An AI website audit is a scan that reads your site and reports what is wrong with its speed, search setup, accessibility and, increasingly, how readable it is to AI assistants. It is a good first pass. It is a poor final answer, because it can list 60 problems and cannot tell you which three are costing you enquiries.
That gap is the whole subject of this post. I run these audits on my own site. They have been useful, and on that same site they have also been wrong, silent or confidently misleading. Here is where, with the numbers.
The short version
- An AI website audit is strong on what a machine can measure: page speed, broken links, missing titles, crawl blocks, missing alt text. Use it for that.
- It is weak on what only a person can judge: whether a page makes sense to a buyer, whether a fix matters to revenue, whether a number it quotes is true.
- A clean score is not a clean site. On my own site, an accessibility scanner reported zero violations on 5 of 6 pages. A manual check found 90 tables with no caption and a keyboard focus ring almost invisible against the background.
- Never act on the list in the order it arrives. Sort it by what it costs you in enquiries, then fix the top three.
- Check every finding against real data: Search Console, GA4 and your CRM. An audit estimates. Your analytics counted.
What is an AI website audit?
It is an automated review of your public pages. A crawler fetches them, runs checks (speed tests, link checks, tag checks, accessibility rules) and an AI model writes up the results in plain language, often with a score and a priority list.
The AI part is mostly the write-up. The measuring is usually older, ordinary tooling. Google’s open-source Lighthouse, for example, is an automated tool that audits performance, accessibility and SEO, and Google’s own documentation tells you to treat a failed audit as an indicator of what to improve, not a verdict. Newer audits add a layer for AI search: can an assistant read your pages, find your answers and cite you.
I used audits throughout the rebuild of my own site, which I wrote up in how I built it with AI. This post is about the audit itself.
What does an AI website audit do well?
It is good at anything that is a fact about the page and does not need context.
- Speed. It can measure how long the main content takes to appear on a phone, which I cover in how to test your website speed. When I audited my own homepage before the rebuild, it scored 46 out of 100 on mobile, with the main content taking 11.3 seconds to show. That was real, repeatable and exactly the kind of thing a tool should catch. (It sits at 95 on mobile now.)
- Broken things. Dead links, missing page titles, duplicate descriptions, blocked pages.
- Consistency. It will check all 200 pages the same way on a Tuesday afternoon, which a person will not.
- A starting list. It stops you staring at a blank page wondering where to begin.
If your site has never been audited, run one today. Several are free. You will learn something.
What does an AI website audit miss?
Everything that needs judgement. Four examples, all from my own site.
It can say “clean” when the site is not
In October I ran an accessibility audit on the new theme before launch. The scanner, axe-core 4.11.4, found zero violations on 5 of the 6 pages I tested. A scan like that looks like a pass.
Then I checked by hand and scanned the content. 90 of 198 data tables had no caption. The keyboard focus ring in the mobile menu had a contrast ratio of 1.39 to 1 against its background. The WCAG minimum for interface components is 3 to 1, so that ring is close to invisible to anyone who navigates by keyboard. The scanner did not flag either problem.
This is not a quirk of my site. In the WebAIM Million 2026, 95.9% of the top million home pages had detectable WCAG 2 failures, 56.1 errors on an average page. Automated tools find a lot. They do not find everything. Deque, the company behind axe, reported in 2021 that its own tools caught about 57% of accessibility issues by volume across more than 13,000 pages. That is a vendor measuring its own product, so read it as a ceiling, not a promise. Even on their own numbers, nearly half the issues need a person.
It can repeat a wrong number
When I rebuilt my site with AI, a stat card on my post-click value page read 43%. The source, Ahrefs’ study of AI assistant traffic, said 0.43%. The leading “0.” was lost when the card was generated. The error made AI traffic look 100 times bigger than it is.
No speed test, link checker or SEO scanner flags that. It is valid HTML on a fast page with a good title. It was caught in a review that checked the page against its source, and I corrected it on 29 October. (The current figure, for October 2026, is 0.46% of visits across 118,671 sites.)
An audit tells you the page is healthy. It does not tell you the page is true.
It can find real problems and still not know their weight
The night before launch, a second AI (OpenAI’s Codex) reviewed my staging site and found seven problems. Form labels that vanished as you typed. A metadata promise the page did not keep. Three dead links. A few smaller things. All seven were real, and fixing them took an evening.
But the tool listed them as a flat set. Only a person could say that the form labels mattered most, because the form is how someone becomes an enquiry. A dead link in a footer and a form nobody can read are not the same size of problem. They look the same in a list.
It uses estimates as if they were counts
Audit and SEO tools often include a traffic number for your site. On mine, Ahrefs estimated more than ten times the Google clicks that Search Console recorded. Search Console counts clicks. The tool models them. Different jobs.
If an audit tells you what you are losing in dollars, ask what it is based on. A modelled range is a scenario. It is not a measurement of money you actually lost.
Can ChatGPT do a website audit?
It can review what you give it. Paste in a page’s text and ask it to critique the structure, and you will get a sensible answer. That is useful for wording and clarity.
What it does not have by default is your data: your conversion rate, which pages your enquiries start on, what a customer is worth to you. Without those it can describe problems. It cannot weigh them. Treat a chatbot audit like any other audit: a prompt for your attention, not a decision.
How should you use an AI website audit?
This is the sequence I use.
- Run the audit and keep the raw findings. Do not skim the score. Read the list.
- Throw out what does not apply. Every audit contains advice written for a different kind of site.
- Sort what is left by revenue. Ask of each finding: does this sit on the path from a visitor to an enquiry? Pages and forms on that path go first.
- Check the top findings against your own data. Open Search Console and GA4. Is the slow page one that people actually land on? If nobody visits it, a slow page can wait.
- Verify any number or claim yourself. If the audit or the AI-written text states a figure, find the source. I would have saved myself the 43% if I had done this for every card.
- Fix three things, not thirty. Then run the audit again in a month and compare.
Notice that steps 3 to 5 need someone who knows your business. That is not a flaw in the tools. It is the division of labour: the machine measures, a person decides.
A free audit I built, and what it does
I built PostCliq, WP Creative’s free website audit, around that division. It checks up to 50 public pages for how fast the site loads for real phone visitors, where visitors get stuck on the way to an enquiry, whether the page messaging supports that journey, the technical setup and whether search engines can reach the site. You do not need a plugin or a WordPress login.
What I care about most is how it reports. Each finding comes with its evidence and a suggested fix, and the top three quick wins come first, so you are not handed 60 items with equal weight. It gives a dollar opportunity range, and it says plainly that this is a modelled scenario rather than a measured loss. If you connect GA4 or Search Console (optional, read-only), it can set real visitor behaviour beside each finding.
It is free, with one standard audit per site every 7 days, and you need a free account. I built it, so weigh my view of it accordingly. Try PostCliq and judge the output yourself, then apply the sequence above to what it finds.
Questions people ask about AI website audits
What is an AI website audit?
An automated scan of your public pages that reports problems in speed, search, accessibility and AI readiness, with an AI-written summary and usually a score. It measures what a machine can measure and leaves the weighing to you.
How accurate is an AI website audit?
Accurate on facts about a page, such as load time, broken links and missing tags. Less reliable on anything that needs context or a source. My own audit missed 90 uncaptioned tables and a near-invisible focus ring, and a generated stat was 100 times too high. Check what matters by hand.
How much does an AI website audit cost?
Many are free, including PostCliq. A human-led audit costs more because it is a person’s time and judgement. I have not seen an independent, dated price source for the Australian market, so I will not quote a range. Ask what you get for the money: a list, or a ranked plan with evidence.
Can an AI audit replace an SEO or web expert?
No. It replaces the slow, repetitive checking. It does not replace the person who decides what matters for your revenue, verifies the claims and knows what you tried last year. An audit is an input to that conversation.
Should I fix everything an audit finds?
No. Fix the few findings that sit on the path to an enquiry, and leave the rest. Sixty equal-looking issues are a prompt to prioritise, not a to-do list.
If you want a second opinion on your audit
Run the audit. Then, if you would like someone to go through it with you and tell you which three findings matter to your revenue and which can be ignored, book a free call. Send the report ahead. I will tell you honestly whether it is worth your time to fix any of it.
