Transparent AI

Source-cited AI RFP answers reviewers can trust

Every AI-generated answer in RFP.ai includes citations to the exact documents it came from. No black box. No guessing. Complete transparency for compliance and trust.

Last updated May 2026

Quick Answer

RFP.ai enforces cited answers so reviewers can inspect the source behind every buyer-facing response before approving it. That makes AI drafting faster to trust in procurement, security, and compliance-heavy workflows.

Source citations make the answer easier to verify before it leaves your workspace.
Confidence thresholds reduce the risk of AI guessing when evidence is weak.
The cited-answer model is especially useful for security questionnaires, DDQs, and regulated responses.
This page is part of the public proof path buyers can inspect before procurement friction starts.

Key Facts

Every AI-generated answer links directly to source documents in your knowledge base
Confidence scores (0-100) help you prioritize review time
When confidence is low, the system flags it for expert review instead of guessing
Built on retrieval-augmented generation (RAG) to ground answers in verified content
Supports ISO 27001-style traceability and review workflows for AI-assisted outputs

How It Works

1. Full-Text Search

When you ask a question, RFP.ai searches your content library using FTS5 (full-text search). We find the most relevant passages from your uploaded documents, past RFPs, and approved answers.

2. Source Grounding (RAG)

The AI receives only the retrieved passages—not general internet knowledge. We explicitly instruct: "Use only the provided sources. Cite specific documents." This is called retrieval-augmented generation (RAG).

3. Confidence Threshold

Every answer receives a confidence score (0-100) based on source relevance and completeness. If confidence is below 35%, we display "Needs SME Review" instead of generating an answer. The AI says "I don't know" rather than hallucinate.

4. Citation Display

The final answer includes clickable citations. Each citation shows the document name, page number, and relevant passage. You can verify everything before sending to your client.

Why It Matters

Compliance & Audit Trails

Traceable, cited answers support ISO 27001-style review and evidence workflows for AI-assisted content. They create an audit trail showing what was generated, which sources were used, and who approved the final answer.

Trust & Verification

Your reputation is on the line. With citations, you can verify every claim before it reaches your client. No black box AI—just transparent, verifiable answers.

Faster Review

Confidence scores help you prioritize. High-confidence answers need minimal review. Low-confidence answers get flagged for expert input. Focus your time where it matters.

Prevent Hallucinations

The 35% confidence threshold prevents the AI from guessing. If your content doesn't answer the question, we tell you—we don't make something up.

Technical Implementation

Retrieval-Augmented Generation (RAG)

RAG combines search (retrieval) with AI generation. Unlike standard LLMs that rely on training data, RAG dynamically fetches relevant content from your knowledge base and uses it as context for generation.

AI Provider: Mistral AI (EU)

We use Mistral AI (mistral-large-latest), an EU-based provider. Your data is never used to train AI models, and processing is governed by the contractual safeguards described in our privacy and sub-processor documentation.

Confidence Calculation

Confidence is based on search relevance scores, source diversity, and passage completeness. A score below 0.35 (35%) indicates insufficient information in your content library to answer the question reliably.

Frequently Asked Questions

RFP.ai uses retrieval-augmented generation (RAG). First, we search your content library using full-text search to find relevant passages. Then, the AI generates an answer using only those passages—never from general knowledge or the internet. Every answer includes clickable citations showing exactly which documents were used.

The confidence score (0-100) indicates how well your existing content answers the question. High confidence (>70) means strong, relevant sources. Medium (35-70) suggests review recommended. Low (<35) triggers 'Needs SME Review' instead of generating an answer. This prevents hallucinations.

Yes. Click any citation to view the source document, with the relevant passage highlighted. You can review, edit, or replace any AI-generated content. Nothing goes to your client without your approval.

Our three-layer approach minimizes hallucinations: (1) Only search your documents, not the internet. (2) Instruct the AI to use only provided sources. (3) Apply a 35% confidence threshold—below this, we say 'I don't know' rather than guess. However, as with any AI system, we recommend human review of all outputs.

This approach supports ISO 27001-style control objectives around traceability, review, and evidence for AI-assisted content. It does not mean RFP.ai is independently ISO 27001 certified today. Teams still need to apply their own approval workflows and compliance reviews.