Verify authenticity and detect hallucinations in literature references of any type and any language: journal articles, books, web articles, conference papers, preprints, newspaper/magazine articles, dissertations, government documents, and more. Full support for both English and Chinese (Chinese) literature verification.
Verification Workflow
- Detect language: Determine if the reference is Chinese or English (or mixed)
- Parse the reference to extract structured fields (title, authors, year, journal/venue, DOI, URL, ISBN)
- Verify existence using appropriate sources:
- English: DOI → CrossRef; URL → HTTP check; title → CrossRef search
- Chinese: title → CNKI via Chrome CDP (primary), WebSearch (fallback); DOI → CrossRef (if available)
- Cross-check metadata between claimed and actual values
- Assess confidence and report findings
- Verify claims if the user asks about specific content attributed to the source
Verification by Identifier Type
Has DOI
Run scripts/verify_doi.py "<doi>". Compare returned metadata (title, authors, journal, year) against the user's citation. Flag any mismatch.
Has Title (No DOI)
Run scripts/search_crossref.py --title "<title>". If results return, compare the top match's metadata against the citation. If no match, try adding --author "<last_name>".
Has URL
Run scripts/verify_url.py "<url>". Check reachability, page title, and extracted metadata (citation_doi, citation_title, article_author). If URL is dead, suggest Wayback Machine: web.archive.org/web/<url>.
Has ISBN (Books)
Use WebSearch to query "ISBN <number>" site:openlibrary.org OR site:worldcat.org and verify the book record.
Minimal Information
Use WebSearch to search for the claimed title + author + year. Cross-reference results from Google Scholar (site:scholar.google.com), publisher sites, and library catalogs.
Chinese Literature Verification (Chinese Literature Verification)
Chinese literature requires different verification strategies because most Chinese publications are not indexed in CrossRef. When the Chrome MCP tool is available, CNKI CDP direct search must be used as the preferred verification method; when the MCP tool is unavailable, use WebSearch as the fallback. Do not merely generate search links or brush off with "needs CNKI verification". Each Chinese reference must be given a clear verdict.
CNKI CDP Direct Search (preferred method for Chinese journal articles)
When the Chrome MCP tools are available (navigate_page, take_snapshot, take_screenshot, fill, click, evaluate_script, wait_for), use this method first for each Chinese journal article. This method queries the CNKI database directly, with accuracy far higher than WebSearch.
Pre-check: try executing on any page take_snapshot, if the MCP tool responds normally, use the CDP method; if unavailable (error/timeout), skip to the WebSearch fallback process below.
Batch verification note: when searching multiple papers consecutively, wait 2-3 seconds between each search to avoid triggering CNKI's anti-scraping mechanism.
CDP Step 1: Open the CNKI advanced search page
navigate_page → https://kns.cnki.net/kns8s/AdvSearch
CAPTCHA handling: the snapshot's DOM may always contain the hidden text "drag the puzzle below to complete verification",do not judge whether a CAPTCHA has appeared based solely on the snapshot text. You must use take_screenshot screenshots to check the actual display state of the page:
- If in the screenshotis visibleCAPTCHA slider popup blocking the page, prompt the user: "CNKI requires security verification, please complete the slider verification in the browser and tell me when done."
- If the page in the screenshot normally displays the search form (no blocking), continue directly.
CDP Step 2: Enter the exact title of the paper to be verified
Use take_snapshot to find the search input box, then:
fill → [检索输入框] → "<待核查论文的完整标题>"
The default search field is "Subject", which is sufficient for exact title verification; no need to change the field type or add other filter conditions.
CDP Step 3: Execute the search
Click the "Search" button and wait for the results page to load:
wait_for → "检索结果" 或结果列表出现
If a CAPTCHA appears at this point, handle it as in Step 1.
CDP Step 4: Read and parse the results
First use take_snapshot to view the results page, then use evaluate_script to extract structured data:
evaluate_script → () => {
const rows = document.querySelectorAll('.result-table-list tbody tr');
if (!rows || rows.length === 0) return JSON.stringify({found: false, results: []});
const results = [];
rows.forEach((row, i) => {
if (i >= 10) return;
const titleEl = row.querySelector('.name a');
const authorsEl = row.querySelector('.author');
const sourceEl = row.querySelector('.source');
const dateEl = row.querySelector('.date');
results.push({
title: titleEl ? titleEl.textContent.trim() : '',
authors: authorsEl ? authorsEl.textContent.trim() : '',
source: sourceEl ? sourceEl.textContent.trim() : '',
date: dateEl ? dateEl.textContent.trim() : ''
});
});
return JSON.stringify({found: results.length > 0, count: results.length, results});
}
If the JS selector does not work(CNKI may have updated its DOM structure), fall back to directly reading take_snapshot 's text output, which contains readable title, author, journal, and date information that can be parsed manually.
If the result is zero: the paper was not found in CNKI. This is an important signal but not the sole basis; continue with the WebSearch fallback process to check Wanfang and Baidu Scholar before drawing a conclusion.
CDP Step 5: Metadata comparison
Compare the CNKI results item by item with the claimed citation information:
| Field | Comparison method |
|---|---|
| Title | Exact match after removing whitespace. If the CNKI result title contains the claimed title (or vice versa), it is considered a match. |
| Author | Check whether the claimed first author appears in the CNKI author list. Author order may differ. |
| Journal | Exact name match. Note abbreviation differences (e.g., "北京大学学报" vs "北京大学学报(哲学社会科学版)"); substring matching in either direction counts as a match. |
| Year | The year in the CNKI date field must match the claimed year. |
CDP Step 6: Determine based on CDP results
| CDP Result | Verdict |
|---|---|
| Exact title match + author match + journal match + year match | Confirmed |
| Exact title match + 2 of the remaining 3 items match | Likely Real(note the differing items) |
| Exact title match + journal or author mismatch | Metadata Error(the paper exists but the citation information is incorrect) |
| No title match among the first 10 results | Proceed with the WebSearch fallback process before making a determination |
CDP Exception Handling
- MCP tool unavailable: skip entirely and use the WebSearch fallback process
- CAPTCHA persistently blocking: ask the user to handle it; if the user cannot handle it, skip to WebSearch
- Search error after page load: retry once; if it still fails, skip to WebSearch
- CNKI inaccessible: skip to WebSearch
- DOM selector returns empty: switch to
take_snapshottext for manual parsing; if that also fails, skip to WebSearch
WebSearch Chinese Verification Process (fallback method / used when CDP is unavailable)
When the Chrome MCP tool is unavailable, or when CNKI CDP retrieval returns zero results and cross-validation is needed, or when additional reinforcement of the CDP determination is needed, use the following WebSearch process. Execute all of the following searches; do not skip any step, and do not mark as "Uncertain" before completing all searches.
Step 1: Multi-source WebSearch (required — at least 3 searches)
Execute these WebSearch queries in sequence for each paper:
- Exact title search:
"<完整论文标题>"(with quotes, no site restriction) - Title + author search:
<论文标题> <作者姓名>(without quotes, broader match) - Title + journal search:
<论文标题> <期刊名>(cross-validate venue) - CNKI-targeted search:
"<论文标题>" site:cnki.net(CNKI) - Wanfang-targeted search:
"<论文标题>" site:wanfangdata.com.cn(Wanfang)
If steps 1-3 already confirm the paper exists with matching metadata, steps 4-5 are supplementary. If steps 1-3 yield no results, steps 4-5 are MANDATORY.
Step 2: Metadata Cross-check (Metadata Cross-check)
From the search results, verify:
- Author name matches
- Journal name matches exactly (beware journal bulletin/magazine/periodical confusion)
- Publication year matches
- Volume/issue/page numbers if available
Step 3: Verdict (a clear determination must be given)
Based on search results, assign ONE of these verdicts — "Uncertain" is NOT acceptable as a final verdict for Chinese literature:
| Verdict | Criteria |
|---|---|
| Confirmed | Found on 2+ sources (CNKI, Wanfang, Baidu Scholar, Google Scholar) with matching metadata |
| Likely Real | Found on 1 source with matching metadata, OR found with minor metadata discrepancies |
| Likely Fabricated | No results from any search, OR title/author/journal combination not found anywhere |
| Confirmed Fabricated | Multiple fabrication indicators: journal doesn't exist, author not in claimed institution, impossible date, etc. |
| Metadata Error | Paper exists but with different author/year/journal than claimed |
Chinese Journal Articles (Chinese Journal Articles)
- Preferred: execute the "CNKI CDP Direct Retrieval" process above (requires the Chrome MCP tool)
- Fallback: if CDP is unavailable or returns zero results, execute the "WebSearch Chinese Verification Process"
- Supplementary: run
scripts/search_cnki.py --title "<中文标题>"Try CrossRef lookup (some Chinese journals have DOIs) - If the reference has a DOI, run
scripts/verify_doi.py(Note: most Chinese references have no DOI; absence of a DOI does not mean fabrication)
Chinese Books / Monographs (Chinese Books / Monographs)
- Use WebSearch to query
"<书名>" "<作者>" site:book.douban.com(Douban Books) - Use WebSearch to query
"ISBN <号码>"if ISBN is provided - Search National Library of China:
"<书名>" site:opac.nlc.cn - Verify publisher exists and has published the claimed work
Chinese Dissertations (Dissertations)
- Use WebSearch to query
"<论文标题>" 学位论文 site:cnki.net - Use WebSearch to query
"<论文标题>" 学位论文 site:wanfangdata.com.cn - Verify the degree-granting institution has the relevant discipline
Chinese Government Documents (Policy Documents)
- Use WebSearch to query
"<发文字号>" site:gov.cn - Verify the issuing agency and document number format
Chinese News Articles (News Reports)
- Use WebSearch to query on the claimed media's domain:
"<标题关键词>" site:<媒体域名> - For People's Daily:
site:people.com.cn; for Xinhua:site:xinhuanet.com
Core Journal Verification (Core Journal Verification)
When a reference claims the journal is a core journal (core journal), verify against:
- Peking University Core: 《A Guide to the Core Journals of China》— use WebSearch to check
- CSSCI (Nanjing University Core): Chinese Social Sciences Citation Index source journals — use WebSearch to check
- CSCD: Chinese Science Citation Database source journals — use WebSearch to check
- Note: Core journal status changes across editions; verify for the specific year claimed
Key Differences for Chinese Literature
- No DOI ≠ fabricated: Most Chinese journal articles lack DOIs; absence of DOI is not evidence of fabrication
- Author name formats: Chinese names in English contexts may appear as "Zhang San", "San Zhang", or "S. Zhang"
- Indexing lag: Recently published papers may not yet appear in CNKI
- Paywall: CNKI/Wanfang require paid access for full text, but search results and abstracts are usually visible
- Never defer verification: Do not use "needs CNKI verification" or "Uncertain (needs CNKI verification)" as the final result. You must actually search using CNKI CDP or WebSearch and give a definitive judgment.
- CNKI CDP first: When Chrome MCP tools are available, prefer using CDP to search directly on the CNKI search page; this is more reliable than WebSearch (searching CNKI indirectly via Google) and can find CNKI papers that Google has not indexed. When MCP is unavailable, use WebSearch as a fallback. Python scripts (verify_chinese.py, search_cnki.py) are supplementary only.
Hallucination Detection
When verification fails or metadata doesn't match, consult references/hallucination-patterns.md (English) or references/chinese-hallucination-patterns.md (Chinese) to identify which hallucination pattern applies. Common red flags:
- DOI doesn't resolve → likely fabricated DOI (but NOT for Chinese literature without DOI)
- Title search returns no results → likely fabricated title
- Authors don't match → author hallucination
- Journal doesn't exist → venue hallucination
- Year is off → date hallucination
- Chinese journal claimed as core but not in official list → core journal hallucination
- CNKI + Wanfang + Baidu Scholar all return no results → strong indicator of fabrication for Chinese literature
Batch Verification
When the user provides a reference list, verify each entry sequentially. Produce a summary table:
| # | Citation (short) | DOI verified | Title match | Author match | Year match | Confidence |
|---|-----------------|-------------|-------------|--------------|------------|------------|
| 1 | Smith 2020... | Yes | Yes | Yes | Yes | Confirmed |
| 2 | Jones 2019... | No DOI | No match | - | - | Likely Fabricated |
Confidence Levels
| Level | Criteria |
|---|---|
| Confirmed | DOI resolves AND metadata matches across sources |
| Likely Real | DOI resolves OR title+author match found, minor discrepancies |
| Uncertain | No DOI, no exact title match, but components are plausible. For Chinese literature, this level is NOT acceptable as final verdict — must execute full WebSearch procedure first. |
| Likely Fabricated | DOI doesn't resolve, no matching work found, hallucination patterns detected |
| Confirmed Fabricated | Multiple fabrication indicators, no trace in any database |
Content Claim Verification
When the user asks whether a specific claim is actually stated in a source:
- Retrieve the source (via DOI link, publisher URL, or web search)
- Use WebFetch to read the page content if accessible
- Search for the specific claim, statistic, or quote
- Report whether the claim is supported, unsupported, or contradicted
Report Format
For each verified reference, output:
**Reference**: [original citation text]
**Status**: [Confirmed / Likely Real / Uncertain / Likely Fabricated / Confirmed Fabricated]
**Findings**:
- DOI: [resolves / not found / not provided]
- Title: [exact match / partial match / no match]
- Authors: [match / mismatch / details]
- Journal/Venue: [verified / not found]
- Year: [correct / incorrect (actual: XXXX)]
**Issues**: [list any discrepancies or hallucination patterns detected]
**Actual Source** (if different): [correct metadata if the reference is a distortion of a real work]
Resources
scripts/verify_doi.py— Verify DOI existence via CrossRef and DOI.org APIs. Returns metadata for comparison.scripts/search_crossref.py— Search CrossRef by title/author/keywords. Find whether a claimed work exists.scripts/search_cnki.py— Generate search URLs for Chinese databases (CNKI, Wanfang, CQVIP, Baidu Scholar) and attempt CrossRef lookup. Supplementary tool — always use WebSearch as primary verification method for Chinese literature.- CNKI CDP direct search (via MCP tools) — use Chrome DevTools Protocol to search directly on the CNKI search page; the preferred method for verifying Chinese journal articles. Requires MCP tools: navigate_page, take_snapshot, take_screenshot, fill, click, evaluate_script, wait_for. See the "CNKI CDP direct search" section above.
scripts/verify_chinese.py— Attempt direct HTTP verification against Baidu Scholar, CNKI, and Wanfang. May be blocked by anti-bot measures — if so, fall back to WebSearch. Usage:python scripts/verify_chinese.py --title "<中文标题>" --author "<作者>"scripts/verify_url.py— Check URL reachability and extract page metadata (title, author, DOI from meta tags).references/hallucination-patterns.md— Catalog of common hallucination types for English literature. Read when fabrication is suspected.references/chinese-hallucination-patterns.md— Catalog of hallucination types specific to Chinese literature (Chinese literature hallucination patterns). Read when Chinese reference fabrication is suspected.references/verification-checklist.md— Comprehensive step-by-step checklist for English literature verification.references/chinese-verification-checklist.md— Comprehensive checklist for Chinese literature verification (Chinese literature verification checklist), including CNKI, Wanfang, core journal, dissertation, and government document checks.