Trust & Compliance, Explained Plainly
What's built into My Chat Assistant and verified from source, what depends on how you configure and host it, and what we haven't independently verified yet. We don't claim a certification we don't hold, and we don't answer a hard question with a soft yes.
Built In and Verified From Source
True of the product itself, out of the box. These can be stated with full confidence.
Automatic PII Redaction
Personal identifiers, an email address, a phone number, a UK National Insurance number, a UK NHS number, a date of birth, a payment card number, and its CVV, are stripped from stored messages by default, before anything is written to the database. Card and CVV patterns are checksum-validated so ordinary numbers aren't caught by mistake.
Safeguarding Detection
A visitor's message is checked for safeguarding language before any answer is generated. A match triggers a fixed, operator-reviewed response and a handoff offer to a person, not a generated reply. The visitor is never cut off; they can keep talking.
Searchable, Exportable Records
Conversations can be searched and filtered by date, session, or keyword, and exported from the admin dashboard, supporting a records request.
Retrieval-Only by Default
In the product's default mode, retrieval-only, no generative AI provider connected, every answer comes directly from indexed site content or operator-authored Q&A pairs. Nothing is invented.
Air Gap by Default
No AI provider key is connected out of the box. Until you deliberately connect one of your own, any point where a question would otherwise go to an AI model hands you a composed prompt to copy and paste into a model of your choosing yourself. No network call is made to any AI model on your behalf unless you've connected one.
No Behavioral Profiling
My Chat Assistant does not build behavioral profiles of visitors or serve targeted advertising, of any visitor, of any age.
Non-Clinical Disclaimer Available
An operator can switch on a non-clinical disclaimer, making clear the assistant does not give medical advice, diagnosis, or treatment.
No Cookies
The widget sets no cookies. A live browser capture confirmed zero cookies before and after a full interaction, opening the widget, asking a question, receiving an answer. The only storage used is sessionStorage for a session id, which clears automatically when the browser tab closes.
Depends on How You Configure It
Built to support these outcomes. Whether a specific deployment achieves them depends on your own setup, hosting, and choices, not something the product resolves on your behalf.
UK GDPR
The product is built to support GDPR-conscious operation: retrieval-only by default, redaction of common identifiers, configurable retention, and exportable, erasable records. Whether a specific deployment is GDPR compliant depends on how you configure and use it, including your own legal basis for processing.
Caldicott Principles
The default, retrieval-only mode sends nothing to a third party and trains no model on visitor data. Whether a specific deployment meets the Caldicott Principles depends on your own configuration and any AI provider you choose to enable.
Privileged Communications
In the default, retrieval-only mode, nothing is sent to a third party. If you enable a generative AI provider, that provider's own data handling applies, a decision that's yours to make.
DSPT / Cyber Essentials
The product itself holds no DSPT or Cyber Essentials certification. Whether your hosting environment meets those standards depends entirely on where and how you host it.
Not Yet Independently Verified
An honest "not yet" beats a confident guess, especially here. These aren't claimed either way until they've actually been confirmed.
Accessibility (WCAG)
Automated accessibility testing, Lighthouse, built on the axe-core engine, has found and confirmed the fix for real issues on the live product, and currently passes clean. A full manual audit, including a screen reader pass, hasn't been completed. We don't claim WCAG 2.1 AA conformance; that specifically requires the full ruleset and a human assistive-technology pass, neither of which has been done yet.
Identity Verification (AML/KYC)
My Chat Assistant has no identity-verification or document-ingestion capability. If your use case needs this, it isn't something the product does.
Subject Access Request (DSAR) Lookup
Conversation lookup currently requires a visitor's session ID, not their name or email. Worth knowing before you commit to a specific DSAR process.
Uptime & High Availability
We don't claim a specific uptime or failover guarantee. That depends on your own hosting, not the product.
AI Disclosure Wording
The AI-generated-answer notice is currently fixed wording, not an operator-editable setting.
Questions We Get Asked Directly
The same questions this page answers, in the form a visitor is likely to type into the chat widget itself.
Q: Does the assistant remove personal data from what it stores?
A: Yes. Common identifiers, an email address, a phone number, a UK National Insurance number, a UK NHS number, a date of birth, a payment card number, and its CVV, are automatically redacted from stored messages by default, before anything is saved.
Q: Does the assistant detect if someone is in distress?
A: Yes. Messages are checked for safeguarding language before an answer is generated, and a match offers a handoff to a person rather than a generated reply.
Q: Can conversations be searched or exported for a records request?
A: Yes, by date, session, or keyword, with an export available from the admin dashboard.
Q: Do you redact NHS numbers?
A: Yes, an NHS number is automatically redacted from stored messages, alongside other personal identifiers.
Q: Can the assistant make things up?
A: In the default mode, no: answers are drawn directly from your own site content or your own Q&A entries, not generated freely.
Q: Does the assistant profile visitors?
A: No. It answers questions from indexed content; it does not build behavioral profiles or serve targeted advertising.
Q: Can this be used on a healthcare site without implying medical advice?
A: Yes, a non-clinical disclaimer can be switched on, stating plainly that the assistant does not give medical advice, diagnosis, or treatment.
Q: What happens if someone pastes their card details into the chat?
A: Card numbers and CVVs are automatically detected and redacted before storage.
Q: Does this widget use cookies?
A: No. It uses no cookies at all. It stores a session id in the browser's own session storage, which is not a cookie and clears automatically when the tab closes.
Q: Is this GDPR compliant?
A: The product is built to support GDPR-conscious operation, retrieval-only by default, redaction of common identifiers, configurable retention, exportable and erasable records, but GDPR compliance depends on how each operator configures and uses it, including their own legal basis for processing.
Q: Does this meet Caldicott Principles?
A: The product's default, retrieval-only mode sends nothing to a third party and trains no model on visitor data. Whether a specific deployment meets Caldicott Principles depends on the operator's own configuration and any AI provider they choose to enable.
Q: Is this safe for legally privileged conversations?
A: In the default, retrieval-only mode, nothing is sent to a third party. If an operator enables a generative AI provider, that provider's own data handling applies, which is the operator's decision to make.
Q: Is this DSPT/Cyber Essentials certified?
A: The product itself holds no such certification; that depends on the operator's own hosting environment and practices.
Q: Is the widget accessible / WCAG compliant?
A: Automated accessibility testing has been run against the live product, using Lighthouse and the axe-core engine, and it currently passes clean after two real issues it found were fixed. A full manual accessibility audit, including testing with an actual screen reader, hasn't been carried out, so we don't claim full WCAG 2.1 AA compliance at this time.
Q: Does this handle identity verification or KYC/AML checks?
A: No. The product has no identity-verification or document-ingestion capability.
Q: How is a subject access request (DSAR) handled?
A: Conversation lookup currently requires a visitor's session ID, not their name or email.
Q: What uptime or failover guarantee does this offer?
A: We don't claim a specific uptime or failover guarantee; that depends on your own hosting.
Q: Does the product use cookies or similar tracking technologies (PECR)?
A: No cookies or similar tracking technologies are used by the product itself.
Q: Can the AI-generated-answer notice be customized?
A: Not currently. The notice is fixed wording rather than an operator-editable setting.
Have a Compliance Questionnaire to Get Through?
Send it over and we'll answer it directly against what the product actually does, not a generic template. See our Privacy & Data Handling page for how conversation data is stored, retained, and controlled.
Send Us Your Questionnaire