A weak source can sink a strong essay, a published paper, or a market report you stake decisions on. These 15 prompts run a structured credibility audit before a source enters your bibliography, covering author authority, evidence base, recency, bias, and the difference between a primary finding, a secondary synthesis, and a third-hand repeated claim.
One thing changed in 2026: today’s assistants can browse. ChatGPT Search, Gemini Deep Research, and Claude Research read live pages and attach citations, so “paste the text, the AI cannot fetch” is no longer the whole story. But a browsing model still surfaces content-farm pages, can fabricate or mis-attribute a citation, and will happily summarize an abstract as if it read the full paper. The audit below is the discipline that turns a browsing answer into a defensible citation.
TL;DR
- AI is a screener, not a judge. It runs the structured check fast; the final call on whether to cite is yours.
- Two checks catch most weak sources: the CRAAP test (template 1) plus the primary-vs-secondary tracer (template 3).
- Browsing AI (June 2026) can fetch and cite, but still cites content farms and invents citations. Always trace the chain back at least one level yourself.
- Use a browsing/research mode for outlet profiles and corroboration (templates 8, 12); paste raw text for line-by-line evidence and statistics audits (templates 4, 14).
- For high-stakes citations (thesis, published article, court filing), pair the AI screen with a human librarian or fact-checker.
Who this is for
Students writing essays, journalists fact-checking pieces, analysts producing market or policy reports, researchers building bibliographies, and anyone about to forward an “I read that…” claim.
When not to use these prompts
Skip the full audit for canonical sources you have already vetted (a foundational textbook in your field, a primary statute, an original peer-reviewed study you have read in full). For anything you have only seen second-hand, run at least templates 1 and 3.
CRAAP, SIFT, and which fits your case
Two evaluation frameworks underpin these prompts:
- CRAAP (Currency, Relevance, Authority, Accuracy, Purpose) was created by librarian Sarah Blakeslee at California State University, Chico in 2004. It works best when you already have a specific source in hand and want a per-dimension verdict. Templates 1, 2, 6, 7, and 9 map to it.
- SIFT / lateral reading (Stop, Investigate the source, Find better coverage, Trace claims to the original) comes from digital-literacy researcher Mike Caulfield. Instead of digging deeper into the page in front of you, you leave it and check what independent sources say about it. This is the better fit for a viral post, an unfamiliar outlet, or a too-good-to-be-true statistic. Templates 8, 11, and 12 are lateral-reading moves, and they are exactly where a browsing/research mode earns its keep.
A practical rule of thumb: use CRAAP on academic and report sources you must grade in detail; use SIFT first on anything from social media or an outlet you do not recognize.
Prompt anatomy / structure formula
A credibility prompt should carry six elements:
- Role: who the AI plays — information-literacy librarian, peer reviewer, fact-checker, research tutor.
- Context: your level, subject, target citation style (APA / MLA / Chicago), and what the source will be used for.
- Goal: one concrete deliverable — a CRAAP score, a primary/secondary table, a 4-line bibliography note.
- Constraints: which source types are allowed, what to skip, and what the AI must never assert without evidence.
- Output format: numbered list, table, or graded verdict (use / use with caveat / do not use) you can paste into Notion, Zotero, or Word.
- Examples / signal: a reference paragraph or anti-example so the model calibrates its rigor.
15 copy-ready prompt templates
1. CRAAP test prompt (Currency / Relevance / Authority / Accuracy / Purpose)
Quick general-purpose audit; good first pass.
You are an information-literacy librarian. Run the CRAAP test on the source below. For each of the 5 dimensions, score 1-5 with 1 sentence of evidence. End with one of: "use", "use with caveat", "do not use".
Source: {title, author, outlet, date, URL}
Content: {paste}Variables to swap: title, author, outlet, date, URL, content
Optimization: If output is too soft, add: “Treat a score of 3 as fail-by-default. Be ruthless on Authority and Purpose.”
2. Author authority check
Audit the author of this source for authority on the specific claim ({claim}). Cover: relevant credentials, prior publications on this topic, institutional affiliation, conflicts of interest. If author info is missing, flag as a yellow card.
Source: {paste source + author bio}3. Primary vs secondary vs hearsay tracer
For each claim in the source below, classify it as: (a) primary (data/experiment/firsthand observation), (b) secondary (synthesis citing primaries), (c) third-hand (repeats a claim without citing primaries). Output as a table: claim | classification | nearest primary source if any.
{paste}4. Evidence-base audit
List every factual claim in the source below. For each: is supporting evidence provided in the source itself? If yes, what type (study, dataset, anecdote, expert quote, "research shows")? If no, mark as unsupported.
{paste}5. Citation-chain trace
The source claims "{claim}" and cites {Reference X}. Help me trace it: what would Reference X likely say if I read it, what to look for to verify it is being represented accurately, what would indicate it has been misrepresented.6. Funding / conflict-of-interest screen
Below is a source. Identify funding sources, sponsorships, advertising relationships, or disclosed conflicts. Then assess whether the conclusions align suspiciously with the funder’s interests.
{paste source + masthead / funding section}7. Recency-vs-canon check
The source on {topic} is from {year}. Has the consensus on {claim} changed since then? Name 2-3 newer sources or developments to look for before relying on the original.8. Outlet-bias profile
Profile the outlet {outlet name}: editorial slant (if any), ownership structure, audience, peer-review status, retraction history. Mark which kinds of claims I should accept readily and which to double-check.9. Peer-review status verifier
The source claims to be peer-reviewed. Help me verify: journal impact factor, indexing in {Scopus / Web of Science / PubMed}, predatory-journal red flags (rapid acceptance, no editorial board, suspiciously broad scope, vanity fees).10. Image / quote-mining detector
Below is a quoted passage in the source. Pretend you are reading the original; what context, qualifications, or counter-evidence might have been removed in the quote? List 3 things to look for in the original.
{paste quote + surrounding sentences}11. Wikipedia-as-canary
The Wikipedia article on {topic} says: "{paste excerpt}". Trace the citation it relies on, then assess: is this a well-summarized primary source, a circular citation, or a "citation needed" weak spot? Recommend whether to trust this statement.12. Cross-source corroboration
Find 2-3 independent sources that corroborate or contradict the central claim "{claim}". For each: name source, summarize position, note independence (different funder, different lab, different country). Conclude: claim status (well-supported, contested, isolated).13. AI / blog-of-blogs detection
Audit this source for signs it is AI-generated content farm material or a "blog of blogs" with no original reporting: vague author bio, no source URLs, recycled paragraphs, suspiciously broad topic coverage, repeated phrasings.
{paste source}14. Statistical-claim audit
The source uses these statistics: {paste statistics}. For each: original source if any, sample size, methodology, time period, any obvious distortions (cherry-picked baseline, missing denominator, scale tricks). Flag any I should not cite without verification.15. Final-call bibliography note
Write a 4-line bibliography annotation for {source} covering: (a) what it claims, (b) credibility level (high / medium / use-with-caveat), (c) what to verify if I keep using it, (d) what to cite instead if I drop it.Browsing AI vs paste-the-text: which mode for which check
As of June 2026, the major assistants split into two useful modes for this work:
- Browsing / research modes — ChatGPT Search and Deep Research (GPT-5.5), Gemini Deep Research (Gemini 3.1 Pro, browses 30-60 live sources with inline citations), and Claude Research (Opus 4.7) read live pages and attach links. Use these for lateral-reading checks where the answer lives off the page: outlet profiles (template 8), Wikipedia citation traces (template 11), and cross-source corroboration (template 12).
- Paste-the-raw-text — for line-by-line work the browsing layer adds noise. Paste the actual article (or PDF text) for the evidence-base audit (template 4), the statistical audit (template 14), and quote-mining detection (template 10), so the model judges only what is in front of it.
The catch with browsing mode: a model can fetch a content farm and cite it confidently, or attach a citation that does not actually support the sentence. Click through every link it gives you; treat an unclicked citation as unverified.
Common mistakes
- Treating “found on Google Scholar” as automatic credibility. Predatory journals are indexed there too.
- Trusting a browsing AI’s cited links without opening them. Models still fabricate or mis-attribute citations.
- Stopping at the abstract or headline. The credibility issues usually hide in the methods or funding sections.
- Quoting a secondary source as if primary. Always trace back one citation level.
- Ignoring date. A 2021 source on frontier AI models is already out of date in 2026.
- Confusing “many sources say” with “many independent sources confirm.” Citation cascades manufacture the illusion of consensus.
- Skipping funding disclosure on industry-sponsored research, then being surprised the conclusions favor the funder.
- Letting the AI declare a source “credible” without traceable evidence. Require a citation or a “cannot verify” answer.
How to push results further
- Run the same source through three templates (CRAAP, primary-vs-secondary, recency). Confidence rises when the verdicts agree.
- For numerical claims, trace the chain at least two levels back. “The WSJ said the WHO said…” is brittle until you reach the dataset.
- When the source disagrees with your prior, audit harder. That is exactly when your own bias is most likely to wave it through.
- Build a personal “do not cite” list with a one-line reason each, kept in your notes for future sourcing.
- Use the AI as a screener, not a verdict. The final call sits with you.
- For high-stakes citations (thesis, published article, court filing), pair the AI screen with a human librarian or fact-checker.
FAQ
- Can AI tell me if a source is credible?: It runs a fast, structured audit and, in browsing mode, can fetch the page and add citations. It cannot vouch for a source it has not actually read, and it can cite content farms, so you confirm the verdict yourself.
- Do I still need to paste the text now that ChatGPT and Gemini can browse?: For lateral-reading checks (outlet bias, corroboration) let it browse. For line-by-line evidence and statistics audits, paste the raw text so the model judges only that content, not whatever it happened to retrieve.
- Is a peer-reviewed paper always credible?: No. Predatory journals and flawed peer review both exist. Use template 9 to check indexing and red flags.
- What is the fastest single check I can run?: The CRAAP test (template 1) plus the primary-vs-secondary tracer (template 3). Together they catch the large majority of weak sources.
- Should I cite Wikipedia?: Cite the underlying sources Wikipedia points to, not Wikipedia itself. Template 11 helps you trace and grade that citation.
- What if the source is paywalled?: Get the full text through your library or interlibrary loan. Never audit from the abstract alone, and do not let a browsing model summarize the abstract as if it read the paper.