What You Need to Know
Validation means checking an output's claims against something that can confirm them. This objective gives the review from objective 2.1 its tools. It puts them in order from the cheapest to the most costly. Your own judgement decides how much checking a given output deserves. A primary document is the original record: the official publication, the regulator's text, the organisation's records, the published criteria of a funder (an organisation that gives grants).
This lesson uses one made-up example throughout: an accounting firm. The office staff have Claude draft a quote letter for a new client from the attached meeting notes and price list. The letter states a total fee and the tax authority's filing deadline. Claude also drafts internal newsletter ideas.
You should not treat Claude as a single source of truth, and high-stakes output deserves careful scrutiny. Claude can produce fluent, specific output that looks correct but is badly mistaken. Specificity alone is not evidence of accuracy: an exact fee in the quote letter can still be wrong.
Cheapest First: The Source You Gave It
When Claude builds an output from attached material, the first check is the attachment. The attachment is already in the conversation. Each figure, section and claim matches the attachment or does not. This check is the habit from objective 2.1, used as a technique, and it costs minutes. The hours and rates behind the total are checked against the meeting notes and price list. Use a check that takes more effort only when the given source cannot settle the question.
Independent, Authoritative, Primary
External claims need external checking. The attachments cannot confirm the filing deadline, which did not come from them. Web search responses include citations "so you can easily verify sources yourself". Users are advised to cross-reference cited sources and use authoritative ones for critical decisions. The Help Center's hallucination article says to review Claude's cited sources, because checking the original catches misinterpretation. Citations exist to be checked.
The standard is the primary document. A second AI summary, a forum post, or Claude's own earlier output is not independent validation. A check that shares the errors of the thing that it checks proves nothing. For the filing deadline, the primary document is the tax authority's published page. Web search from within claude.ai is a legitimate way to find that page. The source validates the claim, never the search.
Make the Output Traceable
A traceable output shows where each claim came from, so it is checkable. Do not assume that an output is correct when you cannot trace its claims. For long documents, Anthropic's prompting guidance says to ask Claude to quote the relevant parts of the source before it does its task. For research work, the guidance says to ask Claude to verify information across multiple sources. On the quote letter, ask Claude to show where each claim came from. Then check the important claims at those places yourself. When a decision depends on a single figure taken from a source, confirm the figure in two independent places before anyone sends it. A figure that Claude calculated needs the check under Figures Claude Calculated, below.
What a Self-Check Can and Cannot Do
A self-check is a request that Claude check its own work. It has three forms. You ask Claude to check its answer. You send the same request again and compare the two outputs. Or you send a follow-up that asks Claude to verify its earlier statements. Objective 2.2 describes the last two forms. Anthropic's prompting guidance says a self-check catches errors, especially in coding and maths.
| A self-check can | A self-check cannot |
|---|---|
| find an arithmetic slip | confirm that a fact is true |
| find two parts of the output that contradict each other | replace the check against the source |
| show a figure that is different in two runs | approve an output for sending |
A self-check cannot confirm a fact, because no new information enters the check. Claude uses the same knowledge and the same reading of the source that produced the output. So it can miss an error a second time and still say that the output is correct. Use what a self-check finds as a list of items to check against the source.
At the accounting firm, the office staff ask Claude to check the quote letter. Claude finds that it added the payroll fee twice, which is a real error. Claude also says that the rest of the letter is correct. That statement does not show that the filing deadline is right, so the staff still check it on the tax authority's page.
Figures Claude Calculated
Some figures in an output are in no source, because Claude calculated them. Examples are a total, a percentage, an average and a change from last year. The check against the source cannot find an error in such a figure, because the source does not contain it. So treat a calculated figure as a separate claim. First, check its input numbers against the source. Then do the calculation again yourself from those numbers. Correct input numbers do not prove that the result is correct. A self-check can find a slip in the calculation, but it does not replace your own calculation.
At the accounting firm, the total comes from the hours in the meeting notes and the rates in the price list. The staff check the hours and the rates against those two documents. Then they do the multiplication again on a calculator and compare the answer with the total in the letter.
Proportionality: The Judgement Layer
Proportionality means that the amount of checking depends on what an error would cost. The client decides on the total, so the total gets both checks: its hours and rates against the source, and the calculation done again. The newsletter ideas get a read-through. Too little checking of the consequential claim is the obvious failure. Heavy checking on low-stakes work is also wrong, not safe. Time spent on triple checks of newsletter wording that nobody acts on is time not spent on checking the total.
In the exam
Questions test the techniques. They also test when each technique is proportionate, in both directions. The most common trap is a self-check that looks like verification. An option has the user ask Claude to confirm, re-check or rate its own work, and calls the confident answer validation. The guide's sample rationale says why it fails: self-reported confidence is not a reliable accuracy signal.
Key Concept
When a validation question offers several checking plans, test each plan against two questions. Does independent information enter the check? Does the depth of checking match what depends on each claim? The right answer passes both. The wrong options usually fail one. Self-check and circular sources (sources that share the errors of what they check) fail independence. Checking every claim to the same maximum depth fails proportionality.
Exam Traps
Asking Claude to confirm or re-check its claims and calling that verification
No independent information enters the check, so Claude can miss the same error again. A self-check can find slips and contradictions for you to check. Claims that matter are still checked against the source.
Checking the numbers behind a calculated figure and accepting the result
Correct input numbers do not prove that the calculation is correct. A total or percentage that Claude worked out is a separate claim, so do the calculation again yourself from the checked numbers.
Checking a claim against another AI summary or Claude's own earlier output
A check against a source made from the thing being checked shares its errors. Check against the primary document instead: the official publication, the published criteria, the organisation's records.
Verifying every sentence to the same maximal depth regardless of stakes
The amount of checking depends on what an error would cost. Checking everything to the same full depth spends limited time on prose nothing depends on, which is not the safe choice. Concentrate the checking on the claims a decision rests on.
Practice Scenario
A sales coordinator at a hotel chain has Claude draft a bid to host the annual conference of a professional association. The draft states annual guest numbers, a percentage increase in conference bookings, and that the hotel meets the association's published venue requirements. She has one evening, too little to verify every line in depth, and nothing may depend on Claude's statements alone. What is the right validation plan before submission?
Build Exercise
Build a Validation Plan for One Real Document
What you'll learn
- How to name the authoritative source that settles each claim
- How to sort claims by consequence and assign checking depth
- How to make an output traceable before spot-checking it
- How to see what self-check alone would have missed
- Take a Claude-drafted document from your work containing at least three checkable claims: a figure, an external fact, and a claim about meeting some published requirement
Why: The three claim types need three different validation moves, which makes this a small version of the full set of techniques.
You should see: One document with at least three claims underlined and categorised.
- For each claim, name the authoritative source that would settle it: your organisation’s records, the official publication, the published criteria
Why: A claim with no nameable source is untraceable, and untraceable claims are treated as unverified.
You should see: Each claim paired with a named source, or flagged as untraceable.
- Sort the claims by consequence: give the figure a decision depends on two independent confirmations, give low-stakes phrasing a read-through, and write one sentence on why each got the depth it did
Why: Proportionality is the judgement the exam tests; the sentence makes you state the reasoning.
You should see: A checking plan where depth visibly varies with stakes.
- Ask Claude to show where each remaining claim came from, then spot-check the ones that carry weight
Why: Traceability turns a vague read-through into a checkable review and puts effort where it is useful.
You should see: Claims tied to sources and reasoning, with the important ones confirmed.
- Run the plan and record which claims failed, which source caught them, and what a self-check-only plan would have missed
Why: The comparison against self-check makes the independence principle concrete on your own work.
You should see: A short record: failures found, sources that caught them, and the self-check gap.
Sources
- Enable and use web search — Anthropic
- Claude is providing incorrect or misleading responses. What's going on? — Anthropic
- Use research on Claude — Anthropic
- Prompting best practices — Anthropic
- Reduce hallucinations — Anthropic
- Claude Certified Associate – Foundations Exam Guide — Anthropic