How to find out what is wrong with your website yourself
The complete DIY audit method, in about ten hours. Plus the sample size arithmetic that says what your own testing can and cannot conclude.
You can audit your own website. Most of what a paid audit finds in week one is machine checkable or askable, and you can do both in an afternoon. The part you cannot do is be a stranger to your own page.
That last sentence is the whole catch, and it arrives at the end, not the start.
Can you audit your own website?
Yes, for the countable half. Broken links, slow pages, missing labels, a form with fourteen fields, no visible price, no proof: all of that is findable without paying anyone. What you cannot do alone is judge whether your page makes sense to someone who has never seen it.
So the honest split is that a DIY audit gets you the mechanical problems reliably and the comprehension problems unreliably. Knowing which half you are looking at keeps you from trusting the wrong result.
What can you check yourself in an afternoon?
Work in the order below, because each pass narrows the next and stops you chasing a problem you have already ruled out. None of it needs a specialist, and the first three steps are the same ones a consultant runs on their first day at a considerably higher hourly rate.
- Crawl the site. Screaming Frog free covers 500 URLs. Fix broken links and duplicate titles before anything else.
- Check speed on a phone, not your laptop. Run PageSpeed five times and take the middle score. One run is noise.
- Split your analytics three ways. By source, by device, by new against returning. If one segment drags the average, that is your answer.
- Count the friction. Required form fields, clicks to buy, surprise costs appearing late, anything that asks before it offers.
- Read the first screen aloud. If it does not say what this is, who it is for and what to do next, nothing below it matters.
Rather skip the afternoon? Revslip runs these checks free.
How do you test it on real people?
Ask five people who have never seen the site to do one specific task while talking out loud. Do not explain anything. Do not answer questions during the test. Write down every place they pause, re-read, or ask you something, because each pause is a sentence your page failed to carry.
The talking out loud part is what makes it work. You are not collecting opinions on the design. You are collecting the moments where the page stopped making sense.
If you only have ten minutes rather than an afternoon, run the 5 second test on your website instead. It answers one question, whether your first screen says what you sell, and there is peer-reviewed evidence on where it stops being reliable.
Every question a tester asks you out loud is a question your page is already being asked silently, by people who will not wait for an answer.
How many people do you actually need?
More than five, and the popular answer is shakier than the people repeating it admit. Nielsen Norman Group's five user rule is the most quoted number in this field. It describes an average, and an average tells you nothing about the sample you personally drew.
Laura Faulkner tested 60 people on one task, logged the 45 problems the full group found, then drew 100 random samples at each size to see what smaller groups would have caught.
- 85% average of problems found by five users, matching the rule
- 55% found by the worst five person sample of the hundred
- 82% worst case once you go to ten people
The average holds. The spread is the problem. A practitioner could draw a group of five, find 55% of the faults, and walk away believing they had found 85%, because nothing in the test tells you which sample you got.
Faulkner's fix is cheap for a founder: run ten instead of five. It roughly doubles your afternoon and lifts the worst case from just over half to four fifths.
What does doing it yourself actually cost?
Call it a day and a half, spread over two weeks, plus the awkwardness of asking ten people for twenty minutes each. That is the real price, and it is worth naming because most DIY guides quietly pretend the time is free.
- 3 hours for the crawl, speed and analytics passes
- 4 hours to recruit and run ten short tests
- 3 hours to write up what you saw and rank it
Against a consultant's day rate that is an easy trade. Against your own week when sales are already down, it is a real decision, and some people will look at ten hours and correctly decide to buy the first pass instead.
If ten hours is the expensive part, start with the free scan.
What will you not find on your own?
Anything that depends on not knowing your own business. You wrote the headline, so you cannot read it cold. You know what the product does, so you cannot notice that the page never says. This is the single reason DIY audits plateau, and no amount of care fixes it from the inside.
The other blind spot is comparison. You know why you are better than the alternative. Your visitor is holding two tabs open and no view about either.
Where our incentive shows We sell an automated audit, so read a post that ends with a limit on DIY audits accordingly. In our defence the method above is complete, with nothing held back for the paid version, and the two studies it leans on are a peer reviewed paper and the very consultancy whose rule we are questioning. Across the 134 businesses Revslip has audited we see the comprehension problems constantly, but those owners came to us already suspecting trouble, so that is a biased sample and not a measure of the web.
Checking the phone view is its own job, and whether an AI website audit checks mobile gives you the ten minute version plus what a tool cannot see.
How do you rank what you found?
By money, not by how annoying it is. Put a number on each problem before you fix anything, or you will spend the week on the easy ones. The formula we publish is traffic × CVR gap × AOV × mobile weight.
A page taking 5,000 visits a month at 1.1%, against 1.9% on your best page, with a €60 average order, is 40 sales and about €2,400 a month. Do that for each finding and the list sorts itself.
When hiring someone is the right call
Skip the DIY route when the question is about the business rather than the page. If you cannot say who the product is for, or nobody has told you why they bought, testing the page answers a question you have not reached yet.
Also skip it when you have already done this once and the obvious problems are gone. A second self audit of a clean site returns very little, and the split between what machines and people catch covers where the remaining money sits.
Questions people ask before auditing their own site
These come up in roughly this order, and all three are really asking whether the DIY version is good enough to act on.
Which free tools should I use for this?
Fewer than you think, and they are not interchangeable. The four families of audit tool covers which one answers which question, and what the free ones actually check covers the gaps.
Can I just ask friends instead of strangers?
Only for the mechanical steps. Friends know what you do, which reintroduces the exact bias the test exists to remove. Recruit people who cannot already describe your business.
What would a paid audit cost if I gave up?
Published bands run from free scans to five figures. What a CRO audit costs sets out where each tier stops being worth the money.