled replies.
-
Voice assistants
- A voice-only path with no text alternative, timeouts that cut off slow speakers, and no way to review what was understood.
- Voice input, text alternatives, timing and error recovery in the agreed tasks. Devices and access methods are defined before testing.
How the review can help
Evidence for the work ahead.
- Understand reported barriers
- If users report difficulty with your assistant, testing documents the affected tasks, reproduction steps and findings your developers can investigate.
- The assistant is changing
- A new chat component, model or prompt can change the replies people must navigate. Saved response examples and repeatable checks give your team a baseline for the next release.
- A standard has been requested
- Share the standard, edition and product scope your buyer requested. We agree which evidence this review can provide and whether a wider product assessment is needed.
The package
AI interface audit and record
One chat, copilot or voice interface, its host page and two agreed journeys. Results within 10 business days after scope, test access and response samples are ready. Additional interfaces and wider product coverage are quoted separately.
- One chat, copilot or voice interface and its host page, tested with the access methods agreed for that interface
- Two agreed journeys, with a written list of states and response samples: streaming, stop and retry, errors, citations and action confirmation where present
- An agreed browser, operating system and screen reader matrix, with version details in the report
- Reproduction steps and recorded evidence for confirmed barriers, with the responsible component or vendor identified where possible
- Implementation guidance, saved response examples and a checklist separating automated checks from manual checks
- A dated record for one agreed standard and edition, naming the tested scope, product version and limitations
Fixed price (USD)
$1295
Save $54: $1349 if you book each part separately
- Implementation · additional
- $150 per hour, quoted as a fixed total after the test.
- Retest of agreed fixes · additional
- $199 per cycle. New journeys or changed response behaviour need a scope review. Automated regression checks are added where reliable, with manual steps for the remaining checks.
- The other two standards
- Documentation is $400 per additional standard, reusing applicable evidence. Any extra testing is scoped and quoted before work starts.
Good to know
Coverage agreed before testing
This review covers the agreed interface, journeys and sampled replies. It does not assess answer accuracy or model safety, or establish conformance for untested parts of your product. A product-wide ACR may need a broader assessment.
Third-party widgets
An embedded chat is tested in your site. Findings distinguish your configuration from vendor-owned behaviour where possible. Implementation and retesting are quoted separately; vendor fixes depend on the vendor.
Legal scope is counsel’s call
EAA requirements depend on the covered service and applicable exceptions. The AI Act’s high-risk obligations do not apply to every chatbot. Your legal adviser determines applicability; this review supplies technical evidence.
Show me the assistant
Share the assistant, the two tasks users need to complete, any known barriers, and any standard or deadline you need to meet. You get the scope and a fixed quote before work starts.