Review your AI’s answers to ensure claims match your sources and comply with key stakeholder requirements.
Verify Responses allows you to deploy AI confidently. The feature gives you full visibility into every claim made by your AI, the sources referenced, and compliance across legal, security, risk, PR, and executive perspectives.
The "Verify Responses" feature is an excellent way to build trust and resilience by offering six different perspectives on the same answer and by showing how the answer was constructed and composed.
Extract every factual claim and verify it against your source documents.
See verified claims scores so you know what holds up - and what needs review.
Identify unsupported claims immediately, without manual checking.
Analyze responses across End User, Security IT, Risk Compliance, Legal Compliance, Public Relations, and Executive Leadership.
Run verification after every query or only when you choose.
Turn each response into documentation you can reference in reviews and audits.
Filter conversations in the Customer Intelligence tab for insights
Spot low verified-claims scores fast across real conversations
Pinpoint the cause: Agent persona, answer retrieval, or knowledge gaps
Fix the root cause and improve your AI performance with every iteration
Give Legal, Compliance, and Security the evidence they need - without slowing teams down.
Catch security, legal, and reputational concerns early with stakeholder-based review.
Keep everyday Q&A fast, and run deeper verification only when the stakes are high.
"CustomGPT.ai has continually impressed us. Our own assessor loves it and frequently comments on how it provides the right answers. It's a very valuable tool that people need to understand and utilize."
Learn more about Verify Responses in our docs.
Verify Responses is a transparency feature that shows you exactly how your AI arrived at its answer. It extracts every factual claim from a response, checks each one against your source documents, and analyzes the response from six stakeholder perspectives. You get full visibility into what’s verified, what’s not, and where potential risks might exist.
If you need to trust your AI’s answers – whether for compliance, customer-facing use, or internal decisions – this feature gives you proof instead of assumptions. It helps you catch inaccuracies before your users do, identify gaps in your knowledge base, and provide audit trails for regulated industries.
Reading a response tells you what the AI said. Verify Responses tells you why it said it – which specific documents it pulled from, which claims are actually backed by your sources, and which ones aren’t. It’s the difference between trusting and verifying.
No. Verify Responses is a behind-the-scenes tool designed exclusively for you – the builder and administrator. Your end-users see a clean, standard chat interface. The feature exists to give you confidence that your chatbot is accurate and safe.
This feature is built for you, the builder and manager of the AI system. Think of it as your administrative dashboard for trust and safety. It provides the audit trail, fact-checking, and risk analysis you need to build, test, and manage a reliable AI assistant.
When you click the Claim button, the system reads the AI’s response, extracts every factual claim (like facts, dates, numbers, or policy details), and cross-references each one against your source documents. It then shows you exactly where it found supporting evidence – or flags claims that couldn’t be verified.
The Verified Claim is a simple calculation: verified claims divided by total claims found. If your AI makes 10 factual statements and 8 can be traced back to your source documents, your Verified Claim is 80%. This gives you a quick, quantifiable measure of how well-supported a response is.
The Trust Building button runs your response through a simulated review by six virtual stakeholders: End User, Security/IT, Risk Compliance, Legal Compliance, Public Relations, and Executive Leadership. Each perspective analyzes the response for potential concerns – like legally risky phrasing, security issues, or reputation risks – that you might not have considered.
The six perspectives are:
Claims without source support get flagged in the interface. This doesn’t necessarily mean the claim is wrong – it means it couldn’t be traced back to your uploaded documents. This is your signal to either add that information to your knowledge base or investigate further.
Yes, and that’s by design. A response can be factually accurate (high Verified Claim Score) but still raise concerns from a stakeholder perspective (lower Trust Score). For example, a statement might be true but use legally risky phrasing. The system highlights these conflicts so you can make informed decisions.
You’ll find the Verify Responses button next to AI responses in your builder interface. The shield icon opens the Verify Responses panel. You can also view verification data in the Customer Intelligence tab.
There are two modes:
Use persistent mode while building and testing your agent – you’ll get instant verification data on every response. When you deploy to production, switch to on-demand mode for spot-checking, to reduce cost. However, if you need full auditability for compliance (like for your CISO or CIO), keep it enabled in production. This also surfaces trust scores in your Customer Intelligence analytics.
No. Your standard chat response is generated at normal speed. The Verify Responses analysis is a separate process that runs either automatically (in persistent mode) or when you trigger it manually (on-demand). Your end-users won’t experience any delay.
Yes. You can run Verify Responses on-demand for any past conversation. Results appear in seconds.
Verification data flows into your Customer Intelligence dashboard. You can filter conversations by verified claim score, stakeholder status, or other criteria to quickly identify responses that need attention. This makes it easy to find patterns, spot problem areas, and prioritize improvements.
Go to Customer Intelligence in your dashboard and use the filters to sort by verified claim score. You can quickly identify which conversations scored below your threshold and need review.
Absolutely. This is one of the most valuable use cases. Filter for low verified claim scores to identify where your knowledge base has gaps. Each flagged claim points directly to content you may need to create or improve.
Verify Responses is available on Premium and Enterprise plans.
The Verify Responses features are powerful tools that consume significant resources. For Premium and Enterprise users, enabling them will apply a cost modifier to your queries. This will be clearly communicated in your settings before you proceed.
Verify Responses creates audit-ready documentation for every AI response. The Legal Compliance stakeholder flags liability risks, while verified claim scores provide verifiable source citations. For regulated industries where “the AI said so” isn’t a valid answer, this gives you the proof you need.
The Security/IT stakeholder analyzes every response for sensitive information handling and system security implications. You can prove to stakeholders that your AI isn’t exposing data or making things up.
Yes. Instead of guessing why your agent gave a weird response, you can see exactly which source documents it used and where it went wrong. This makes troubleshooting much faster.
Enable testing mode during development so verification runs on every response. Test your agent with real questions, check verified claim score, and fix issues before your users encounter them. Then switch to on-demand mode for production.
Yes, this is one of the most valuable discoveries you can make. When claims get flagged as unverified, it often means your knowledge base is incomplete. Low Verified claims scores point directly to content you need to create.
This depends on your use case and risk tolerance. For high-stakes applications (legal advice, medical information, compliance), you likely want scores above 90%. For general customer support, 80%+ may be acceptable. Use your scores to set internal benchmarks and improve over time.
A low score is actually valuable information. It means you found problems before your customers did. Use the detailed breakdown to identify whether the issue is your persona configuration, the retrieval system, or gaps in your documentation – then fix it.
A high Verified claims score means the claims in the response can be traced back to your source documents. It doesn’t guarantee the source documents themselves are correct or complete. The feature verifies alignment with your sources, not absolute truth.
This usually means the information exists but isn’t in your uploaded documents. The AI may have inferred something reasonable, but it couldn’t be verified against your sources. Consider adding that information to your knowledge base.
Yes. Analysis is performed securely within the CustomGPT.ai environment. Verification data is not shared with other users or external systems. Only the authenticated user can access explainability analysis for their queries.
Verify Responses operates within CustomGPT.ai’s SOC-2 Type II and GDPR-compliant infrastructure.
No. Only you (the authenticated builder/administrator) can access the Verify Responses analysis for your agents and conversations.
During development and testing, run it on everything (persistent mode). In production, the frequency depends on your needs. High-stakes applications may warrant continuous verification; others may only need periodic spot-checks.
Enable persistent mode while building
It depends on your requirements. If you need governance and full auditability for compliance teams (CISO, CIO), keep it enabled – this also surfaces trust scores in Customer Intelligence analytics. If you’re primarily doing spot-checks, on-demand mode reduces overhead while still giving you access when needed.
Verify claim-level accuracy and stakeholder risk for every answer-so teams can approve, trust, and scale AI with confidence.
