Ask why website audit is important and you get two different answers, depending on which audit the person answering has in mind. One is about whether your site works. The other is about whether it can be found. They share a name, they are often sold by the same people, and buying the wrong one is the most common mistake in this subject.
This page is about the first of those, the site as a thing you built and shipped, and about what an audit of it can and cannot prove. It is deliberately short on statistics, for reasons set out below. If you want the procedure rather than the argument, our guide to conducting a website audit is the run book, and the website audit hub sits behind both.
The phrase covers two genuinely different jobs
Two things get sold as a website audit and they answer to different people, use different instruments and produce different reports.
The first treats the site as a delivered product. Does it load quickly, does it work on a phone, can somebody operate it without a mouse, is it served securely, do the forms submit, do the links resolve, is the content still true. These are properties of the thing itself, and they are true or false whether or not anybody searches for you.
The second treats the site as a search asset. Can the pages be crawled and indexed, is the right page answering the right query, did search behavior change after you acted. Our explainer on why an SEO audit is important takes that side apart properly, and the step by step SEO performance procedure is its run book.
The two overlap on exactly one finding, which is that a page nobody can load is also a page nobody can rank. Beyond that single point they are separate jobs. A site can be flawless as a product and invisible in search. It can also rank well while being slow, fragile and unusable for a portion of the people who arrive.
They get conflated because the same firms sell both and the word audit carries the same authority either way. That is not usually deceptive. It is that the buyer asks for a website audit, the seller has a product with that name, and neither party establishes which of the two problems started the conversation. The result is a report that is perfectly competent and answers a question nobody asked.
Why website audit is important depends on which audit you mean
This is not pedantry about vocabulary. The two audits fail in opposite directions, so the answer to the question changes completely depending on which one is on the table.
A product audit matters most when you already have traffic. Every visitor is arriving at something, and if that something is slow, broken on a phone, or asking for information the user already gave you, the cost lands on every single arrival. Nothing about the search results changes. The loss happens after the click, where a search report cannot see it.
A search audit matters most when you do not have traffic, because it diagnoses why people are not arriving in the first place. Running it on a site nobody can reach is putting effort into the wrong end of the funnel, and running a product audit on a site with no visitors is polishing a shop nobody walks past.
So the useful version of the question is not whether an audit is important in the abstract. It is which of the two failures you are living with, and that is usually answerable in an afternoon from whatever analytics you already have.
Decide which one you need by asking who reads the report
The cleanest test does not involve the site at all. Ask who would receive the finished report and act on it, because the two audits produce work for different people.
If the answer is whoever builds and maintains the site, meaning developers, a design team, or whoever owns the platform, you want the product audit. Its findings are tickets. Compress these images, fix this contrast ratio, this form drops submissions on a slow connection, this template shifts as it loads.
If the answer is whoever owns the content and the search results, you want the search audit. Its findings are a different kind of work, about which pages should exist, what they should say, and which of them is answering a query it was never written for.
Where the honest answer is both, run the product audit first. That ordering is not a preference. Search findings frequently recommend changes to pages, and making those changes on a template that is already broken means doing the work twice. Fix the container, then decide what goes in it.
Why the numbers in most of these articles cannot be used
Almost every article on this topic opens with a percentage. A share of users who abandon a slow page, a share who never return after a bad experience, a share who judge credibility on design. They are quoted confidently and they are usually unusable, for a reason worth understanding rather than just asserting.
The test is simple and you can apply it to any figure, including any figure on this page. Can you name the study, the sample, the population it was drawn from, and the date it was collected. If a citation leads to another article rather than to a study, and that article cites a third, the number has no origin. It has a chain of people repeating each other.
This matters more than pedantry because these figures are load bearing in the argument they appear in. If the only reason to audit your site is a statistic nobody can source, then the case is weaker than it looks, and anybody who checks will find that out. So this page does not reproduce them, including as examples, because a number printed on a page gets remembered by a skimmer long after the caveat around it is forgotten.
The consequences you can actually stand behind
Strip out the unsourceable statistics and there is still a case. It is just a mechanical one rather than a statistical one, and mechanical arguments have the advantage that you can check them against your own site instead of taking them on trust.
- A broken thing fails every time it is used. A form that drops submissions does not drop some of them. It drops all of them, silently, for as long as nobody checks.
- Defects in a template repeat across every page it generates. The cost scales with the size of the site, which is why the same defect is trivial on a brochure and expensive on a catalog.
- Nobody reports the failure to you. Somebody who cannot complete a purchase leaves. They do not open a support ticket explaining which step broke, so the absence of complaints is not evidence of anything.
- Content goes stale without any event to mark it. A price, a policy or a person changes in the business and nothing on the site knows, because publishing is an action and going out of date is not.
None of those require a percentage to be true. They are statements about how software and organizations behave, and each one is checkable on your own site rather than accepted on somebody’s authority.
The weaker version of this argument is worth admitting too. Mechanical reasoning tells you a defect costs something. It does not tell you how much, and anybody claiming to know that figure for your site without your conversion values and your traffic is guessing with more confidence than the situation allows.
Speed, and the thresholds somebody actually published
Speed is where the unsourceable numbers cluster most thickly, which is a shame, because it is also an area where a named organization has published specific targets and said exactly how to measure against them.
Google’s Core Web Vitals documentation sets three. It says Largest Contentful Paint should occur within 2.5 seconds of when the page first starts loading, that pages should have an Interaction to Next Paint of 200 milliseconds or less, and that pages should maintain a Cumulative Layout Shift of 0.1 or less. It also names the measurement rule, recommending the 75th percentile of page loads, segmented across mobile and desktop.
That last part is the half everybody drops, and it is the half that makes the figures usable. A threshold without a percentile is not a target, because an average hides exactly the slow sessions you needed to find. Our explainer on what a URL rating measures covers the same failure in a different metric.
Google’s page experience documentation also connects the two audits at their single point of contact, stating plainly that Core Web Vitals are used by its ranking systems, while answering whether there is a single page experience signal with the words there is no single signal.
Whether the site works for people who are not you
The person who built the site is the worst possible test of it. They know where everything is, they are on a fast connection with a large screen, and they have never once tried to use it the way a stranger does.
So a product audit deliberately tests the conditions the team does not experience. Operate the whole site with a keyboard and nothing else, and see whether you can reach every control and tell where you are. Turn the connection down to something slow and see what appears first and what the page looks like while it waits. Open it on a small screen and try to complete whatever the site exists to make people do.
The keyboard one is not a preference about good practice. The W3C’s Web Content Accessibility Guidelines make it a requirement, stating that all functionality of the content is operable through a keyboard interface without requiring specific timings for individual keystrokes, except where the underlying function requires input that depends on the path of the user’s movement and not just the endpoints. A second criterion covers the other half of the same test, requiring that any keyboard operable user interface has a mode of operation where the keyboard focus indicator is visible. Reaching every control and knowing where you are while you do it are two separate checks, and sites routinely pass the first and fail the second.
Each of those is a genuine check with a yes or no answer, requiring no tool and no license. Google’s own page experience guidance frames self assessment the same way, asking whether pages have good Core Web Vitals, whether they are served securely, and whether content displays well on mobile devices.
What you find this way tends to be embarrassing and cheap to fix, which is the best combination a finding can have.
Content that expired without anybody noticing
This is the least technical part of a product audit and frequently the most valuable, because it needs no specialist to find and it directly contradicts things you are telling customers.
Sites accumulate statements that were true when written. A page describes a service that was retired. A team page lists somebody who left. A guide references a product version that no longer exists, or an interface that has been redesigned since. None of this generates an error, no tool flags it, and the page keeps ranking and keeps being read.
The check is to list every page that makes a factual claim about your business, then ask somebody who works there whether each claim is still true. It is tedious and it is not automatable, which is exactly why it is usually skipped and why the findings are usually real. Tools that promise a content audit are mostly measuring structure and readability, and our comparison of website checkers covers what that class of product actually inspects.
Nothing in this category is a technical defect. All of it is a credibility defect, which costs more and shows up nowhere.
What a website audit cannot tell you
An honest case for something includes its limits, and this one has sharp ones that are worth stating before you commission anything.
- Whether anybody wants what you sell. An audit inspects execution, and a flawless site for a product with no demand is a flawless site with no demand.
- Whether the copy persuades. Readability can be scored, persuasion cannot, and the two get confused constantly because only one of them has a number attached.
- Which fix will earn the most. Ordering by expected value requires knowing what a conversion is worth to you, which is your information and not the auditor’s.
- Whether the design is any good. Contrast and spacing are checkable, and whether the thing looks trustworthy to your particular buyer is a judgment somebody makes, not a measurement.
Anybody selling you an audit that answers all four is selling a strategy engagement with an audit’s name on it. That may be what you need. It is not the same product and it should not carry the same price.
When a website audit is not important
There are situations where the honest recommendation is to spend the money elsewhere, and saying so is the only thing that makes the rest of this page worth reading.
If you are about to rebuild the site, auditing the one you are replacing produces findings against something that will not exist. Audit the new one before launch instead, which is when the findings are cheapest to act on. If the site has almost no visitors, the binding problem is that nobody arrives, and the product audit is measuring a room with nobody in it.
If you already have a list of known defects that nobody has had time to fix, an audit will hand you a longer list and change nothing. The constraint there is capacity, not information, and another document does not create capacity. And if the site is a handful of pages that one person maintains and reads regularly, a formal audit is ceremony. Walk through it with the checks in this page and you have done the same job.
Those four cases are the honest boundary on why website audit is important as a claim. It holds when there is traffic to protect, a change to verify, or an owner to brief. Outside those conditions it is a document produced because documents feel like progress.
How often to run one, without a calendar
Every article on this subject recommends an interval and the intervals disagree, which tells you the number is convention rather than finding. A better rule attaches the audit to events instead, because events are what actually change a site.
Audit before a launch, because that is the last moment a structural finding is cheap. Audit after a redesign or a platform migration, because both change every page at once and both routinely lose things nobody was watching. Audit when a metric moves without an explanation, since an unexplained change is the definition of something you do not understand about your own site. Audit when somebody takes over ownership, because an audit is the cheapest handover document there is.
Between those, a short recurring check beats a long occasional one. A handful of measurements taken consistently will show you a trend, and a trend is what tells you something is degrading while it is still cheap to reverse. Our guide to analyzing a website covers what belongs in that short recurring set.
Telling a real finding from a generated one
Most audit reports arrive long, and length is the thing people mistake for thoroughness. A generated report can produce hundreds of findings in seconds, and the count says nothing about whether any of them matter.
Three questions separate the two. Does the finding name a specific page or element, or does it describe a category. Does it say what happens to a person because of it, or only that a rule was violated. And would fixing it change anything a visitor experiences, or only change a score in the tool that produced it.
A finding that fails all three is not wrong, it is just unranked, and an unranked list is work for you rather than work done for you. Our explainer on checking a site on the web goes through what the automated layer genuinely catches, which is a real and useful set, just a smaller one than the report length suggests.
There is a fair objection to this. A generated report is cheap and catches things a person skims past, so dismissing it wholesale would be wrong. The useful arrangement is to let the tool find candidates and let a person decide which of them are findings, because deciding is the part that needs somebody who knows what the site is for.
The reason this matters to the question at the top of the page is simple. An audit is important when it changes what somebody does, and a report nobody can act on has not audited anything.
What we would check first
We sell audits, so weigh this accordingly and apply the same test to us that this page applies to anybody. Everything in the order below can be done without us, and the first two things need no tool at all.
Complete the thing your site exists to make people do, on a phone, on a slow connection, as a stranger would. Most of what matters surfaces in that one pass and it costs an hour. Then check the pages that make factual claims about your business against somebody who knows whether those claims are still true. Only after those two does measurement help, and then the published thresholds above are a better target than any score a tool invents.
If the answer to who reads the report turns out to be nobody, do not commission one, from us or from anybody else. An audit with no recipient produces a document and no change, which is the same failure this page has described in four other forms.
If you would rather have somebody do that on a specific site, our free website audit is where that is set out, and the report is yours whatever you decide to do next.
