- ClaimLock is a source-locked claim auditing AI agent skill. It checks claims in your draft against source texts you supply.
- Claude extracts claims and proposes evidence, then an offline Python script runs the text-matching checks for numbers, dates, quotations, and overstated wording.
- The output is a claim ledger with verdicts, findings, a coverage percentage, a risk summary, an annotated or clean draft, and a list of checks not measured.
- It does not search the internet, accepts plain text and Markdown only, and does not check every sentence.
- It shows whether a claim matches your sources. It cannot prove the source or the claim is true.
ClaimLock: An AI Agent Skill That Checks Your Claims Against Sources
Writing a factual article, report, or research brief takes more than good writing. Every important claim needs evidence, and even a small mistake in a number, date, quotation, or attribution can undermine the credibility of the entire document.
Checking every sentence manually is time-consuming, especially when a draft draws on several reports, surveys, or research documents. ClaimLock is an AI agent skill designed to help. It audits a draft against source texts you provide, checks claims against the evidence, and identifies wording that needs further review.
Instead of relying only on an AI assistant's judgment, ClaimLock combines AI-assisted claim extraction with an offline Python script that checks whether specific evidence passages and factual details match the supplied sources. You can open ClaimLock on Capafy at any point.
What Is ClaimLock?
ClaimLock is a source-locked claim auditing skill for AI-assisted writing. It helps writers, editors, researchers, consultants, and communications professionals find claims that may not be adequately supported by the source material they provide. It does not search the internet or research a topic itself: you supply the draft and the sources.
The skill is designed for factual writing where accuracy matters, including:
- Research briefs and business reports
- Blog articles and editorial content
- Market analysis
- Communications documents
- Reports prepared from surveys or annual reports
The key idea is simple: make it easier to identify which claims have matching textual evidence and which claims need another look before publication. It supports the evidence-checking stage that professional fact-checking relies on.
🔒 Want to lock your claims to your sources?
Try ClaimLock →How Does ClaimLock Work?
ClaimLock works in four steps: you provide a draft and source documents, Claude identifies checkable claims and proposes evidence passages, an included Python script verifies the evidence with deterministic text checks, and ClaimLock generates an audit report. The AI proposes; the script checks. A claim is never accepted just because it sounds plausible.
1. Provide your draft and source documents
Start with the draft you want to review, then provide the source text that supports its claims. Each source gets a short identifier, such as annual_report or city_survey, and optionally a title and publication date.
ClaimLock works with plain text and Markdown. Its documented workflow does not accept PDF files or web pages directly, so you supply the relevant text. If you have no sources, the skill must report that claims cannot be checked against supplied evidence. It should not invent source material or rely on memory.
2. Identify claims that need checking
Claude examines the draft and finds sentences with checkable information, for example:
- A company opened on a specific date.
- A survey found a particular percentage.
- An organization employed a stated number of people.
- A report contains a quotation.
- A sentence makes a strong or absolute statement.
The agent prepares a list of claims and proposes evidence passages from the supplied sources. The claim text must be copied from the draft, and the evidence passage must be copied from the source.
3. Verify the evidence with Python
The included script checks the proposed claims against the draft and source files. It looks for textual mismatches involving numbers and percentages, dates and years, quotations, evidence passages, unsupported claims, certain absolute or causal statements, and claims that may be stale or weakly connected to the evidence. Claude proposes the claims and evidence, but the script performs the documented text-matching checks.
4. Generate an audit report
After the audit, ClaimLock produces a structured report:
- A claim ledger with individual verdicts.
- Findings explaining problems.
- A coverage percentage and an overall risk summary.
- A clean or annotated version of the draft.
- A list of checks that were not measured.
What Does ClaimLock Check?
ClaimLock checks five kinds of problems in a draft: number mismatches, date and year mismatches, quotation mismatches for quotes of three or more words, unsupported claims that have no evidence attached, and overstated wording such as "everyone," "always," "never," or "guaranteed." Each finding is a flag for review, not proof that a claim is false.
Number Mismatches
Flags figures that differ from the supplied source.
Date & Year Mismatches
Catches a wrong opening year or reporting date.
Quotation Mismatches
Checks quotes of three or more words against the cited source.
Unsupported Claims
Marks claims with no evidence attached.
Overstated Wording
Flags absolutes and causal language the evidence may not support.
Number mismatches
Suppose your draft says, "According to the city survey, 71% of residents shop there weekly," but the supplied survey says, "62% said they shop at Harbor Market weekly." ClaimLock can flag the difference between 71% and 62%. This helps when figures were copied from different documents or changed during editing.
Date and year mismatches
If a draft says a business opened in 2022 but the supplied evidence says 2021, the script can identify the discrepancy. Dates matter most in news articles, financial reports, company histories, and market research.
Quotation mismatches
ClaimLock checks whether quotations of three or more words appear in the cited source. If a source says, "We source most produce from farms within fifty miles," but the draft says, "We buy all produce from farms within fifty miles," the altered wording creates a stronger statement, and ClaimLock can flag it.
Unsupported claims
If a claim has no evidence attached, the audit can mark it as unsupported. That helps you find sentences that need a source, a correction, or removal before publication.
Overstated wording
Words such as "everyone," "always," "never," or "guaranteed" can make a statement stronger than the evidence allows. ClaimLock flags certain absolute words and causal wording for review, especially when the source does not support such a strong claim.
Understanding ClaimLock's Verdicts
ClaimLock labels each claim with one of seven verdicts: Supported, Mismatch, Unsupported, Overstated, Weak, Stale, or Altered-claim. They help you prioritize review. A high-severity finding does not mean a claim is false, and a Supported verdict does not mean it is objectively true; each only describes how the claim fared under the documented checks.
| Verdict | Meaning |
|---|---|
| Supported | The documented text checks found no mismatch in the supplied evidence. |
| Mismatch | A number, date, or quotation differs from the supplied evidence. |
| Unsupported | No evidence was supplied, or the cited passage could not be found in the source. |
| Overstated | The wording includes an absolute or causal claim that the evidence does not support. |
| Weak | The claim includes names or topics that are not adequately connected to the supplied evidence. |
| Stale | A present-tense claim may rely on a source that is too old. |
| Altered-claim | The claim text does not match the draft exactly. |
A Practical Example: Auditing a Business Report
Imagine a draft about a local market with four claims: an opening date, a headcount, a survey result, and a sales increase with a stated cause. You give ClaimLock the draft, an annual report, and a city survey. It extracts the claims, proposes evidence, and the script flags any number mismatch or overstated wording before you publish.
Your draft contains these statements:
- Harbor Market opened on 4 March 2021.
- The market employed 48 people at the end of 2023.
- A city survey found that 62% of residents shop there weekly.
- Sales rose 12% in 2023 because of a new loading dock.
If the sources contain the same date, number, quotation, and wording, a claim may receive a Supported verdict under the documented checks. Now imagine you change the survey result to 71%, or add a statement that everyone agrees the market is the best in the region. ClaimLock can flag the number mismatch or the overstated wording, so you find those issues before publishing.
Tip: Once your claims check out, polish the page itself. Preview your snippet with the SERP Preview Tool and review keyword balance with the Keyword Density Checker.
How to Use ClaimLock
To use ClaimLock, prepare your draft in plain text or Markdown, supply sources with short identifiers, let Claude extract claims and evidence, run the audit script, review the claim ledger, then correct, soften, source, or remove problem claims and re-run the audit. The script runs offline with Python's standard library and makes no network requests.
- Prepare your draft in plain text or Markdown.
- Paste or upload the source texts you want to use.
- Give each source a short identifier and publication date when known.
- Let Claude identify the checkable claims and propose evidence passages.
- Run the ClaimLock audit script.
- Review the claim ledger and findings.
- Correct, soften, source, or remove problematic claims.
- Re-run the audit after accepting changes.
The package includes sample files and a deliberately flawed audit example that shows how mismatches and unsupported claims are reported. The script uses the Python standard library and does not contact external websites.
Check Your Claims Before You Publish
Add a disciplined evidence checkpoint to your writing workflow, with a claim ledger and an annotated draft.
Try ClaimLock on Capafy → More AI Agent SkillsChecks textual matches only · Not a complete fact-checking system
Is ClaimLock Free?
ClaimLock's Python script uses the standard library and runs offline, so the script itself needs no paid API or network connection. The AI-assisted part of the workflow depends on the model and runtime configured for the skill, and the Capafy listing specifies Claude Sonnet 5. Your total cost depends on the platform through which you run it.
That means you should distinguish between the offline auditing script and the AI service used to prepare claims and explain results. The complete workflow may depend on the platform you use to run the skill.
Important Limitations
ClaimLock checks textual matches and flags potential problems; it is not a complete fact-checking or research system. It cannot verify that a source is true, cannot fully judge whether a paraphrase preserves meaning, does not check every sentence, and does not replace professional review for legal, medical, financial, or other high-stakes content.
It does not verify whether sources are true
A source may contain incorrect information. ClaimLock can identify that a claim matches a passage in that source, but it cannot establish that the source itself is accurate.
It does not establish that a paraphrase preserves meaning
A claim may use different wording from its source and still be accurate. Conversely, wording may match while the meaning has changed. ClaimLock does not fully evaluate whether the evidence logically proves the claim.
It does not check every sentence
The documented checks focus on particular markers: numbers, dates, quotations, absolutes, causal language, and certain attribution or present-tense patterns. A factual statement without those markers may not be identified automatically.
It does not replace professional review
For legal, medical, financial, or other high-stakes content, treat ClaimLock as an additional review aid, not a substitute for qualified professional judgment. Remove passwords, API keys, ID numbers, and confidential material you cannot share before submitting drafts or sources. For editorial standards, see the IFCN Code of Principles.
Remember: a Supported verdict means the supplied text matched under the documented checks. It does not mean the claim is objectively true.
Who Should Use ClaimLock?
ClaimLock suits bloggers, content writers, editors, researchers, consultants, and communications teams who regularly prepare factual documents and want a structured way to check them against source material. It is most useful when the draft and its supporting source texts are available as plain text or Markdown, which are the formats the skill can process.
- Bloggers and content writers: review articles that contain statistics, dates, or quotations.
- Editors: identify claims that need additional evidence before publication.
- Researchers: organize evidence checks for reports and research briefs.
- Consultants: review factual statements in business documents.
- Communications teams: identify potentially overstated claims in public-facing materials.
Looking for more skills in the same spirit? Read about ScoreTrace AI for auditing forecasts and T&C Liability Shield for reviewing terms.
ClaimLock — Frequently Asked Questions
These answers cover what ClaimLock is, what input it accepts, whether it searches the internet, what a Supported verdict means, and whether it is free. In short, ClaimLock compares your draft with source texts you supply and flags mismatches, but it cannot prove that a source or a claim is true, and it does not check every sentence.
Final Thoughts
ClaimLock organizes the evidence-checking stage of factual writing by pairing AI-assisted claim extraction with an offline Python script that finds mismatches in numbers, dates, quotations, and wording. It can show whether a claim matches the source text you supplied, but it cannot guarantee the claim is true. Use it as an extra checkpoint alongside human judgment.
Review the flagged findings carefully, and verify important claims with appropriate sources and human judgment. That makes ClaimLock a useful starting point for a more disciplined and transparent content-review workflow. Browse more in our AI Agent Skills category and the Free AI Online Tools collection.
Next step: Try ClaimLock on Capafy and audit your next draft against its sources.
