A source credibility checker turns "can I trust this?" into answerable questions. This one scores a source across the five CRAAP tests — Currency, Relevance, Authority, Accuracy and Purpose — weights the answers to a 0–100 score, lists every weak answer with a concrete fix, and compares two sources side by side. A SIFT tab adds the four off-page moves professional fact-checkers actually use. It runs in your browser and saves nothing beyond your own device.
The score is a structured self-check. You answer the questions; the tool adds up your answers. A high score means your judgments were favorable — it cannot verify a single fact on the page you are checking, and it cannot catch a forgery built to pass a checklist.
The weights are this page's own teaching judgment, not a standard published anywhere. Professors weight the same categories differently, and two honest people can answer the same question differently. The point of the exercise is the argument you make while answering, not the final digit.
For sources that only look credible — polished design, journal-shaped layout, confident tone — pair the score with the SIFT moves in the second tab. Lateral reading catches what vertical checklists cannot.
Score a source with the CRAAP checker
Give the source a label (optional — it shows up in the compare view), then walk the fifteen questions: five categories, three to four questions each, every answer Yes, Partly or No. Yes scores the question's full weight, Partly scores half, No scores zero, and the weights sum to 100. Hit Score source A for the verdict, the category bars and the red-flag list. If you want to see the tool behave first, load the demo source — an invented, illustrative newsletter feature tuned to land mid-band. Your labels and answers save in this browser only (key csw-scc-v1) until you press Reset.
Source credibility scorer
100% private — runs in your browser0 of 15 answered — Yes scores full weight, Partly half, No zero.
—
Red flags to fix before you cite
The scorecard is plain text you can paste into research notes or an annotated bibliography draft. It records your judgments and the fix advice — it never claims to have verified the source.
SIFT is the four-move companion to any checklist, popularized by digital-literacy researcher Mike Caulfield. It works because it sends you off the page: researchers who study professional fact-checkers found they judge a source less by what the page says about itself and more by what independent sources say about it. Make the four moves on any web source before it earns a place in your references, and tap each line as you go — your ticks save in this browser alongside your scores.
- Stop. When a headline provokes you, pause and ask two questions: do I actually know this source, and is this where the claim should come from?
- Investigate the source. Open a new tab and read about the publication or author elsewhere — reputation, funders, track record — not just their own about page.
- Find better coverage. Look for the best available reporting on the claim itself: who confirms it, who disputes it, and whether a consensus exists at all.
- Trace claims to the original. Follow quotes, figures and studies back to their first context; an accurate quote can still mislead when it is ripped from context.
The SIFT steps are described generically here; the method itself is Caulfield's, and it is free to learn. If a source fails a SIFT move, no CRAAP score can rescue it — the problem is off the page.
A full pass takes five to ten minutes per source — slower than trusting a hunch, much faster than discovering in feedback week that your anchor source was a content farm. The rest of this page explains where CRAAP comes from, what each category actually measures, where checklists fail, and how to write the evaluation sentence your professor wants to see.
What the CRAAP test is and where it came from
CRAAP is a five-part checklist for judging a source before it enters your reference list. The letters name the tests: Currency (is it fresh enough for this topic), Relevance (does it answer your actual need), Authority (who stands behind it), Accuracy (does the evidence hold together) and Purpose (why does this page exist). Librarians at California State University, Chico assembled the acronym, and it spread through composition and research courses because it turns criteria librarians already applied into questions a first-year student can answer without a library science degree.
The checklist's real contribution is not any single question — librarians asked them all before 2004. It is packaging: fifteen-ish concrete questions, an acronym memorable enough to survive exam week, and a rubric-shaped format a writing center can teach in one handout. The version on this page keeps the five categories and adds three things handouts cannot do: weights that sum to 100 so the categories compete honestly, a fix list that turns every weak answer into a next action, and an A/B mode for the moment when you have two plausible sources and one citation slot.
What it does not add is authority. The score reflects your answers, and your answers reflect how carefully you looked. Treat the number as a mirror with categories printed on it — useful for finding your own blind spots, worthless as a certificate.
Why checklists alone miss things: lateral reading
Every checklist question inspects the page in front of you — its dates, its byline, its citations. That is vertical reading, and it has a blind spot exactly the shape of modern misinformation: sources that are built to pass. A predatory journal imitates the surface signals of peer review. A lobbying front imitates the neutral tone of a think tank. A content farm imitates the structure of news. On the page itself, the forgery is often flawless.
Research on how professional fact-checkers work — most famously the studies out of Stanford — found that they barely read the page at all before judging it. Within seconds they open new tabs and read about the source: who funds it, what its reputation is, whether anyone credible echoes the claim. This is lateral reading, and it inverts the student instinct of finishing the page first. The fact-checkers' advantage is not better judgment; it is better geography.
Mike Caulfield's SIFT method — the four moves in the second tab of the scorer — packages that behavior for students. The honest division of labor: CRAAP audits what the source shows about itself, SIFT checks what the rest of the web says about the source. Run both and the forgery's advantage mostly evaporates. Run CRAAP alone and you are grading the disguise.
Currency: is the source fresh enough to build on?
Currency sounds trivial until you meet its two traps. The first is the undated page: no byline date, no revision note, a copyright line frozen years ago. Undated is not always disqualifying — an archive of historical documents does not need freshness — but you cannot know that until you know when it was written, so the first move is always the same: check the header, the footer, the about page, and if all fail, an archive.org capture that shows when the content actually changed.
The second trap is assuming fresh design means fresh data. A page rebuilt last month can still be serving figures gathered a decade ago, and the layout will not tell you. The question that matters is whether the evidence is current for your topic's clock: sources on social media behavior age like produce, sources on the Peloponnesian War age like wine (both comparisons illustrative).
A strong currency profile, illustratively: a data page that shows a "last updated" line refreshed within the month and revises its numbers when the underlying agency revises. A weak one: a statistics page headlined "latest figures" whose newest citation is a decade old. The fix for the weak case is not always abandonment — for a history-of-the-field paragraph the older source may be perfect — but the age has to be known, never assumed.
Relevance: the most skipped test, the most expensive to fail
Relevance fails quietly. Students skip it because everything looks relevant in a results list — the title contains your keywords, the abstract contains your topic — and the failure surfaces weeks later as a grader's comment: "sources mentioned but not used." Related is not the same as answering. A source can be current, authoritative and accurate, and still be the wrong source because it answers a neighboring question.
Three sub-questions keep it honest, and the scorer asks all three. Does it answer the exact prompt, or a cousin of it? Is the depth right — a literature review needs scholarly density a first-year essay would drown in, and vice versa? And is the genre right: primary sources carry data and voices, secondary sources carry context and synthesis, and citing one where the argument needs the other produces that hollow "the research shows" sentence with nothing behind it.
Illustratively: for an essay on whether a city's bike-share program reduced traffic, a national report on cycling trends is currency-perfect and authority-perfect and still off-target — the program-level data is the relevant unit. When your source list keeps resolving into "the same overview cited five different ways," that is relevance debt, and an organized literature review outline is the structural cure: each slot in the outline names the job a source must do before you go hunting for sources to fill it.
Authority: credentials, track record and venue
Authority questions get answered badly in both directions. The gullible direction: a title, an institutional logo or a ".edu" domain treated as proof of expertise. The cynical direction: every source discounted because the author is not famous. Both skip the actual work, which is checking three separable things.
Credentials are what the author claims — degrees, posts, affiliations. They are claimable in seconds, so they open the investigation rather than end it: search the name with the institution, and see whether the institution's own pages confirm the person and the role. Track record is what the author has done in this specific area — prior work on the same question, cited by people other than themselves. A Nobel laureate writing outside their field has authority to spare and expertise to check; the reverse happens too, which is why the two travel separately. Venue is the reputation being borrowed: a journal your library's databases index, a university press, a government statistical agency, a newsroom with a corrections page. An unknown venue is not automatically bad, but the burden of proof moves to the source.
Illustratively: a strong profile is a named researcher whose staff page lists the exact research group, publishing in a journal your database carries. A weak one is "Staff Writer" with a first initial, no bio page, on a site whose about section is a contact form. The scorer's four authority questions weight this category highest (24 of 100) because every other test inherits from it: an unaccountable author can fabricate dates, citations and disclosure, and nothing on the page will contradict them.
Accuracy: what the source does with its evidence
Accuracy is the category students audit least and professors audit first. The page-level tells are learnable in one pass: Does the source cite where its numbers come from — a named study, dataset, or agency — or does it say "studies show" and hope? Was anything reviewed by eyes that could say no: peer review, an editor, a fact-check desk? And do the details cohere internally, because errors in dates and arithmetic predict errors in judgment?
Illustratively, a weak accuracy profile: an article asserting that "80% of employers prefer" something, citing no study at all — the figure is decoration. A strong one: a report whose each key figure links to a named dataset, which states its own limitations in a section a hurried reader would skip, and whose numbers reproduce when you follow the links. The strong source is not one without errors; it is one whose errors are checkable.
When the source under review is a study rather than a story about a study, accuracy lives in the methods section — design, sample, measures, limitations. If that section reads as vague, the accuracy questions above will not save it; the qualitative case study methodology guide shows what a competent methods section actually contains, and the research paper rubric breakdown shows how graders reward cited, verifiable evidence when they score your own work.
Purpose: why does this page exist?
Purpose asks the question the other four avoid: what is this source for? To inform, to persuade, to sell, to entertain, to give the appearance of informing while selling. The answers live in the seams: the about page's mission line, the advertising load, whether "sponsored" or "partner content" labels appear, who funds the operation, and whether the piece's conclusion conveniently requires the reader to buy, donate or vote.
Purpose is also the category students most want to turn into a disqualifier, and that is a mistake. Advocacy sources are citable — some of the most useful primary documents are pamphlets written to persuade — but they must be cited as advocacy: "the industry group's position paper argues…" rather than "research shows…". The scorer's purpose questions do not ask whether the source has an agenda; they ask whether the agenda is disclosed and whether the informational and persuasive layers are separable.
Illustratively: a health page whose "natural remedies" section ends in a shopping link, with no sponsorship disclosure anywhere, fails both purpose questions — the informational layer exists to route you to the commercial one. The same publisher's page with a visible "we earn commissions" line is still promotional, but now it is honest promotional, and your evaluation sentence can price that in.
SIFT: four moves for web sources
SIFT is a set of moves, not a score: Stop, Investigate the source, Find better coverage, Trace claims to the original. Caulfield designed it for the speed of the open web — the whole sequence can run in the time a checklist takes to open — and it works because each move happens away from the page being judged.
Stop is the metacognitive move: when a headline spikes an emotion, notice it, and ask whether you know this source and whether this is where the claim should come from. Most bad citations are planted in the moment a strong feeling meets a confident sentence. Investigate means learning what the source is before trusting what it says — new tab, search the publication's name, read what critics and librarians and prior readers have written about it. Find better coverage reframes the goal: you are not grading this page, you are answering your question, and the best available answer might live somewhere else entirely. Trace follows quotes, figures and studies back to their origin, because a claim can be accurately quoted and still be a distortion of its context — the original paper may be about a subpopulation, a preliminary result, or a hypothesis the authors themselves later revised.
The scorer's SIFT tab gives you the four moves as a tappable checklist. Use it on every web source during gathering; use the CRAAP score on the two or three survivors that might anchor a paragraph. Fast filter first, deep audit second — that ordering is the whole method.
Preprints and predatory journals: two traps around peer review
The accuracy questions lean on peer review, so two boundary cases deserve their own section. Preprints are legitimate research shared before review — common in fast-moving sciences, and often the first public word on a new result. They are citable, but they are unreviewed: cite with the caution stated, and check whether a peer-reviewed version has since appeared, because the published edition is usually revised, sometimes in its conclusions. Treating a preprint as settled science is the error; treating it as radioactive is the opposite error.
Predatory journals are the inverse problem: outfits that mimic peer review to collect fees, publishing anything that pays. They are built to pass vertical inspection — real-looking DOIs, editorial boards, impact-sounding names — so the checks that work are lateral ones. Is the journal indexed in the databases your library licenses? Does the editorial board list people who confirm their own membership on their institutional pages? Is the review timeline advertised in days where honest review takes months? A DOI proves a document exists, not that anyone vetted it.
The scorer catches the crude versions of both: a preprint answers the review question with an honest "Partly," and the flag text tells you to look for the published version. A predatory journal usually trips authority and accuracy together — unknown venue, uncheckable reviewers, guaranteed-acceptance turnaround. When both categories crater on a journal-shaped source, believe the pattern.
Evaluating AI-generated and AI-summarized content
AI text stresses every category at once. Currency looks perfect — the text was generated seconds ago — while carrying the information-age of its training data, which the reader cannot see. Authority is genuinely ambiguous: no author exists to investigate, so the venue hosting the output inherits all the scrutiny. Accuracy is the hard one: generated prose can state nonexistent studies with perfect fluency, and fluent confidence is not evidence. Purpose is usually to answer you, which sounds harmless until the answer is confidently wrong.
Three habits cover most of the ground. First, never cite an AI summary when you can cite what it summarizes — if a chatbot describes a study, find the study, and if the study does not exist, you have learned the summary's value. Second, verify every quote and figure against a source that existed independently of the conversation; invented citations fail this check in seconds. Third, treat whole AI-written web pages as sources of last resort: they pass surface currency by construction and surface authority by design, which is precisely why the SIFT moves — investigating the publisher rather than the prose — matter more, not less.
One boundary stays fixed regardless of the score: if your course has rules about AI assistance, this tool evaluates sources, not authorship, and no credibility score addresses who wrote what. That question lives in your syllabus, and the honest answer is the one you already know.
The source-evaluation sentence your professor wants
Annotated bibliographies and literature reviews ask you to show the evaluation, and the difference between a passing and a strong entry is usually one sentence's architecture. The weak version names a verdict without a reason: "This is a reliable source." The strong version names a judgment, a category, a cause and a consequence — it reads like one line of the scorer's report written in prose.
A formula that holds up: judgment + category + because + how to use it. Two invented examples of the shape (illustrative):
- "Peer-reviewed and method-driven, so strong on accuracy; but the sample predates the policy it is used to explain, so cite it for background rather than as the key claim."
- "An industry-funded report — the sponsorship is disclosed, so the data is usable, but I pair every figure with the independent agency's numbers."
Both sentences would score directly onto the scorer's category bars: the first is authority and accuracy up, currency partly; the second is purpose down, accuracy preserved by pairing. That is what "evaluating sources" means in practice — not a verdict, a located judgment with a usage note. The annotated bibliography guide shows where the sentence sits in a full entry, and its rubric section explains why evaluative annotation earns marks that summary annotation cannot.
What this checker cannot do
- It cannot verify anything. The score adds your answers; it never checks the source itself. A confident wrong answer produces a confident wrong score.
- It cannot catch a well-built fake. Checklists audit the page; forgeries are engineered to pass page audits. Lateral reading — the SIFT tab — is the necessary second opinion.
- The weights are a teaching choice. Twenty-four points for authority reflects this page's argument, not a published standard. If your professor's rubric weights differently, their rubric wins.
- A score is not fit. Eighty-five on the wrong question is still the wrong source; relevance is the one category no tool can judge from the outside.
- It does not format citations. Scoring a source and citing it are different jobs — the citation generators on the free writing tools hub handle the formatting side once the source passes.
Every tool on this site states its limits, and this one's are the widest, because the judgment at the center of the page is yours by design. The scorer's job is to make that judgment structured, recorded and arguable — if your score surprised you, it did its work. The study guides hub collects the rest of the research-process guides this page assumes.
Source list strong, deadline still short?
If the sources are evaluated and the writing is what is behind schedule, a quoted, deadline-bound writer can work from your scored source list — your citations, your evaluation sentences, your argument. Get a free price quote and see the terms first.
Quiz: five questions on evaluating sources
Five questions on the CRAAP categories, lateral reading, preprints and the evaluation sentence. Wrong answers explain themselves, and four options keep the guessing honest.
1. What do the five letters in CRAAP stand for?
2. A source carries no date anywhere on the page. What is the best first move?
3. Lateral reading means:
4. A preprint differs from a published journal article because a preprint:
5. Which evaluation sentence would earn marks in an annotated bibliography?
Frequently asked questions
What does CRAAP actually stand for?
Currency, Relevance, Authority, Accuracy and Purpose — the five tests a source has to pass before you cite it. Librarians at California State University, Chico assembled the acronym, and it stuck because it compresses the evaluation criteria librarians already used into questions a first-year student can answer in minutes. The letters are the categories; the questions under each one are where the judgment happens.
Is 80 the magic number that makes a source safe to cite?
No. The bands on this page are judgment aids, not certificates. A source scoring 85 can still be wrong for your argument, and a source scoring 55 can still be exactly the right primary document for one paragraph. Use the score to find the weak categories, use the red-flag list to decide whether the weakness is repairable, and use your own lateral reading to check what no checklist can see.
Can a source score well and still be a bad choice?
Yes — most often on relevance and fit. A peer-reviewed study can score in the high seventies and still be the wrong genre, the wrong scope or the wrong depth for the sentence you need it to carry. Checklists also judge what a source shows about itself; a well-designed fake or a predatory journal that imitates the surface signals can pass every question. That is why the page pairs the CRAAP score with the SIFT moves, which send you off the page to see what other sources say.
How is SIFT different from CRAAP?
CRAAP inspects the source itself — its dates, authorship, evidence and intent. SIFT is a set of moves you make around the source: Stop before trusting a provoking claim, Investigate the source in other tabs, Find better coverage of the claim, and Trace quotes and figures back to their original context. CRAAP runs vertically, down the page in front of you; SIFT runs horizontally, across the web around it. Researchers who study fact-checking found that professionals lean heavily on the horizontal moves — which is exactly what a checklist on the page itself cannot do.
Does this tool store or send my source names?
Neither. The scorer runs entirely in your browser — no account, no upload, nothing leaves your device. Your labels, answers and SIFT ticks are saved in this browser's local storage under the key csw-scc-v1 so a refresh does not wipe your work, and the Reset button deletes them on demand. Clearing your browser data removes them too.
Did the score change how you'll use the source?