checklist

AI chatbot disclosure notice template for small teams

A practical disclosure notice template for customer-facing AI chatbots, covering placement, customer wording, data use, transcript notices, human support paths, consent routing, and review records.

Use this template before a customer-facing AI chatbot appears on a website, app, help center, customer portal, onboarding flow, checkout flow, support page, or account screen.

The notice does not make the chatbot safe by itself. It sets customer expectations, explains what the bot can and cannot do, gives a human support path, and points sensitive requests to the right workflow. Before publishing a notice, run the AI Tool Risk Checker and attach the result to the chatbot launch packet.

Bottom line

Every customer-facing AI chatbot needs plain customer-facing wording that answers:

1. Is this AI-assisted or human support?
2. What can the chatbot help with?
3. What should customers avoid entering?
4. Can the chatbot make decisions or only provide guidance?
5. How can customers reach a human?
6. Are chats retained, reviewed, or used to improve service?
7. Where should privacy, deletion, export, security, billing, legal, or account access requests go?

Use the Small Team AI Security Checklist for baseline approval, owner assignment, incident response, and access controls. This page focuses on customer-visible chatbot notices.

When to use this template

Scenario	Use this template?	Why
Website support chatbot answers product questions	Yes	Customers should know they are interacting with automated assistance.
Chatbot collects support context before creating a ticket	Yes	Customers should know what data they are providing and how to reach support.
Bot handles billing, cancellation, refund, or account questions	Escalate	Notice must route sensitive or high-impact requests to human owners.
Bot can use account data or ticket history	Yes	Customers need clear context about account-specific assistance.
Bot summarizes conversations into tickets or CRM	Yes	Transcript and summary handling should be disclosed in plain language.
Bot can call tools or trigger actions	Escalate	Customer-visible action and confirmation wording need owner approval.
Bot handles privacy, security, deletion, export, or AI training questions	Escalate	Use approved trust/privacy/security wording.
Internal employee-only support bot	Maybe	Use an internal notice if employees may paste customer or regulated data.
Static FAQ page with no chatbot	No	Use normal website and privacy notices.

If the customer can rely on the answer, the notice should not make the bot sound more capable than it is.

Disclosure placement matrix

Location	Required notice
Chat launcher	Short label that the assistant is AI-assisted.
First bot message	Plain disclosure, scope, sensitive data warning, and human path.
Input placeholder	Short warning not to enter passwords, payment data, or sensitive data.
Before customer-visible action	Confirmation that explains what will happen and whether a human reviews it.
Before account-specific lookup	Authentication and account data context.
Before transcript or summary is saved	Transcript, summary, or ticket record notice.
Help center page	Link to support policy, privacy page, and human contact option.
Error or uncertainty state	Clear handoff to human support.
Privacy/security/trust question	Route to approved owner or page.

The first notice should be visible before the customer shares sensitive details, not buried after the conversation.

Customer notice template

Use this as starter copy and edit it for your product.

Part	Copy block
Bot identity	”You are chatting with an AI-assisted support bot.”
Scope	”It can help with product questions, basic troubleshooting, and support intake.”
Limits	”It may be wrong or incomplete, and it cannot make legal, billing, security, privacy, or account-access decisions.”
Sensitive data warning	”Do not enter passwords, payment details, private keys, recovery codes, or sensitive personal information.”
Human path	”You can ask for a human at any time.”
Records	”We may save this chat or a summary to help answer your request and improve support quality.”
Correction path	”If an answer looks wrong, tell us and a human will review it.”
Privacy route	”For deletion, export, privacy, or AI data-use questions, contact [approved support/privacy path].”

Short version for the first message:

You are chatting with an AI-assisted support bot. It can help with product questions and support intake, but it may be wrong or incomplete. Do not enter passwords, payment details, private keys, recovery codes, or sensitive personal information. You can ask for a human at any time.

Do not say “always accurate,” “approved by our legal team,” “secure by default,” “private by default,” or “not stored” unless the responsible owner has evidence for that exact claim.

Data use disclosure checklist

Question	Notice or routing requirement
Are raw transcripts saved?	Say that chats may be saved, or route to the retention notice.
Are summaries saved into tickets or CRM?	Tell customers a summary or support record may be created.
Are chats reviewed by humans?	Say support or quality reviewers may review chats if true.
Are chats used for product improvement?	Use vendor-reviewed wording and avoid vague claims.
Are chats used to train models?	Use exact approved wording; do not guess.
Can customers request deletion or export?	Route to privacy/support process.
Can customers opt out of AI chat?	Provide human path or alternative support channel.
Can the bot access account data?	Explain authenticated/account-specific context.
Can vendor support staff see chats?	Review vendor terms and support access before answering.

The notice should match actual settings. A small team should not promise more privacy, deletion, or training restrictions than it can verify.

Human support path

Customer signal	Required path
”I want a human”	Hand off immediately.
Bot is uncertain or source is missing	Offer human support.
Customer disputes the answer	Create a human review path.
Billing, refund, cancellation, renewal, or pricing exception	Route to billing/account owner.
Security, privacy, deletion, export, AI training, or trust question	Route to approved trust/privacy/security owner.
Account access, admin transfer, permission change, or identity issue	Route to authenticated workflow.
Legal, regulated, health, finance, HR, children, or government topic	Refuse or route to owner-approved human path.
Angry customer, churn threat, complaint, or public escalation	Route to support lead or customer success.

Use the AI chatbot human handoff policy template for the handoff workflow behind the notice.

Sensitive topic notice rules

Topic	Default wording rule
Security questionnaire or compliance claim	”A human trust/security owner will review this.”
Privacy, deletion, export, or AI training	”We route privacy and data-use requests to the approved support/privacy process.”
Billing, refunds, credits, or cancellation	”A billing/account owner must review this request.”
Account access, admin transfer, or permissions	”Use the authenticated account support path.”
Legal, regulated, health, finance, HR, children, or government data	”The bot cannot provide this guidance; a qualified owner must review.”
Passwords, payment data, recovery codes, private keys, or secrets	”Please do not enter this information here.”
Incident, outage, vulnerability, or data exposure claim	”We will route this to the security/incident owner.”

Do not let the bot improvise notices for sensitive topics. Use approved scripts.

Consent and opt-out routing

This section is operational guidance, not legal advice.

Situation	Route
Customer wants human support instead of AI	Provide human path without requiring an argument.
Customer asks whether use of chatbot is optional	Explain available support alternatives.
Customer asks to delete or export chat	Route to privacy/support process.
Customer asks not to use their chat for training	Use exact vendor-reviewed wording and process.
Customer asks who reviewed the chat	Use approved access summary.
Customer refuses transcript retention	Route to retention/privacy owner.
Customer provides sensitive data after warning	Redact/delete or escalate according to retention policy.

If your jurisdiction, industry, contract, or customer terms require specific consent language, get owner review before publishing.

Transcript notice

Use this when chats or summaries are stored.

Use case	Copy block
Support record	”We may save this conversation or a summary as part of your support request.”
Quality review	”Support owners may review chats to improve answer quality and routing.”
Human handoff	”If we hand this to a human, the conversation context may be attached to the support ticket.”
Retention	”Chat records are handled according to our support data retention process.”
Sensitive data	”Please do not include passwords, payment details, private keys, recovery codes, or sensitive personal information.”
Correction	”If an AI answer is wrong, a human can review and correct it.”

Use the AI chatbot conversation log retention policy template for the retention workflow behind these statements.

Vendor wording questions

Ask these before approving the notice.

Question	Why it matters
Does the vendor store raw transcripts, summaries, feedback, source traces, or files?	Determines what the notice should say.
Can the vendor use chats for training or product improvement?	Affects customer-safe data-use wording.
Can admins disable training or product-improvement use?	Affects opt-out and approval wording.
Can vendor staff access conversations for support, abuse, or quality review?	Affects human review and access claims.
Can transcripts be deleted by customer, date, conversation, or workspace?	Affects deletion request routing.
Are source traces or tool calls retained separately from transcripts?	Hidden records may outlive the visible chat.
Are regions, subprocessors, or retention settings configurable?	May affect customer commitments.
Can the bot show a custom notice before the first message?	Needed for practical placement.

Do not publish vendor-specific promises until the tool owner has verified the settings.

Approval record

Copy this into the chatbot launch packet.

Field	Entry
Bot name and location	Website, app, help center, portal, or support page.
Notice version	Date and exact approved wording.
Placement	Launcher, first message, input placeholder, account lookup, action confirmation, privacy route, or help page.
Bot scope	Topics the bot can answer and topics it must route.
Sensitive data warning	Approved warning text.
Human path	How customers reach a human and which queue owns it.
Transcript statement	What is saved, summarized, reviewed, or retained.
AI data-use statement	Training/product-improvement wording and owner approval.
Privacy route	Deletion, export, correction, opt-out, and privacy contact path.
Vendor evidence	Settings, terms, retention, support access, and training controls reviewed.
Decision	Approve, approve with limits, hold, deny, or escalate.
Next review	Date and trigger conditions.

Keep the record with the chatbot launch approval, not only in a copy document.

Monitoring checklist

Review these during pilot and monthly after launch.

Signal	Action
Customers keep asking if the bot is human	Make disclosure more visible.
Customers paste sensitive data	Improve input warning and route cleanup.
Customers ask for human support repeatedly	Review handoff path and first message.
Bot answers privacy/security/training questions incorrectly	Route those topics to approved owner.
Notice says chats are not stored but logs exist	Stop and correct wording immediately.
Vendor settings change	Reapprove notice and data-use wording.
Customers complain about misleading automation	Review placement, wording, and human alternative.
Transcript retention changes	Update notice and approval record.
Tool actions are added	Add action confirmation notice.

Notices should be maintained like product controls, not one-time marketing copy.

Metrics to track

Metric	Why it matters
Human handoff requests	Shows whether notice and scope are clear.
Sensitive data warnings triggered	Shows whether customers need stronger guidance.
Privacy/data-use questions	Shows trust friction.
Deletion/export requests tied to chatbot	Shows operational workload.
Incorrect answers about AI use or data	Shows high-risk notice failure.
Customer complaints about chatbot identity	Shows disclosure visibility problem.
Notice version changes	Shows governance history.
Vendor setting changes	Shows drift risk.
Bot actions requiring confirmation	Shows where notice scope expanded.
Customer corrections after bot notice	Shows downstream trust impact.

Track notice failures separately from answer-quality failures. A bot can answer correctly and still mislead customers about who or what they are interacting with.

Evidence checked

This template is aligned with:

1. NIST AI RMF Core, which emphasizes transparency, accountability, documentation, human oversight, feedback, incident communication, and continuous AI risk management.
2. NIST AI Risk Management Framework, which frames AI risk management across design, deployment, use, evaluation, and operation.
3. NIST Generative AI Profile, which identifies generative AI risks and risk management actions for organizations deploying generative AI systems.
4. FTC artificial intelligence guidance, which tracks FTC business guidance, policy statements, and enforcement activity related to AI accuracy, privacy, confidentiality, and consumer protection.
5. FTC AI privacy and confidentiality guidance, which warns AI companies to honor privacy and confidentiality commitments, including commitments about customer data use.
6. Cybergiz templates for chatbot launch review, knowledge base review, human handoff, answer correction, conversation log retention, prompt injection response, tool action approval, customer impact assessment, and incident response.

This page is practical operating guidance, not legal, procurement, privacy, compliance, audit, certification, or security assurance advice.

FAQ

Do we need to tell customers the chatbot is AI-assisted?

For practical trust and support operations, yes. Customers should understand when they are dealing with automation and how to reach a human. Specific legal requirements depend on jurisdiction, industry, contract, and use case.

Where should the disclosure appear?

Put a short label on the launcher and a clear notice in the first bot message. Add extra notices before account lookups, transcript storage, customer-visible actions, and sensitive-topic routing.

Can the notice say chats are private?

Only if the owner can prove exactly what that means. If support staff, vendor staff, logs, summaries, or quality reviewers can access chats, use more precise wording.

Should the bot answer questions about training or product improvement?

Only with approved wording. Training, product improvement, human review, transcript retention, and deletion are different controls. Do not let the bot guess.

What should the input warning say?

Tell customers not to enter passwords, payment details, private keys, recovery codes, or sensitive personal information. Route sensitive workflows to approved support paths.

What if the bot can take actions?

Add confirmation wording before the action. Explain what will happen, whether a human reviews it, and how the customer can correct or cancel if needed.

Who owns chatbot notice approval?

The chatbot business owner owns the launch decision, but privacy, security, support, product, trust, legal, and vendor owners may need to approve specific wording.