Website AI Audit
Score your website across 25 AI-readiness criteria: whether AI systems can read your pages, conversion and lead capture, support and self-service, and operations. Returns a prioritised roadmap, not a feature list.
One email, no spam. Unlocks every gated tool on this site.
Score your site against 25 AI-readiness criteria and get a roadmap ordered by what pays back first. Half the list is about features you could add. The other half is about whether AI systems can read the site you already have, which is the half most audits skip.
What This Audit Does, And Who It Is For
The audit reads your site and scores it against twenty-five criteria grouped into five areas: whether machines can read your pages at all, conversion and lead capture, support and self-service, back-office operations, and governance. Each criterion gets a score from 0 to 4. The output is not a league table. It is a roadmap ordered by what pays back soonest for the least work.
It is built for people who own a site that already earns something and now have to decide where AI fits into it. That is usually a small agency, a professional services firm, a local business with a booking flow, or a founder running the whole thing alone. If your site has no traffic yet, this audit will tell you true things you cannot act on. Fix traffic first.
The audit is deliberately unbalanced in one direction. Most AI-readiness checklists are lists of features you could bolt on: a chatbot, a recommendation block, an AI search box. Those are the easy half. The harder half, and the one this audit puts first, is whether AI systems can read the site you already have. If they cannot, every feature you add is decoration on a page nobody outside your own browser ever sees in full.
The Half Most Audits Skip: Can Machines Actually Read Your Pages
This is the criterion that decides the value of the other twenty-four, so it is worth showing what a pass and a fail look like on real pages rather than describing it.
The test is simple and you can run it yourself. Fetch one of your URLs the way a crawler does, with no JavaScript execution, and look at what comes back. Not the rendered page in your browser. The raw bytes the server sends.
Here is that test run against a live page on this site, /tools/word-counter, on 22 August 2026. The left column is what a machine needs in order to quote, index, or summarise a page. The right column is what actually came back.
| What the crawler needs | Present in raw HTML? |
|---|---|
| The article prose, including tables and lists | Yes, complete |
A page-specific <title> |
Yes |
A page-specific <meta name="description"> |
Yes |
| Open Graph and Twitter tags with a per-page image | Yes, including image alt text |
<link rel="canonical"> |
Yes |
| Breadcrumb trail | Yes |
| Author attribution block | Yes |
| Internal links to related pages | Yes, in the prose and in a related grid |
Eight out of eight. That page is prerendered to static HTML at build time, so the server hands over a finished document.
Now the failure case, which is far more common. A site built as a single-page application with client-side rendering and no prerendering step serves this, near enough, for every URL on the domain:
<!DOCTYPE html>
<html lang="en">
<head>
<title>My Company</title>
<script type="module" src="/assets/index-a3f9c1.js"></script>
</head>
<body>
<div id="root"></div>
</body>
</html>Score that against the same eight rows and you get zero out of eight. One generic title reused across every URL, no description, no canonical, no prose, no links. The content exists. It arrives a second later, assembled in the browser from a JSON API. But the document that left the server contains none of it.
Google is explicit that this is survivable on its own crawler. Its guidance on AI features in Search states that "Google is able to process content within JavaScript as long as it isn't blocked." Rendering happens, and pages built this way do get indexed. The problem is narrower and more awkward than "you will not be indexed". It is that rendering is a second pass with its own queue, that it is not guaranteed to be timely, and that Googlebot is not the only thing reading your site any more. Other retrieval systems and agent fetchers vary in whether they execute scripts at all, and none of them publish a rendering budget you can plan against.
So the honest way to read that eight-versus-zero comparison is this: prerendered HTML removes a dependency. It does not buy you a ranking. It means the correctness of your page does not rest on someone else's renderer running before the crawl expires.
How To Read The Score
Each criterion is scored 0 to 4:
| Score | What it means |
|---|---|
| 0 | Absent. The capability is not there at all. |
| 1 | Present but broken, or present in a way that misleads. |
| 2 | Working at a basic level, no measurement attached. |
| 3 | Working and measured, with an owner. |
| 4 | Working, measured, and improving on a cycle. |
Two rules about the arithmetic matter more than the numbers themselves.
A criterion that does not apply is excluded, not zeroed. A one-person consultancy with no support inbox should not be punished for scoring nothing on ticket deflection. The denominator shrinks instead. An audit that zeroes inapplicable rows produces a low total for a healthy site, and low totals push people towards buying things they do not need.
A score of 1 is treated as worse than a 0. This is the counter-intuitive one and it is the rule that changes the roadmap most often. A missing FAQ page costs you the queries it would have answered. An FAQ page full of answers that contradict your current pricing costs you the queries plus the trust of everyone who reads it. A chatbot that is not there is neutral. A chatbot that confidently gives wrong opening hours is a liability with a support cost attached. Broken beats absent for repair priority, every time.
The Twenty-Five Criteria
Here is the full list, counted. The previous version of this page announced twenty-five criteria and then listed thirteen, which is the sort of thing that survives on a page for a year because nobody counts. I counted.
Machine readability (6)
- Page content present in the raw HTML response, without JavaScript execution
- Per-page
titleandmeta description, unique across the site - Canonical URL declared, self-referencing and absolute
- Structured data present and valid for the page type
- Sitemap complete, and every URL in it resolving to a prerendered page
robots.txtnot blocking the resources needed to render the page
Conversion and lead capture (6)
- A primary action identifiable on every commercial page
- Enquiries captured to a store you own, not only an inbox
- Enquiry source attributable to the page and query that produced it
- Response time to a new enquiry measured
- Follow-up sequence exists and runs without a human remembering
- Booking or scheduling reachable without an email exchange
Support and self-service (5)
- The questions you answer most often written down somewhere public
- Those answers reachable from search, not buried in a portal login
- A self-service path for the top repeat request
- Escalation to a human always available and obvious
- Support volume by topic recorded, so the list can be re-ranked
Operations (5)
- Repetitive internal handoffs identified and counted
- At least one of them automated end to end
- Data moving between tools without a manual export step
- Failures surfacing to a person rather than failing silently
- A record of what each automation is for, so it can be retired
Governance (3)
- Someone named as responsible for what the automated systems say
- Public claims traceable to something you can evidence
- A review cadence, so the site does not drift out of date
Governance looks like the soft category and it is the one that produces the ugliest failures. An assistant confidently quoting a price you stopped charging eighteen months ago is a governance failure, not a technical one.
Common Mistakes
Buying a chatbot as the answer. A chatbot is a delivery mechanism. If the answers do not exist in writing, the bot invents them, and now your wrong answers have a friendly interface. Write the answers first. The FAQ page is the asset. The bot is a wrapper on the asset.
Scoring the feature rather than the outcome. "We have a contact form" is not a 3. A contact form nobody watches on a Friday is a 1. The question is never whether the thing exists. It is whether anything measurable happens after someone uses it.
Confusing AI-readiness with AI features. These are separate axes. A plain, fast, well-marked-up site with no AI features on it at all is highly AI-ready: retrieval systems can read it, quote it, and cite it. A site stuffed with AI widgets that serves an empty div to anything without a JavaScript engine is not ready at all, whatever the feature list says.
Chasing tactics the platforms have said they ignore. This is where most of the current advice fails, and it is worth quoting the primary source rather than paraphrasing it. On dedicated machine-readable files, Google's own guidance says: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them." On structured data specifically: "Structured data isn't required for generative AI search, and there's no special schema.org markup you need to add. However, it's a good idea to continue using it as part of your overall SEO strategy." And on rewriting your pages into machine-friendly chunks, the same guidance says there is no ideal page length and no chunking requirement. None of that means structured data is worthless. It means the case for it is rich results in ordinary search, which is a real and sufficient reason, and not a hidden AI-citation lever.
Missing the eligibility gate that sits outside your site. Two conditions have to hold before a page can appear in generative AI features. The page must be indexed and eligible to be shown with a snippet, and separately the site has to be included in Search generative AI features in Search Console. The second is an account-level setting. No amount of on-page work substitutes for it, and it is worth checking before concluding that your content is being ignored.
Auditing once. A score is a photograph. Sites drift, platforms change what they show, and a page that scored 4 in March can be quietly wrong by September because the underlying facts moved. The review cadence in criterion 25 exists because of this, and it is the criterion most often skipped by people who score well everywhere else.
Frequently Asked Questions
How long does the audit take to run?
The automated pass takes a couple of minutes. Reading the result properly takes longer, because the useful part is not the score but the ordering, and the ordering only makes sense once you have matched it against what you already know about where your enquiries come from.
Do I need to fix everything on the list?
No, and trying to is the wrong instinct. The list is ordered so that the first three or four items carry most of the value. Machine readability failures come first because they are cheap to fix and they gate everything downstream. Governance items come last in sequence but they are the ones that stop a fixed site from breaking again.
Will this tell me my rankings or my traffic?
No. It reads the site, not your analytics. It can tell you that a page is invisible to a crawler that does not run JavaScript. It cannot tell you what that page currently earns. Pair it with your own Search Console data, which is the authoritative source for what is actually being served and clicked.
Is a low score bad news?
Usually the opposite. A low score on machine readability means there is a large, mechanical, one-off fix available. A high score everywhere with flat results is the genuinely difficult position, because it means the constraint is somewhere the audit cannot see.
Why does this tool ask for my email when your other tools do not?
Because the output is a report rather than a number, and it gets sent to you rather than rendered in the page. That is a real trade-off and not one I am entirely happy with: gating anything costs you the visitors who will not hand over an address, which is most of them. The calculators and developer utilities on this site are open for exactly that reason. If you want the criteria without the report, the list above is the whole list, and you can score yourself against it in an afternoon.
Can I just run the eight-row raw HTML test myself?
Yes, and you should, because it is the single highest-value check on the list. Fetch your own URL with any tool that does not execute JavaScript, and read what comes back. If you see your article text, you pass. If you see an empty container and a script tag, you have found the thing worth fixing first.
Where This Audit Stops, And What Comes Next
The audit is a reading exercise. It tells you which of the twenty-five things are missing, broken, or unmeasured, and it puts them in an order. What it cannot do is the part that follows, because that part is not a checklist. Deciding that criterion 19 should be your first automation, choosing which handoff to take end to end, and building the thing so that it fails loudly instead of silently are all judgement calls made against your specific tooling.
That is the work I do. Most of it is unglamorous: an automation audit of the internal handoffs rather than the site, then wiring the two or three that actually cost hours into n8n or Make, then a voice or chat layer on top only once there is something reliable underneath it for it to talk to. Where the answers need to come from your own documents rather than a prompt, that means a retrieval system, and the practical guide to building one covers how that is put together.
Two other tools here are worth running alongside this one. The follow-up revenue calculator puts a number on criterion 11, which is usually the criterion with the largest gap between what a site scores and what the owner assumed. The AI consultant widget is a faster, rougher pass if you want a starting point before committing to the full audit.
If the report comes back and you would rather talk through the ordering than work from it alone, that conversation is free and I do not need you to have fixed anything first.
Want this built against your real numbers?
A 30-minute call to scope the workflow, agent, or automation you actually need.
More lead tools
All tools
Agency Retainer Calculator
Work out what a retainer needs to be to hit a target margin. Accounts for delivery hours, software costs, scope creep, non-billable time and the effective hourly rate you end up with once everything is counted.
AI Agent Cost Calculator
Estimate what an AI agent really costs to build and run each month — model usage, platform fees, development time and maintenance — with a worked example and the metered costs most quotes leave out.
AI Receptionist Cost Calculator
Compare the monthly cost of a human receptionist against an AI voice agent handling the same call volume, including out-of-hours cover, holiday and the calls a person cannot answer while already on a call.
Automation Readiness Assessment
Score your readiness to automate across five pillars, then see which single fix moves you up a band. Includes the API request arithmetic that decides whether an automation is affordable to run.
Automation ROI Calculator
Calculate the return on investment, payback period and net gain on an automation project, software purchase or campaign. Includes a worked example and the common mistakes that make ROI estimates too optimistic.
Business Process Analyzer
Find which of your business processes are costing the most and should be automated first
Have a workflow that's burning hours every week?
Bring me one real bottleneck. I'll tell you whether it's worth automating, and what it would take.