Framework Principle

Source Discipline

Source discipline is the practice of checking AI-assisted work before treating it as fact, evidence, authority, or final guidance.

A fluent answer is not proof. Serious work requires knowing which claims are being made, what evidence supports them, what remains uncertain, and what should be verified before the output is used.

Core Idea

AI can help with research, but it is not the source itself.

Artificial intelligence can help users explore topics, summarize provided material, classify claims, identify possible source categories, and prepare questions for further review.

That support can be valuable. It can make unfamiliar subjects easier to approach, help organize complex material, and reveal what needs closer attention. But research support is not research replacement.

An AI-generated response may contain a mixture of facts, assumptions, interpretations, recommendations, outdated information, and unsupported claims. Source discipline separates those categories before the output becomes part of real work.

The goal is not distrust for its own sake. The goal is responsible trust: confidence earned through evidence, review, and appropriate limits.

Why It Matters

Polished language can hide weak evidence.

AI output can feel finished before the evidence is ready.

Fluency can mislead.

A response can be clear, confident, and well structured while still being incomplete, outdated, unsupported, or wrong.

Precision can create false confidence.

Dates, names, citations, statistics, titles, version numbers, and quotations require careful checking because they can look authoritative even when they are incorrect.

Currency matters.

Laws, policies, prices, software versions, product features, public roles, medical guidance, and market conditions can change.

Sources have scope.

A source may support one claim without supporting every statement near it. Related material is not automatically evidence.

Exploration can harden into authority.

Draft language, brainstorming, and first-pass summaries should not quietly become final claims without review.

Responsibility remains human.

The person using the output remains responsible for checking what they publish, submit, send, implement, or rely upon.

Claim Classification

Not every sentence needs the same kind of support.

Source discipline begins by identifying what kind of statement is being made.

Some statements are personal experience. Some are interpretation. Some are recommendations. Some are factual claims. Some are current claims that may have changed. Some are technical, legal, medical, financial, academic, or professional claims that require a higher standard of review.

AI often blends these categories into smooth paragraphs. The reader may not see where experience ends, where interpretation begins, or where a factual claim needs evidence. A source discipline pass restores those distinctions.

The higher the consequence of being wrong, the stronger the evidence standard should be.

View Practice Tools

Classify the statement as:

Personal experience
Interpretation
Assumption
Recommendation
Factual claim
Current claim
Expert-review claim

Evidence Standards

Match the source to the claim.

A useful source discipline practice asks what kind of evidence would actually support the statement being made.

A personal reflection does not require the same support as a legal claim. A historical statement does not require the same source as a software command. A product recommendation does not require the same review as a medical or financial decision.

The source should fit the claim, the risk, the audience, and the intended use.

Examples of source matching

Official documentation for software behavior
Primary sources for laws, policies, and institutional requirements
Current reputable sources for changing facts
Scholarly or professional sources for academic claims
Qualified experts for high-stakes professional judgment
Clearly marked personal experience for reflective claims

Research Support

Use AI to map the terrain, not to replace the evidence.

AI can be useful at the beginning of research because it can help organize the field of inquiry.

It can suggest terms to investigate, identify likely source categories, prepare questions, summarize material the user provides, and help distinguish what appears known from what requires verification.

But the generated response should not become the source of record. The user still needs to consult appropriate materials, check current information, read sources in context, and verify whether the evidence supports the final claim.

AI can help prepare the working surface. It should not be treated as the foundation unless the foundation has been independently checked.

Read About After the Prompt

AI can help by:

Listing terms to research
Identifying possible source categories
Preparing questions for experts
Summarizing provided source material
Extracting claims from a draft
Flagging statements that need verification

Evidence Pass

A practical review before use

An Evidence Pass helps prevent AI-assisted research support from turning into unsupported authority.

Evidence Pass Checklist

Extract the claims.

Identify factual statements, current claims, interpretations, assumptions, and recommendations.

Classify the risk.

Ask what could happen if the claim is wrong, incomplete, outdated, or misleading.

Identify source needs.

Determine what kind of source would actually support the claim.

Check currency.

Mark claims that may have changed since the source, model response, or prior draft was created.

Separate experience from evidence.

Personal experience can be valuable, but it should not be presented as universal proof.

Revise or remove.

Do not keep claims that cannot meet the evidence standard required by the intended use.

Final question: Does the support behind this claim match the seriousness of using it?

Citations and Attribution

A citation is not decoration.

A citation should help the reader understand where a claim came from and whether the cited source actually supports it.

AI-assisted writing can create citation risk when the tool invents sources, misstates source details, quotes inaccurately, or attaches a source to a claim the source does not support. Source discipline requires checking the relationship between claim and citation, not merely adding references at the end.

This matters in academic work, publication, business communication, public claims, technical documentation, and professional settings where readers rely on source integrity.

The more public or consequential the work, the more carefully citations should be checked.

Citation review asks:

Does the source exist?
Is the source reliable for this claim?
Does the source support the exact statement?
Is the quote accurate?
Is the citation current enough?
Is the source being used fairly in context?

High-Stakes Work

Some claims require more than a quick check.

When AI-assisted output touches legal, medical, financial, tax, academic, engineering, compliance, employment, safety, or other high-stakes areas, ordinary confidence is not enough.

The appropriate response may be to consult official sources, seek qualified professional advice, verify current rules, test technical instructions, preserve records, or avoid using the output until it can be reviewed properly.

Source discipline is not meant to slow down every ordinary use of AI. It is meant to scale the review standard to the level of risk.

Escalate review when the output affects:

Legal rights or obligations
Medical or health decisions
Financial, tax, or investment choices
Academic submission or publication
Technical systems or security
People who may rely on the result

Human Authority

The user remains responsible for what is kept.

Source discipline is one expression of human authority.

AI can help identify claims, suggest source categories, flag uncertainty, and prepare review checklists. It cannot take responsibility for whether the claim should remain in the final work.

The human user must decide whether a statement should be retained, rewritten, qualified, sourced, escalated, or removed. That decision becomes especially important when the work will be published, submitted, implemented, cited, or relied upon by others.

The final standard is not whether the AI sounded right. The final standard is whether the person using the output can responsibly stand behind it.

Explore Human Authority

Human decisions include:

Keep
Revise
Qualify
Verify
Escalate
Remove

In the Series