AI Source Verification Guide for Factual Answers

Choose whether an AI answer should rely on uploaded files, cited public sources, or academic-style evidence review before you use it.

Purpose

This guide helps you verify the source basis of factual AI answers before using them. It separates three workflows: answers based only on provided files, answers supported by cited authoritative sources, and academic-style evidence review. It also shows where to configure the selected source rule in AI tools so the rule is reusable and not left as a one-off chat instruction.

When to use this

Use this section to decide whether this workflow is the right fit before you configure prompts, policies, or reference material.

Use case
You need to verify a factual AI answer before using it
Use this when an answer must not rely on memory, assumptions, unstated prior context, or unsupported general knowledge.
Use case
You need to choose between provided files and public sources
Use this before deciding whether the answer should come only from uploaded files, logs, screenshots, repository snapshots, or pasted excerpts, or whether authoritative public sources are allowed.
Use case
The output may be published, reused, or used for a decision
Use this when unsupported claims, weak citations, or invented evidence would create review, publication, product, legal, academic, or engineering risk.
Use case
You are setting up a reusable AI source-verification workflow
Use this when the source rule should be placed in project instructions, assistant instructions, a Gem, a Skill, or an API system/developer layer instead of being pasted only into one chat.

Decision guide

Choose which sources the AI answer may use

Select one option before opening the matching guide: provided files only, cited authoritative sources, or academic-style evidence review.

Option 1 · Provided files only
Answer only from provided files
Use this when every factual claim must be supported by uploaded files, logs, screenshots, excerpts, repository snapshots, tables, or other material supplied in the current task.
Evidence boundary: Allowed sources: provided material only; no public web sources, no memory, no prior knowledge, no inferred project state.
Best for: CI logs, uploaded PDFs, screenshots, repository ZIPs, local documents, pasted excerpts, private data, and file-based analysis.
Use when: The question is about material supplied in the current task.
Open guide
Option 2 · Cited authoritative sources
Answer with cited authoritative sources
Use this when the answer depends on public facts and important factual claims must be supported by cited authoritative sources such as official documentation, standards, papers, or recognized institutions.
Evidence boundary: Allowed sources: authoritative public sources with stable locators, plus any user-provided material that is explicitly allowed.
Best for: Current or niche facts, technical product behavior, standards, framework documentation, public policy, scientific claims, and source-backed explanations.
Use when: The task asks what public sources, official documentation, standards, or scholarly material support.
Do not use when: The task is only about private files or local material and external sources are not allowed.
Open guide
Option 3 · Academic evidence review
Run academic-style evidence review
Use this when the answer needs formal register, authoritative citations, VERIFIED / NOT VERIFIED separation, missing-evidence handling, and confidence reporting.
Evidence boundary: Allowed sources: authoritative public sources and explicitly supplied material, with academic evidence labels and confidence reporting.
Best for: Research-style answers, scholarly claims, claim assessments, technical explanations, source-backed article review, and publication-oriented evidence checks.
Use when: The output must be formal, source-backed, explicit about uncertainty, and structured for review.
Open guide

Workflow assets

Required workflow assets

Open the prompts, policies, and reference pages needed to run this workflow correctly.

Required prompt
Provided-material-only prompt
Restricts factual answers to files, logs, screenshots, excerpts, or other material supplied in the current task.
Type: Instruction-layer prompt
Belongs in: Instruction layer
Use when: The selected option is provided material only.
Open prompt
Required prompt
Authoritative-sources prompt
Requires cited authoritative sources for important factual claims.
Type: Instruction-layer prompt
Belongs in: Instruction layer
Use when: The selected option is cited authoritative sources.
Open prompt
Required prompt
Evidence-Gated Academic Mode prompt
Adds academic register, evidence labels, citation discipline, fail-closed behavior, and confidence reporting.
Type: Instruction-layer evidence mode
Belongs in: Instruction layer
Use when: The selected option is academic-style evidence review.
Open prompt
Optional prompt
Instruction hierarchy and source rules
Keeps instruction priority, untrusted content, and the active source policy separated.
Type: Source-rule component
Belongs in: Instruction layer
Use when: The workflow combines user input, uploaded files, retrieved content, or multiple prompt components.
Open prompt
Optional prompt
No simulation / no fabrication
Blocks invented facts, fabricated sources, simulated scans, guessed results, and unsupported system behavior.
Type: Guardrail component
Belongs in: Instruction or runtime prompt layer
Use when: The task has risk of unsupported reconstruction or fabricated evidence.
Open prompt
Optional prompt
Confidence score
Adds evidence-based confidence reporting after source coverage and uncertainty are checked.
Type: Verification component
Belongs in: Verification layer
Use when: The output needs explicit confidence reporting.
Open prompt
Required policy
Use only the files you provide for factual answers
Defines when factual answers must stay inside provided files or material.
Type: Files-only source policy
Controls: Provided-material source rule, missing-source handling, and no-external-source behavior.
Read policy
Required policy
Use cited public sources
Defines when public factual claims need cited authoritative sources.
Type: Authoritative-source policy
Controls: Source authority, stable locators, citation requirements, and fail-closed behavior.
Read policy
Required policy
Academic-style output with cited sources and confidence
Defines formal register, evidence labels, source-backed claims, and confidence reporting for academic-style output.
Type: Academic evidence policy
Controls: Academic register, claim taxonomy, evidence labels, citation rules, missing evidence, and confidence score.
Read policy

Implementation procedure

Step-by-step implementation procedure

Follow the workflow in order. Each step gives one action and one verification check before continuing.

  1. Runtime prompt layer

    Identify what the answer must prove

    Decide whether the question is about supplied files, public facts, or academic-style evidence review.

    Action
    Classify the task before opening a policy or prompt. Do not choose a prompt based only on the filename.
    Verify
    The selected option matches the type of source support the answer requires.

  2. Instruction layer

    Choose which sources the answer may use

    Choose whether the answer may use provided files only, cited authoritative sources, or academic-style evidence review.

    Action
    Use the decision cards to open the matching procedure and policy.
    Verify
    The workflow does not mix incompatible source rules.

  3. Reference layer

    Provide the source material

    Supply files for files-only work, provide or permit authoritative sources for public-source work, and provide research/source material for academic evidence review.

    Action
    Upload, paste, or configure the allowed source material in the relevant project, knowledge, file, or retrieval surface.
    Verify
    The assistant can point to the allowed source material before answering.

  4. Instruction layer

    Configure the rule where it will persist

    Place stable source rules in the instruction layer and source material in the reference layer.

    Action
    Use the platform mapping for ChatGPT, Claude, Gemini, or API/internal systems.
    Verify
    The rule is not left only inside a one-off user prompt when it must be reusable.

  5. Verification layer

    Test the missing-source behavior

    Ask a factual question without providing the required source and confirm that the assistant does not answer from unsupported memory or assumptions.

    Action
    The expected behavior is a fail-closed response requesting the missing file, citation, or source.
    Verify
    The workflow fails closed before real use.

Verification checklist

Use this checklist before accepting the output, publishing it, or using it as evidence for a downstream workflow.

Allowed sources
The allowed sources are explicit
The task states whether the assistant may use provided files only, cited authoritative sources, or academic-style evidence review.
Placement
Reusable rules are not left in one chat
Reusable source rules are placed in Project/GPT/Gem/Skill/API instructions when the tool supports that layer.
Sources
Source material is available before generation
Files, policies, papers, standards, logs, screenshots, and source excerpts are configured as files, knowledge, retrieval, or context material.
Fail closed
Unsupported claims stop the workflow
If the selected source option cannot support an important factual claim, the assistant must identify the missing support instead of completing the answer.
Citations
Public claims have stable source locators
When public sources are allowed, citations must point to stable source material such as official documentation, standard sections, papers, or other authoritative references.

Next step