How to tell whether a source online is worth believing

Judge a source by what independent people say about it, not by what the page says about itself. The quickest reliable method is to open a second tab, search the name of the site or the author, and read two or three results from somewhere else before you go back. That is called lateral reading, and it beats every checklist of on page warning signs, because everything on the page is controlled by whoever made it. Design, tone, credentials and citation counts are all cheap to fake. Reputation held by other people is not.

Lateral reading: leave the page to judge the page

Most people read a source vertically. They start at the top, work down, and form an impression from the writing itself. That is exactly how a weak source wants to be read, because it gets to set the terms.

Professional fact checkers do something different. Within seconds of landing, they leave. They search the publication name, the organization behind it, and the author, and they read what unrelated parties say. Only then do they come back and decide how much attention the page deserves. The whole move takes about ninety seconds and it settles most questions before you have read a single paragraph of the actual article.

Three searches do most of the work: the site name plus the word funding, the site name plus the word criticism, and the author name on its own. If nothing at all comes back about any of them, that is itself information. A source nobody else has ever mentioned is not necessarily wrong, but it has no track record for you to lean on.

The two minute check, in order

Run these in sequence and stop early when the answer is obvious.

1. Who is behind this? Look for a named organization and a named human. Open a new tab and search both. An about page that describes a mission but never names a person is a real signal.
2. What do they get if I believe this? Subscriptions, clicks, product sales, affiliate commissions, donations, votes, regulatory outcomes. Everyone has an incentive. The question is whether theirs points the same way as the claim.
3. What is the evidence, and can I reach it? Follow one link out of the article. If the trail dead ends in another article on the same site, or in nothing at all, treat the claim as unsupported.
4. When was this written, and has it changed? Look for both a publication date and a revision date. Pages with neither are usually not maintained.
5. Does anyone else say this? One independent source that reached the same conclusion by a different route is worth more than ten sites quoting each other.

Reading the web address helps at step one more than people expect. The real owner of a page is the part just before the first single slash, and everything before that can be set to anything, which is why learning to read a URL properly saves you from lookalike domains that borrow a trusted name.

Who pays, and what they want from you

Funding is not disqualifying. A trade body publishing research about its own industry can still be accurate, and a nonprofit can still be sloppy. What funding tells you is which direction to check hardest, and which questions the source will never ask.

The pattern worth recognizing is the neutral sounding name. Words like institute, council, center, foundation and alliance carry an air of public service and cost nothing to adopt. Search the name with the word funded, or look for an annual report, and you will usually find the answer in under a minute. Where a site is selling something, notice whether the article ever mentions a case where you should not buy. A source that never reaches a conclusion against its own interest is a catalog with paragraphs.

From study to press release to headline

Most bad health, diet and technology claims are not lies. They are a real finding that lost its conditions on the way to you, through three steps.

The paper says something narrow and hedged: an effect was observed in a particular group, under particular conditions, with a stated margin of uncertainty. The press release, usually written by the institution's communications team to attract coverage, drops the hedges and often the population. The headline, written to be clicked, converts a correlation into a cause and a possibility into a fact.

So when a claim rests on research, ask three things. Who was studied, and were they people? How many, and for how long? Did the study measure the thing you care about, or a stand in for it, such as a blood marker instead of an actual illness? If the article will not tell you, the claim is not yet something you can pass on. Beware too of confident references that do not exist at all, a failure mode that has become more common since AI tools started inventing plausible sources complete with authors, journals and page numbers.

Expertise is not the same as a credential display

A row of letters after a name answers a different question from the one you are asking. Qualifications show that somebody was trained in a field. They do not show that the field is this one.

Look instead for three things. Does this person work in the specific area of the claim, rather than a neighboring one? Do other people in that area cite, quote or argue with them? Have they ever publicly been wrong and said so? The last is the strongest signal available, because it is the one thing a promotional source will never do.

Be careful with the reverse error too. A person without formal qualifications who links to primary documents you can check is more useful than a professor who links to nothing. The credential is a shortcut, not the thing itself.

The signals that mean nothing

Several things people treat as proof carry no information at all about accuracy. They are worth naming, because unlearning them frees up attention for the checks that work.

Two of these have got harder recently. Photos of people and places can now be generated on demand, so a friendly team page proves less than it used to, which is why it helps to know what still gives an AI generated image away. Volume is no longer evidence of effort either, because a site can publish a hundred plausible articles a week, and the tells in machine written text are worth recognizing before you treat an archive as a track record.

When the source is an AI answer

Increasingly the page you are judging is not a page. It is a generated summary sitting above the results, with a few citations attached. Treat it as a starting point rather than a source, because the summary can be confident and fluent while resting on a weak page, or on nothing. Open the citations and check that each one actually says what the summary claims it says, using the same five minute routine as checking an AI answer before you rely on it. It also helps to know when a generated answer is the right tool and when a list of links is better: for anything contested, recent or high stakes, you want the links.

What to do when you cannot place a source

Sometimes ninety seconds of searching returns nothing useful. The right response is not to reject the source or accept it, but to hold it at arm's length and say so when you repeat it: this is one site I could not verify.

Then do one of three things. Find the primary document the claim rests on, such as the actual filing, ruling, dataset or paper, and read the relevant paragraph yourself. Or find one genuinely independent source, meaning one that did its own work rather than quoting the first. Or wait. Almost nothing you read today needs a decision today, and a claim that is true at nine in the morning will still be true by evening, when more is known. If the claim is already spreading fast and you are tempted to share it, the tighter routine for checking a viral claim before you pass it on takes about five minutes and catches most of what matters.