AI Website Audit: What a Model Can Check, What It Cannot, and What to Do With the Report
An AI website audit can crawl every page, flag the mechanical failures and summarise them in minutes. It cannot tell you whether the site does its job. Here is how to run one, what to trust, and what to decide by hand.
On this page
An AI website audit is a review in which a model crawls every page of a site, runs the mechanical checks a person would run on a sample, reads the copy on each page for whether it answers a clear question, and produces a report ranked by severity. It is fast and it is thorough in the way a person cannot afford to be across three hundred pages. What it cannot do is tell you whether the site does its job, because that question is about the business, not the pages.
Direct Answer
An AI website audit is worth running because a model can check every page for the mechanical failures a person would only sample: speed, broken links, missing metadata, heading structure, alt text, structured data, mobile rendering, and copy that does not answer a question. It is not worth trusting on the questions that decide whether a site works: is the offer clear, do the right visitors arrive, does an enquiry get answered. Run the model on the mechanics, read the report, then decide the business questions by hand with the enquiry data in front of you.
Key Takeaways
- A model audits every page; a person audits a sample. Use the model for coverage.
- Mechanical findings are trustworthy. Judgments about the offer are not.
- The audit that matters includes what happens after the form, not only what ranks.
- The report is an input to a decision, not the decision.
What can a model check on a website?
| Area | What the model checks | Trust level |
|---|---|---|
| Performance | Core Web Vitals per page, heavy assets, unused scripts | High, it is measured |
| Technical SEO | Titles, descriptions, canonical tags, headings, indexability | High |
| Structured data | Presence and validity of schema per page type | High |
| Accessibility | Contrast, labels, alt text, keyboard paths | High for the mechanical subset |
| Content | Whether each page answers a clear question, thin pages, duplicates | Medium, needs a person to confirm |
| Enquiry path | Form present, fields, what happens on submit, reply time | Medium, the reply time needs your data |
| Offer clarity | Whether a visitor understands what you do in five seconds | Low, this is judgment |
The first four rows are where the model earns its place. Web Vitals are measured numbers. Google’s structured data introduction defines what valid markup looks like. WCAG specifies contrast ratios. A model applying those across every page finds the failures a person skims past on page forty.
What can a model not judge in a website audit?
Whether the site works. That has three parts, and none of them is on the page.
The first is the offer. A model can tell you the homepage has a headline. It cannot tell you whether a prospect reading it understands what you sell and why it is different. That needs a person who knows the customer.
The second is the traffic. A site can be mechanically perfect and receive the wrong visitors. The audit does not see the search queries, the ad sources or the referrals. That needs the analytics and someone who reads them.
The third is what happens after the form. The most common finding in the audits I run is not on the site at all. The form works, the enquiry arrives, and it waits a day for a reply. No page check finds that. The enquiry data does, and a model can help there too, but as the system that answers the enquiry rather than the one that audits the page.
How to run an AI website audit
- Crawl every page. Get the model the full list of URLs and the rendered HTML of each. Not a sample.
- Run the mechanical checks per page. Performance, technical SEO, structured data, accessibility. Record a severity for each finding.
- Have the model read each page’s copy for one question. What question does this page answer? Pages with no clear answer are flagged as thin.
- Map the enquiry path. Every form, what it asks, where it goes, and, from your own records, how long until a reply.
- Read the report as a person. Sort by severity. Confirm the content flags by reading three of them yourself.
- Decide the business questions by hand. Is the offer clear? Are these the right visitors? Does the enquiry get answered? The model’s report is context, not the answer.
- Fix mechanics first, then decide on the rest. Speed and broken links are a week’s work. A wrong offer is a redesign, covered in AI website redesign.
What an AI website audit is not
It is not a verdict on the site. A report full of red is normal for a site nobody has audited in two years, and most of the red is cheap to fix.
It is not a substitute for reading your own enquiry data. The site’s job is to produce enquiries that get answered. The audit sees the first half.
And it is not a reason to rebuild. Most audited sites need fixes, not replacement. The performance side specifically is covered in how to speed up your small business website.
Summary
An AI website audit gives you coverage a person cannot afford: every page checked for performance, technical SEO, structured data, accessibility and thin content, ranked by severity. It does not tell you whether the offer is clear, whether the right visitors arrive, or whether enquiries get answered, and those are the questions that decide whether the site works. Run the model on the mechanics, read the report, decide the business questions by hand. The wider library is under AI websites; if you want the enquiry path itself mapped, that is workflow discovery.
Frequently asked questions
What does an AI website audit actually check?+
Everything mechanical across every page: speed metrics, broken links, missing titles and descriptions, heading structure, alt text, structured data, mobile rendering, and whether the copy on each page answers a clear question. It summarises the findings by page and by severity.
What can an AI audit not tell me?+
Whether the site does its job. Whether the offer is clear, whether the right people arrive, whether enquiries get answered. Those need a person reading the site as a customer and reading the enquiry data.
How long does an AI website audit take?+
The crawl and the mechanical checks take minutes. The part that takes time is reading the report and deciding what to fix first, which is a person's job.
Is an AI website audit the same as an SEO audit?+
It overlaps. An SEO audit focuses on ranking factors. A useful website audit also covers accessibility, performance, the enquiry path and what happens after the form, because a site that ranks and loses the lead has still failed.
What should I do after the audit?+
Fix the mechanical failures in severity order, then decide by hand whether the site's offer and enquiry path need changing. If they do, that is a redesign question, not an audit one.
Sources
- Web Vitals: web.dev
- Introduction to structured data markup in Google Search: Google Search Central
- Web Content Accessibility Guidelines (WCAG): W3C Web Accessibility Initiative

