User testing a website: watch what people do

Feedback from people who know you is kind, and it measures the wrong thing. A friend has heard your pitch, knows what you meant, and fills in the gaps on the page without noticing them. A headline that confuses a stranger reads clearly to them. Recruited testers are better, but they arrive with a task in hand, hunting for the thing you named. A real visitor from a search has a question and no task at all. Neither group tells you where the stranger gives up, unless you watch them act.

• Show the homepage to someone who has never heard of the product. Ask what it does, before anything else.
• Ask a tester to find one thing, such as the price, and say nothing while they look.
• Note the first click, and how far the tester scrolls before they touch anything.
• Yes or no questions invite polite answers, so rewrite them as tasks the tester can finish or fail.

The people nearest to you are the least useful testers for the first screen. They have read the plan and want to be kind. When a friend says the page is clear, they are often reading it with what they already know, and that knowledge is exactly what a stranger does not have. The same is true of a recruited tester who was told the product in advance, because the briefing sits in their head while they browse.

Choose testers who match the person who would arrive from a search or a post. Someone who runs a similar business, who has the problem, and who has never seen your site is worth more than ten people who admire it. Ask them to tell you, after the first screen, what the product does in their own words. If they cannot, the headline has failed, and no amount of praise changes that.

Ask for an action, not an opinion. Give the tester a task such as finding the price of the smallest plan, then keep quiet. Note where the cursor or finger goes first, whether the tester reads the headline or skips it, and how far they scroll before they touch anything. A tester who clicks the wrong menu three times has told you more than any sentence they could say.

Ask questions after the task, and make them about what happened. What did you expect to find after that click? Where did you think the price would be? Do you know what happens after you press that button? Questions such as do you like the design invite a polite yes, and a polite yes tells you nothing about the page.

A simulated visitor arrives with no history and no briefing, which is the position of a real visitor from a search. It reads, scrolls, clicks and types, and the report gives the step where it stopped, a screenshot and any error on that step. You can ask a simulated visitor a question after its run, and it answers from what it saw on the page. That makes it a fast way to catch the first stop before you invite anyone.

It does not show taste, the reasons a real person has for leaving, or whether they would pay. It cannot tell you what a real stranger thinks of your prices in their own words. Use the simulated pass to find the stop, then take the same task to real people and ask why they stopped there. That way the real testers spend their time on the one question the simulated visitor cannot answer: what they were thinking.

Test a landing page

Every new headline, hero image or pricing change can move the first stop. Run the same first screen task after each change, and before you show the site widely. Behind a login, a paid run can use a test account, so the first-use screens get the same walk as the homepage.