GannetGitHub

A guide to source credibility: what it is and how to judge it

Source credibility is how far you can believe a source. Learn the theory, the CRAAP test, SIFT and lateral reading, with worked examples and a checklist.

Updated

Source credibility is how far you can believe what a source tells you. It rests on two main things: whether the source knows the subject (expertise) and whether it is honest about what it knows (trustworthiness).

Researchers in psychology and communication use the term to explain why people accept some messages and reject others. Teachers and librarians use it for a practical job: deciding whether a page, paper or video is fit to use. This guide covers both. It explains the theory, the three methods most students are taught, three worked examples, the signals that mislead, and a stopping point so you know when you have done enough.

What source credibility means

Credibility is a judgement made by the person receiving the message. McCroskey and Young (1981) describe the common view that source credibility is the “attitude toward a source of communication held at a given time by a receiver”. So it is not a fixed property of the source. It depends on who is judging, what they know and what they need.

Three points follow from that.

  • Credible is not the same as correct. A credible source is one you have good reason to expect accuracy from. It can still be wrong. Check important claims anyway.
  • Credibility is about a topic. Mike Caulfield makes the point with an example: if you are reading a piece on economics by a Nobel prize-winning economist, you should know that before you read it. The same author has no special standing on medicine.
  • Credibility depends on purpose. Caulfield also notes that a quick check is enough if you only want to read an interesting story. If you are doing deep research, you may need to chase down individual claims.

Source credibility theory

The roots in rhetoric

The idea is old. McCroskey and Young write that Aristotle suggested ethos, the term now usually called source credibility, had three dimensions: intelligence, character and good will. Much modern work is a long argument about whether those three can be measured and how many of them survive.

Hovland and Weiss, 1951

The modern research starts at Yale in the 1940s and 1950s. Carl Hovland and Walter Weiss published an experiment in Public Opinion Quarterly in 1951. They gave Yale history students the same article on four topics, but changed who was named as the author. Some authors were chosen as “trustworthy” and some as “untrustworthy”. On atomic submarines, for example, the trustworthy author was Robert J. Oppenheimer and the untrustworthy one was Pravda.

Their findings were these.

  • Under the conditions of the experiment, neither learning nor remembering the facts was affected by how trustworthy the source was.
  • Opinion change was affected. Immediately after reading, there was a large difference in agreement between the trustworthy and untrustworthy sources.
  • Four weeks later, the authors report that agreement with the two types of source was “almost identical”. People tended to separate the content from the source over time. This is known as the sleeper effect.

The paper itself is about trustworthiness. The two-part idea came next.

Hovland, Janis and Kelley, 1953

Hovland, Janis and Kelley set out their model in the book Communication and Persuasion (Yale University Press, 1953). It is the foundation of most later work. Hocevar, Metzger and Flanagin (2018) summarise it and quote the definitions from page 21. Expertise is “the extent to which a communicator is perceived to be a source of valid assertions”. Trustworthiness is “the degree of confidence in the communicator’s intent to communicate the assertions he considers most valid”.

Put simply: can this source know, and will it say what it knows? McCroskey and Young read the Yale team as having suggested three dimensions: expertness, trustworthiness and intention toward the receiver.

Later measures

Researchers then spent decades trying to measure credibility with rating scales.

  • 1966. James McCroskey published scales for ethos. Per his own measures page, they measured competence and trustworthiness.
  • 1969. Berlo, Lemert and Mertz found “safety” and “qualification” dimensions, roughly equal to trustworthiness and competence, and added a third, “dynamism”. Berlo and colleagues themselves noted the “relative instability” of dynamism, as McCroskey and Young quote them.
  • 1981. McCroskey and Young argued that the historical idea of the construct had not been seriously challenged and that adequate measures of it already existed.
  • 1999. McCroskey and Jason Teven argued that “goodwill” had been wrongly dismissed. They offered an 18-item scale with three scores: competence, trustworthiness and goodwill (also called caring). Competence items include “informed” and “expert”. Trustworthiness items include “honest” and “ethical”. Goodwill items include “has my interests at heart” and “self-centered”. McCroskey notes that the three scores should not be added into one number.

Does credibility change what people believe

A review by Chanthika Pornpitakpan (2004) covers five decades of evidence. Studies of the main effect “seem to indicate the superiority of a high-credibility source over a low-credibility one”. But studies of interaction effects “show source credibility to be a liability under certain conditions”. So credibility usually helps persuasion, with exceptions.

What the theory means for a researcher

Persuasion research asks how credibility changes an audience. As a student you ask the reverse: how do I judge credibility before it persuades me? The three dimensions give three plain questions.

  1. Competence. Is this source in a position to know about this topic?
  2. Trustworthiness. Is it honest, and does it have a record that shows it?
  3. Goodwill or motive. Whose interests does it serve? Who pays for it?

The methods below are ways of answering those questions with evidence rather than a feeling.

Methods people are taught

The CRAAP test

The CRAAP test was written by librarian Sarah Blakeslee at the Meriam Library, California State University, Chico. It appeared in LOEX Quarterly in Fall 2004. The Meriam Library handout describes it as “a list of questions to help you evaluate the information you find”. The letters stand for five criteria.

  • Currency. The timeliness of the information. When was it published or updated? Does your topic need current information?
  • Relevance. The importance of the information for your needs. Who is the intended audience? Is it at the right level?
  • Authority. The source of the information. Who is the author, publisher or sponsor? What are their credentials?
  • Accuracy. The reliability, truthfulness and correctness of the content. Is it supported by evidence? Has it been reviewed? Can you verify it in another source?
  • Purpose. The reason the information exists. Is it to inform, teach, sell, entertain or persuade? Are there biases?

The handout says different criteria will be more or less important depending on your situation. It marks some questions as being for web sources only. One asks whether the URL reveals anything about the author or source, with examples “.com .edu .gov .org .net”.

CRAAP is a good list of things to think about. Its weakness is that most of its questions are answered by looking at the source itself. The next two methods exist because of that.

SIFT

Mike Caulfield, who writes on online digital literacy, describes SIFT as four moves. His 2019 write-up says students have been taught the approach since 2017. It names the moves like this.

  1. Stop. Before you read, ask whether you know the site and its reputation. If not, use the other moves. Stop again if you feel yourself going down a rabbit hole, and remember your purpose.
  2. Investigate the source. Know what you are reading before you read it. Caulfield says taking sixty seconds to find out where media is from will help you decide if it is worth your time.
  3. Find better coverage. Sometimes you care about the claim, not the page that carried it. Ignore the source that reached you and look for trusted reporting on the claim, or scan several sources to see where expert opinion sits.
  4. Trace claims, quotes and media back to the original context. If an article cites a study, find the study and see whether it says what the article says.

His open textbook, Web Literacy for Student Fact-Checkers, gives a version of the same idea with other names: check for previous work, go upstream, read laterally and circle back.

Lateral reading

Lateral reading is the research finding that SIFT builds on. Caulfield credits Sam Wineburg for the phrase “read laterally”. The study comes from Stanford. Sam Wineburg and Sarah McGrew first released it as a working paper in 2017 and published it in Teachers College Record in 2019.

The design was an expert and novice comparison. They sampled 45 people: 10 professional fact checkers, 10 historians with PhDs, and 25 Stanford undergraduates. Each person thought aloud while judging live websites and searching on social and political issues such as bullying, minimum wage and teacher tenure. The paper analyses three of the tasks.

Their findings, as stated in the abstract:

  • Historians and students “often fell victim to easily manipulated features of websites, such as official-looking logos and domain names”.
  • They read vertically, “staying within a website to evaluate its reliability”.
  • Fact checkers read laterally. They left a site “after a quick scan” and opened new browser tabs to judge the original site.
  • Fact checkers “arrived at more warranted conclusions in a fraction of the time”.

The sample is small, and this page relies on the published abstract. The full paper has the task-level detail. Treat the study as evidence about how different groups behave, not as a precise measure of how often.

Why lateral reading beats checklists on the web

A checklist keeps your eyes on one page. That is the wrong place to look when the page is the thing in doubt. Caulfield makes three arguments.

  • Sites describe themselves kindly. In Web Literacy for Student Fact-Checkers he writes that if a site is untrustworthy, what it says about itself is most likely untrustworthy as well. Even a trustworthy site is inclined to paint the most favourable picture of its expertise.
  • Checklists overload people. In a 2019 interview he says that CRAAP asks about an overwhelming number of things, and students then apply it mechanically. Checklist approaches, he says, come from library collection development criteria and “never really worked online”.
  • The gap is context, not understanding. He argues that most people already know that authority and conflicts of interest matter. What fact checkers have are habits that quickly reveal context, such as that a broadcaster is a state propaganda arm.

There is also evidence that it can be taught. Breakstone and colleagues (2021) taught 87 college students lateral reading in an online nutrition course. At the start, 3 of 87 left the original site to consult another source. After four one-hour modules, 67 of 87 did. Students who believed a .org domain made a site reliable fell from 23 to 7. This was one course at one university, so it shows the method can be learned, not that it works everywhere.

Lateral reading answers “who is behind this?” It does not tell you whether a claim is true. For that you still read the content, trace the claim and compare it with other coverage.

Worked examples

A peer-reviewed paper

Peer review is a good sign. It is not a guarantee. A documented case shows why. In 1998 The Lancet published a paper titled “Ileal-lymphoid-nodular hyperplasia, non-specific colitis, and pervasive developmental disorder in children”. PubMed lists it as a study of 12 children. On 6 February 2010 The Lancet retracted it. The retraction notice says that “the claims in the original paper that children were ‘consecutively referred’ and that investigations were ‘approved’ by the local ethics committee have been proven to be false”.

Here is how to judge any paper, using that case as a warning.

  1. Stop. Do not accept a paper because of its journal alone.
  2. Check the record. Search PubMed and the publisher’s page for the title, and for the word “retraction”. In this case the Lancet’s own page lists the paper with the title prefixed “RETRACTED:”, and the PubMed record of the notice names the 1998 paper (PMID 9500320) as the one retracted.
  3. Read what it actually claims. Look at the sample size and method. Does the design support the conclusion? Twelve children cannot show how common something is.
  4. Look at who funds it and who the authors are. Check for declared competing interests.
  5. Find better coverage. Look for systematic reviews and later studies on the same question. One paper is one data point.

The lesson is that credibility is a record, not a label. It can be lost.

A news article

Suppose you find an article headlined “Scientists confirm the city’s tap water is unsafe”. It is on a local news site you have not met before. This example is invented to show the steps.

  1. Stop. Do not share it yet.
  2. Investigate the source. Open a new tab and search for the site’s name together with words like “about” or “owner”. Check whether it is a news organisation with named editors, or a page set up to attract clicks.
  3. Find better coverage. Search for the claim itself. If the water supplier, the regulator or several established outlets report the same thing, that counts for far more than the first article. If only this site says it, be careful.
  4. Trace the claim. Which scientists? Which study? Find it. Check whether the article’s wording matches the original.
  5. Check the type of piece. Is it news, opinion or paid content? Each deserves a different level of trust.

Caulfield’s own example is about a made-up headline that koalas have been declared extinct, from a foundation you do not know. He says your best bet may be to find the best source you can on the topic, not to investigate the foundation.

A site that looks official but is not

Suppose a website presents itself as an institute for a public health topic. Its address ends in .org. It has a clean design, a padlock, a donate button and an About page listing an “advisory board”. Every one of these is easy to get. The checklist questions all seem to pass.

Now read laterally.

  1. Leave the site. Search the institute’s name, plus words such as “funded by” or “criticism”.
  2. Look for independent coverage: news reports, a Wikipedia article with citations, regulator records.
  3. Look up two or three people on the advisory board. Do they hold the expertise the site claims, and where do they work?
  4. Check whether the same people or funders sit behind several similar sites.

This is not only hypothetical. Caulfield describes a natural healing centre with a .org domain and a practitioner with an “NMD” qualification. A 30-second search showed it had been criticised by academics, that the American Cancer Society had debunked its treatment, and that the treatment’s inventor had been sentenced to five years for manslaughter after a patient died. In the Breakstone study, a student judged a climate site by its disorganised layout and its About page and never left it. After the course, the same student left the site, found a Wikipedia article and wrote: “They are funded by Exxon so automatically that raises flags of the reliability of the info.”

Signals that mislead

Some features feel like evidence of credibility. They are not. Here is what each one tells you.

  • A .org domain. RFC 1591, the 1994 document that set out domain use, describes .org as the “miscellaneous TLD for organizations that didn’t fit anywhere else”. Wikipedia’s entry on the domain says it has no registration restrictions and can be used by anyone. A .org tells you almost nothing. Many students still think it does: 23 of 87 in the Breakstone study.
  • The HTTPS padlock. It shows that your connection to the site is encrypted. Let’s Encrypt, which issues certificates, says its certificates are standard Domain Validation certificates. In 2015 it wrote that it was not sure certificate issuance is the right level at which to police phishing and malware sites. The Chrome Security team reported that only 11 per cent of study participants correctly understood what the lock meant, and that “nearly all phishing sites use HTTPS”. It tells you about the connection, not the site.
  • A professional design. Design costs little to copy. In the Wineburg and McGrew study, historians and students were taken in by “official-looking logos”. In the Breakstone study, some students judged a site by its look. Neat layout says nothing about funding or accuracy.
  • An About page. The site writes it. Breakstone and colleagues note that groups craft About pages to reflect positively on their aims, and Caulfield says even a trustworthy site paints the most favourable picture of itself. Use the page to learn what the site claims, then check the claim elsewhere.
  • Links to authoritative sources. Breakstone and colleagues list this as a misconception: links alone do not make a source credible. Click them and see whether they say what the page says.
  • Credentials. A title looks like expertise. In Caulfield’s example the “NMD” looked like a doctor’s degree. Check what the qualification is, who awarded it, and whether it covers the topic.

How to tell you have judged it

Stop when you can say all of the following about a source, in a sentence each, and point to where you found the answer.

  1. Who is behind it. A named person or organisation, confirmed by something outside the source itself.
  2. Why it exists. Who pays for it and what they gain.
  3. Why it can speak on this topic. Expertise shown by record, such as past work, role or training, and not by the source’s own claim.
  4. Where the claim comes from. You have traced it to the original, and the original says the same thing.
  5. What others say. Independent coverage agrees, or you know who disagrees and why.
  6. Whether it still stands. It is current enough for your purpose, with no retraction or correction.
  7. How you will use it. For example: “I will cite this for the 2024 figure but not for the explanation.”

If you cannot answer one of these, that is the next thing to search for. If you can answer all seven, stop. More checking has a cost. Caulfield’s advice is to match the depth of your check to what you need the source for.

How Gannet scores a source

Gannet is an open-source deep research tool that DBHQ is building in C#. It is in development and has not been released. The plan is that, given a question, it plans the research, searches the web, weighs each source and writes a report in which every claim is cited to a source the run actually fetched. The repository is at github.com/dbhq-uk/gannet (MIT licence).

One part already exists in merged code: a score from 0 to 100 worked out from a source’s web address alone. It lives in CredibilityScorer.cs, with its numbers in credibility-rules.json. A source starts at 50. Then it gains or loses points:

  • a domain on the trusted list, such as .edu, .ac.uk, .gov or bbc.co.uk: +30
  • HTTPS: +5, or plain HTTP: -10
  • a suspicious top-level domain (.xyz, .tk, .ml, .ga, .cf, .gq) or a suspicious domain (bit.ly, tinyurl.com, t.co, blogspot.com, wordpress.com): -20
  • four or more labels in the host name, when the domain is neither trusted nor suspicious: -5
  • a path containing /papers/, /research/ or /publications/: +10

A score of 70 or more is high, 40 to 69 is medium and anything lower is low. The code matches whole domains, so notcnn.com does not get the bonus meant for cnn.com. According to its tests, https://www.bbc.co.uk/news/article (external site) scores 85 and http://free-stuff.xyz/ (external site) scores 20.

A domain-and-path score is a first filter, not a verdict. It cannot see who funds a site, whether a paper was retracted or whether a claim is true. A good post on a blogging platform will score low and a weak page on a trusted domain will score high. The methods in this guide are what a person does after the filter.

Frequently asked questions

What are the 5 criteria for source credibility?

The most taught set is the CRAAP test: currency, relevance, authority, accuracy and purpose. Sarah Blakeslee wrote it at the Meriam Library, California State University, Chico, in 2004. It is a good list of questions, but it mostly looks inside the source. For web pages, add lateral reading: leave the page and check what others say about it.

What makes a source not credible?

Usually one of these: nobody accountable is named, the author has no record on the topic, a hidden interest drives it, the claims cannot be traced to evidence, or reliable sources disagree with it. A paper can also lose credibility after publication if it is retracted. Signs such as a poor design are weaker clues than any of these.

Is Wikipedia a credible source?

Treat it as a starting point, not a final citation. Wikipedia’s own guidance says caution is advised when using it as a source, and that many academic institutions do not accept it in research papers. It is useful for background and for following its cited references. Cite those original sources instead, after you have checked them.

Is a .org website credible?

Not because of the .org. RFC 1591 describes it as a miscellaneous domain, and Wikipedia’s entry says anyone can register one. Look at who runs the site, who funds it and what independent sources say about it. In one study, 23 of 87 college students believed a .org domain made a site reliable.

How do you quickly check if a website is credible?

Use SIFT. Stop, then investigate the source by opening a new tab and searching for its name and funding. Next, find better coverage of the claim, and trace any study or quote back to the original. Caulfield says that about sixty seconds spent finding out where media comes from helps you decide whether to read on.

What is source credibility theory?

It is the research idea that people accept a message more or less readily depending on who they think sent it. It began with Hovland and colleagues at Yale. Hovland, Janis and Kelley (1953) named two parts, expertise and trustworthiness. Later work by McCroskey and Teven (1999) added goodwill, meaning whether the source has your interests at heart.

Sources

All pages were read on 21 October 2026. Copies of the text relied on are saved in site/drafts/research/source-credibility/.