Website usability testing: how a founder runs one on real people, free
Website usability testing means watching real people try tasks on your site and writing down where they get stuck. Run it in five steps: pick 3 tasks, recruit 5 testers, watch without helping, rank what broke, then fix and retest. When we checked 378 YC startup homepages on a phone, 122 had no main button on the first screen.
This guide is for founders and small teams who build their own site, often with a coding agent such as Claude Code, Codex or Cursor, and have no researcher and no testing budget. Every step below can be done for $0. Where a paid tool helps, its price is printed with the date we read it. UserProof, our product, is free for 3 sites and 5,000 recorded visits a month, and Pro is $29 a month.
Jump to: what it is · the 5 steps · how many testers · which method · what tests find · a real fix · what to measure · where paid platforms win · sources · FAQ
What is website usability testing?
Website usability testing is watching someone who matches your buyer try to finish a real task on your site, such as finding the price or signing up, and noting every place they hesitate, misread or give up. Nielsen Norman Group describes the core of every session the same way: a facilitator, a set of tasks and a participant, with the facilitator watching behavior and listening rather than helping.
It comes in three forms. In a moderated test you sit with the tester, in person or on a video call, and ask them to think aloud. In an unmoderated test a platform gives the tester the tasks and records their screen while you are not there. The third form uses no recruited testers at all: you record the visits real buyers already make and read where they stop. A good first round combines the first and the third.
How do you run a website usability test, step by step?
You run it by writing a few realistic tasks, watching five matching people attempt them, and fixing what most of them hit before testing again.
- Pick one goal and write 3 tasks as scenarios. Use the buyer's words, not your button labels: "You run two small websites. Find out what this costs for both" tests your pricing page; "Click Pricing" tests nothing.
- Recruit 5 people who match the buyer. Recent sign-ups, people from a community you are part of, or a panel. Friends who already know the product skip the confusion you are trying to see. One screener question filters the rest, for example "Which tool did you last pay for to do this?"
- Watch without helping. Ask them to use their own phone if most of your traffic is on phones, share the screen, think aloud, and record with their permission. When they ask "should I click this?", answer "what would you do if I weren't here?" Open each session with the same short script: you are testing the site, not them; there are no wrong answers; and may you record the screen. Run phone and desktop testers separately, since each sees a different first screen.
- Write down each problem and count it. One line per problem: the task, what happened, and how many of the 5 hit it. Rank by count first, then by whether it stopped the task.
- Fix the top problems and test again with 5 new people. Nielsen Norman Group's advice is several small rounds, because the second round checks the fixes and finds what the first round's problems were hiding.
On a live site you can shorten step 2 for the problems that do not need a voice. Our free homepage grader runs the 10 phone checks from the table below on your own page, and the dead click finder marks every click on a page as worked, slow, dead or error. Neither replaces a person reading your copy.
How many testers do you need?
Five testers per round is enough to find most problems in a qualitative test, and Nielsen Norman Group recommends 3 rounds of 5 over one round of 15. Its model rests on one number: across the projects it studied, a single tester exposed 31% of the usability problems on average. With that rate, the first five testers surface about 85% of the problems, and finding all of them takes about 15.
| Testers | Problems found | What it means |
|---|---|---|
| 1 | 31% | Better than nothing, misses most |
| 3 | 67% | Two thirds after three sessions |
| 5 | 84% | NN/g's "about 85%": stop and fix here |
| 10 | 98% | Few new problems for the time spent |
| 15 | over 99% | The point NN/g uses for "all" |
The five-tester rule is for finding problems, not for proving a change worked. Nielsen Norman Group says a quantitative study needs at least 20 users, and a conversion change measured on live traffic usually needs more. In the Sales.co test below, a 39-point difference against the site's other pages needed about 53 visits a side to call. The sample size calculator gives the number for your own change.
Moderated, unmoderated or real-visit recordings: which should you use?
Use a moderated test to hear why people struggle and an unmoderated platform to reach many testers fast. Once your site has traffic, recordings of real visits show what buyers do when nobody is watching.
| Method | Who you watch | Cost to start | Finds | Misses |
|---|---|---|---|---|
| Moderated, on a video call | 5 people you recruit | $0 with your own recruits and any video call tool | Why they hesitate, in their own words | Behavior when nobody is watching; it takes your time |
| Unmoderated: Lyssna | Recruited testers, self-recruited or from its panel | Free plan $0 with 15 self-recruited responses and 3 seats; Growth from $166 a month; panel responses priced separately | Task success and clicks across more testers | Follow-up questions in the moment |
| Unmoderated: Maze | Recruited testers | Pricing page prints no plan prices; offers "Try Maze for free" and a sales contact | Prototype and live-site tasks, surveys | Plan prices are not stated on its pricing page |
| Unmoderated: UserTesting | Its participant network | Quote only; its plans page says the price varies by users, types and features | Think-out-loud videos at enterprise scale | A self-serve price for a small team |
| Recordings: Microsoft Clarity | Your real visitors | Free; up to 100,000 sessions a day per project, per its FAQ | Where real buyers scroll, click and leave | Why: visitors don't talk |
| Recordings: Hotjar | Your real visitors | Free plan records 5% of sessions, up to 10,000 | Replays, heatmaps (5% of sessions on free) | Why; 95% of visits on the free plan |
| Recordings: UserProof ours | Your real visitors | Free: 3 sites, 5,000 recorded visits a month; Pro $29 a month for 50,000 | Dead clicks, rage clicks, errors on click and slow responses, ranked, with the file and line for your coding agent | Why; sites with no traffic yet |
For a site with a few hundred visits a week, the cheapest strong setup is a recorder on the live site plus five moderated sessions a month. If you are choosing a recorder, the best session replay tools for startups compares the free plans and what each records.
What do usability tests find on startup homepages?
The most common faults are mechanical, and a phone shows them in seconds: on October 28, 2026 we loaded 378 YC startup homepages on a phone and ran 10 checks on each. A third had no main button on the first screen, and over a third of the buttons we could measure were too small to tap.
| Problem | Share | Homepages | Would a 5-person phone test catch it? |
|---|---|---|---|
| Main button too small to tap | 35% | 95 of 268 | Yes, as missed taps |
| No main button on the first phone screen | 33% | 122 of 366 | Yes, as scrolling to find what to do |
| Slow to load | 25% | 96 of 378 | Often, on a phone network |
| Dead clicks | 8% | 29 of 378 | Yes, if a task passes the dead element |
| Images or files fail to load | 8% | 29 of 378 | Sometimes |
| JavaScript errors | 5% | 19 of 378 | Only when the error breaks the task |
| Main button covered | 4% | 11 of 290 | Yes |
| Broken links | 3% | 13 of 378 | Only on the links a task uses |
| Main button does nothing | 1% | 2 of 298 | Yes |
| Wider than a phone | 0% | 0 of 378 | Yes |
A script finds these on every page at once. People find the rest: a headline nobody understands, a price that reads as per user when it is per site, a sign-up form that asks too much. The full results and each startup's score are in the YC homepage report.
What does a fix found this way look like?
At Sales.co, recordings of real visits showed the problem without any recruited testers: visitors from x.com bounced from the home page 76% of the time, against 28% from other sources, and 69% of them were on a phone. On a phone the first screen held the headline and a seven-line paragraph, and both buttons sat on screen two; 136 phone visits produced zero button clicks.
The fix was a one-line subhead and the buttons moved up, on phones only, shipped with Claude Code the same afternoon. In the six days after, bounce from x.com visits fell from 90 of 118 (76%) to 45 of 85 (53%), a 23-point drop with a 95% interval of 10 to 36 points. Over the same days the share taking a next step fell on sales.co's other pages, so the lift is not site-wide drift. Button clicks from phone visitors went from 0 to 4. The Sales.co story has the before and after screens. A five-person phone test could also have shown the buried buttons; the recordings added the size of the loss and the check that the fix worked.
What should you measure in a usability test?
Measure whether each tester finished each task, how long it took and where they went wrong, then add one standard score if you want to compare rounds.
- Task success: did each tester finish, finish with help, or give up? Report it as "4 of 5", not a percentage, at this size.
- Time on task, and the page or element where each tester first went the wrong way.
- The System Usability Scale (SUS), a 10-item questionnaire released by John Brooke in 1986. MeasuringU reports an average score of 68 across about 500 studies, so a score above 68 is above average.
- On live traffic, bounce on the page, how far down visitors scroll, rage clicks and dead clicks. Our dead clicks report shows how often each kind turns up.
Where do paid testing platforms win?
Paid platforms win when you need testers you cannot find yourself, or answers before the site exists. Lyssna says its panel has 690,000+ participants you can filter, priced per use; Maze and UserTesting run tests on prototypes and add AI analysis on their larger plans. UserProof does none of that: it needs a live site with visitors, and it shows what people did, not what they were thinking. For why visitors leave once the site is live, our guide to why users leave your website goes cause by cause.
Sources, read November 16, 2026
- Nielsen Norman Group, Usability (User) Testing 101: facilitator, tasks and participant
- Nielsen Norman Group, Why You Only Need to Test with 5 Users (Jakob Nielsen, 2000): 31% per tester, about 85% with 5, 15 for all, 3 studies of 5
- Nielsen Norman Group, How Many Test Users in a Usability Study?: at least 20 users for quantitative studies
- MeasuringU, Measuring Usability with the System Usability Scale: 10 items, John Brooke 1986, average 68 across about 500 studies
- Lyssna, Pricing: Free $0 (15 self-recruited responses, 3 seats), Growth $166 a month, panel priced separately, 690,000+ panelists
- Maze, Pricing: no plan prices printed, "Try Maze for free"
- UserTesting, Plans: price set by quote
- Microsoft Clarity, FAQ (100,000 sessions a day per project), and Hotjar, Pricing (Hotjar plans are sold on Contentsquare's page): free plans, from our daily pricing ledger read November 16, 2026
- UserProof, YC homepage benchmark: 378 homepages on a phone, October 28, 2026; Sales.co story; plans
FAQ
What is website usability testing?
It is watching people who match your buyer try real tasks on your site, such as finding the price or signing up, and noting where they hesitate or give up. It can be moderated (you watch live), unmoderated (a platform records them) or done on real visits with a session recorder.
How many users do you need for a usability test?
Five per round for finding problems. Nielsen Norman Group's model says five testers surface about 85% of usability problems and recommends 3 rounds of 5 over one round of 15. To measure a rate with statistics it says to test at least 20 users.
Can you do website usability testing for free?
Yes. Recruit five matching people yourself and run the sessions on any video call tool. Lyssna's free plan allows 15 self-recruited responses, Microsoft Clarity records real visits for free, and UserProof is free for 3 sites and 5,000 recorded visits a month (prices read November 16, 2026).
What is the difference between moderated and unmoderated usability testing?
In a moderated test a facilitator watches live and can ask follow-up questions. In an unmoderated test a platform gives the tasks and records the session with nobody present, which reaches more testers faster but loses the follow-up questions.
What is a good SUS score?
Above 68. MeasuringU reports 68 as the average System Usability Scale score across about 500 studies, so a score above it is above average and one below it is below average.