Find unsupported numerical claims
Scan draft copy for percentages, counts, dates, and research-style statements that appear to need evidence.
Citation Gap Checker
Free, no signup. Find claims that need a source before an answer engine repeats them, and flag dated evidence for a human re-check.
Before
“Studies show conversion rose 42% in 2021.”
After
“A 2021 study reported a 42% increase [source]” — then verify the statistic remains applicable.
Pasted content is analyzed in your browser. A URL is sent only to the protected fetch endpoint; paste content takes priority. After a successful URL check, only its domain is remembered locally.
Checks run from our server; we fetch the URL you enter and don't keep the results. The claim pass runs in your browser. Up to 25 cited URLs are sent to the bounded status checker; pasted prose is not sent. Anonymous run-level outcome counters may be used for aggregate research; URLs, domains, IPs, and identifiers are never included, and no statistic is released below 100 runs.
Sample report Deterministic local example
For “A 2021 study shows checkout changes increased conversion by 42%.”, evaluated in 2026, the engine returns one likely claim and two findings:
1 likely claim checked · 2 findings
Missing adjacent source — This claim has no adjacent outbound source link.
Aging evidence — This numeric claim references evidence at least three years old; confirm it still applies.
Both findings can be true at once: the sentence needs a source, and the source’s continued applicability needs checking.
How to use it
- Paste article text or HTML. The content extractor removes markup before sentence analysis.
- Select Check citation gaps.
- Review every highlighted sentence; use the search link as a research starting point, not as evidence.
- Add the most direct primary source beside the claim, or revise the claim when no suitable evidence exists.
Site passport Local context for this saved site
Local data
Saved targets, named lists, and recent check summaries remain only in this browser.
Rate this tool
What the results mean
- Missing adjacent source — a likely factual claim contains no URL in the same sentence.
- Aging evidence — a detected year is at least three years behind the current year.
- Broken citation / Verify citation — supported by the engine when link-status evidence is supplied, but live status is not evaluated by this page.
- No finding — no matching pattern was found; it is not an accuracy certificate.
How it works
The claim pass runs in the browser. It extracts readable blocks, splits them into sentences, and selects likely claims using explicit markers: numbers, percentages, research language, and strong superlative or absolute wording. It checks the same sentence for an HTTP(S) URL, extracts four-digit years, and sends at most 25 cited URLs to the bounded status checker. Evidence at least three years old receives a review flag. No model writes or judges the result.
Features
- Accepts plain text or HTML.
- Finds numerical, comparative, absolute, and research-style claims.
- Separates missing sourcing from aging-evidence review.
- Builds a prefilled source-search query for each finding.
- Keeps pasted content in the browser.
Limitations
Adjacency is a crude proxy: a URL in the sentence may be irrelevant, while a footnote or nearby citation may be perfectly valid but still missed. The three-year threshold is intentionally generic. The bounded status check does not read papers, assess source quality, or determine whether a citation entails the exact claim. Always verify the primary source and the claim’s scope yourself.
Frequently asked questions
What counts as a likely claim?
The checker conservatively looks for sentences with numbers, percentages, research language, or strong comparative and absolute terms such as best, largest, always, and proves. Short fragments are ignored.
Does the tool verify that a citation supports a claim?
No. It does not fetch links or interpret source material. It only checks whether a likely claim has an adjacent URL and whether dated evidence merits a new human review.
Why is a three-year-old statistic flagged?
Three years is a review trigger, not an expiration date. Some evidence remains valid for decades; volatile market, product, and behavior statistics can become stale much sooner.
Can a clean report prove an article is accurate?
No. A clean result only means the local heuristics did not find one of their known patterns. Qualitative assertions, misleading framing, and irrelevant sources still require editorial review.
Does the checker detect broken citation links?
Yes. After the local claim pass, the page sends up to 25 cited URLsAn AI citation is the visible source link an AI answer engine shows next to its generated text — the clickable reference that credits the web page it used. A citation's presence is a separate thing from whether the cited page actually supports the statement, and from being retrieved (read behind the scenes) or merely mentioned (named without a link); citation is driven more by brand mentions and being retrievable than by traditional ranking. to the bounded status checker. A failed status can flag a broken citation or a DOI that needs manual verification; source relevance still requires human review.
Feature requests for Citation Gap Checker
Upvote what you want most. New ideas can be submitted from the floating Feedback menu; requests appear here once approved, and the most-wanted rise to the top.
You won't be emailed about that request anymore.
Loading…
➕ Request a feature
New requests are reviewed before they appear here.
Where this tool helps
Common use cases
Refresh aging statistics
Identify dated claims that should be verified or replaced before a content update goes live.
Build a fact-checking queue
Turn likely citation gaps into a focused list for editorial or subject-matter review.
Recheck a sourced revision
Run the updated copy again after adding citations, qualifying language, or removing unsupported claims.
Watch the full workflow
Citation Gap Checker walkthrough
Read the transcript
Citation Gap Checker
Unsourced and aging claims can weaken readers’ trust and answer-engine reuse. I’ll show you when this checker helps, how to paste or fetch content, read the live deterministic findings, distinguish missing, aging, and broken evidence, use source-search handoffs, understand limitations, and complete a careful editorial review.
Step 1
This tool conservatively finds sentences with numbers, percentages, research language, superlatives, or absolute terms, then checks for adjacent U-R-Ls and aging years. Use it as a citation review queue, not an accuracy score or a substitute for reading sources.
Step 2
Use it before publication, during content refreshes, after importing legacy copy, or when reviewing statistics, market claims, product comparisons, medical or legal statements, and absolute language. Prioritize claims that affect decisions, money, safety, reputation, or rapidly changing conditions.
Step 3
Paste plain text or H-T-M-L, or fetch a public page, then select Check citation gaps. Claim detection and prose stay in the browser. Only up to twenty-five cited U-R-Ls can be sent for bounded status checks; those checks do not read or judge the sources.
Step 4
The sample sentence produces one likely claim and two findings. It lacks an adjacent outbound source, and its twenty-twenty-one statistic is at least three years old in twenty-twenty-six. Both can be true: first find the source, then verify that its evidence still applies.
Step 5
The live local example also flags absolute wording. Each finding preserves the sentence, explains the trigger, provides rationale and a verification step, and offers a search handoff. Treat the search as a starting point. Do not cite a result page or a convenient secondary summary without checking provenance.
Step 6
Missing adjacent source means a likely claim has no H-T-T-P U-R-L in the same sentence. That is a crude proximity heuristic. Locate the most direct primary evidence, confirm it entails the wording, and place a clear citation beside the claim or rewrite the unsupported assertion.
Step 7
Aging evidence means the detected year is at least three years behind the current year. It is not an expiration rule. Foundational research can remain valid; product, market, behavior, price, and adoption statistics may stale quickly. Review volatility, method, and present applicability.
Step 8
A bounded status check can identify unreachable citations or D-O-I links needing manual review. A two-hundred response does not prove relevance, quality, accuracy, or entailment. Read the source, locate the supporting passage, and confirm the claim does not exceed the study’s population or result.
Step 9
Features include text and H-T-M-L extraction, conservative numerical and comparative detection, separate missing and aging findings, bounded citation status checks, prefilled searches, detailed guidance, page fetching, privacy notes, and downloadable ticket C-S-V for assigning research and revision work.
Step 10
No finding only means the known local patterns did not fire. Qualitative assertions, misleading framing, omitted context, footnotes, nearby citations, images, tables, and subtle causal claims may be missed. A captured-prefix result also cannot clear the unobserved remainder of a page.
Step 11
Sentence-level adjacency can flag valid footnotes or overlook irrelevant nearby links. The three-year trigger is generic, extraction can lose structure, and status checks do not inspect papers or source quality. The tool does not verify facts, causal inference, study design, or whether wording matches evidence.
Step 12
Record the sentence, claim type, source U-R-L, publication date, author, primary or secondary status, supporting passage, methodology, population, geographic and temporal scope, link status, reviewer, and next review date. Revise unsupported certainty, cite directly, rerun, and retain the editorial evidence trail.
Source the claim—or narrow it until it is true.
Open the original source, verify the exact measure, population, date, method, and scope, then add the citation beside the claim or revise the wording. Recheck links and volatile evidence regularly, and keep human editorial judgment in the loop even after a clean report.
Common issues & how to fix them
- Warning Factual claim has no adjacent source Fix: Add a relevant authoritative citation directly beside the factual claim.
- Warning Numeric claim relies on aging evidence Fix: Replace or supplement the aging evidence with a current authoritative source.
- Error Citation link is broken Fix: Replace or repair the citation URL and rerun the live status check until it returns a successful response.
- Warning Citation needs verification Fix: Open and verify the cited source or DOI, then replace it with a direct valid reference if it cannot be confirmed.