How to judge an unfamiliar website: a routine, not an impression

Design tells you almost nothing: a convincing page is cheap. Four questions, asked in the same order every time, tell you most of what matters.

This is the oldest skill on the site and the one that transfers furthest. It applies to a shop, a health claim, a homework source and a news story, and it does not require any technical knowledge, only the discipline of asking the same four questions before forming a view.

One: who is behind it, and can that be checked elsewhere?

Look for a named organisation or person, a postal address, and a way to make contact that is not a form. Then look them up somewhere else: the check is not whether a name appears but whether it exists independently.

The absence of any traceable identity is the single strongest signal, and it is the one a well-made page cannot fake cheaply. A site that expects to be contacted makes it easy; a site that does not, hides it behind a form.

Two: when was this written, and does it say?

An undated page is a page that cannot be judged. On anything that changes (a menu path, a price, a piece of medical or legal guidance) the date is a large part of the value, and its absence is a decision the publisher made.

This is worth teaching children explicitly for homework, because an undated source and a dated one look identical in a search result and are not the same evidence.

Three: what does it want from me?

Every page wants something: a purchase, an email address, an advert viewed, a belief adopted. That is not sinister, and knowing which one changes how you read the rest.

The warning shape is a mismatch between what the page appears to be and what it is for: a review that only ever recommends, a health article that ends at a single product, an information page that cannot be read without an account.

Four: who else says this?

One page asserting something is a claim. The same claim traceable to a named study, a regulator, or a body with a reason to know is different in kind. Following a claim back to its source is often the point at which it dissolves.

For a shop, the equivalent is uneven, dated reviews somewhere the shop does not control. For a factual claim, it is a citation that leads to something real and not merely to another page like the first one.

Teaching it rather than doing it for them

Run the four questions out loud on a site your child picked, in that order. A routine transfers to the next site; a verdict does not, which is why doing the checking for them protects one page and teaches nothing. And for the youngest readers, a web filter takes the accidental encounters out of the picture while the routine is being learned; what filtering can and cannot carry is on the filtering software page.

It also pairs directly with the scam-site tells, since a fraudulent shop fails questions one and four before it fails anything visual; that side is on the scam websites page.

The machine-written page, and why the four questions still work

A great deal of what now ranks for ordinary questions was assembled quickly, at scale, by software. It reads fluently, cites nothing that resolves, carries no date and names no author, and it is often confidently wrong in a way that a careless human page is not.

Nothing about the four questions changes. Fluency was never the evidence, and this kind of page fails questions one, two and four immediately: no traceable publisher, no date, and claims that lead only to other pages of the same kind.

For homework this is the most useful thing to teach right now. The instruction is not to distrust anything that reads well; it is to follow one claim to its source. A page with nothing behind it collapses on the first pull.

What to do

1. Ask who runs it, and check that name somewhere else.
2. Look for a date. An undated page on a changing subject cannot be judged.
3. Name what the page wants from you before deciding what to believe.
4. Follow one claim back to its source and see whether it survives.
5. Do it out loud with your children, in the same order, every time.
6. Treat design quality as no evidence at all.