Most AI SEO proposals start with a promise. Ours start with this checklist, because the same twelve things decide whether a page can be cited by ChatGPT, Gemini, Perplexity or a Google AI Overview, and most sites fail four of them. Here is the full list, what each check looks for, how to score it and what to fix first.
What an AI SEO audit is, and what it is not
An AI SEO audit checks whether your website can be found, read, trusted and quoted by the engines that write answers: Google AI Overviews and AI Mode, ChatGPT, Gemini, Perplexity and Copilot. It sits on top of a normal technical and content audit rather than replacing one. If the site cannot be crawled or has no pages worth citing, no amount of AI-specific work helps.
It is not a tool that scores your site out of 100 against a secret formula. The engines do not publish how they select sources, so an honest audit is a structured set of checks against the patterns that cited pages consistently share, each one tied to a fix. These are the twelve we run, grouped into five areas. You can run most of them yourself in an afternoon.
Access: can the engines reach and render your pages?
- 1. AI crawler access. Check robots.txt and any firewall rules for GPTBot, ClaudeBot, PerplexityBot, Google-Extended and Bingbot. Many sites blocked them in 2023 by default and never revisited the decision. Blocking a crawler means its engine cannot retrieve your page live, so it cannot cite you from current content.
- 2. Rendering. Fetch your key pages with JavaScript disabled, or through the URL inspection tool in Search Console. If the main content only appears after scripts run, assume several engines will see an empty page. Server-rendered HTML is the safe default.
- 3. Speed and stability. Core Web Vitals still gate crawl budget and rankings, and rankings feed retrieval. A page that fails the mobile thresholds is less likely to be on the shortlist at all.
Structure: is the answer easy to lift?
- 4. Answer-first openings. For each page, read the first two sentences under the H1 and under every H2. Do they answer the question the heading implies? Engines extract passages; the passage should stand alone.
- 5. Self-contained sections. One idea per heading, with the supporting evidence inside the same section. A page that spreads one answer across five sections is hard to quote accurately.
- 6. Structured data that mirrors the page. Validate your Organization, LocalBusiness, Service, Article and FAQPage markup, and confirm that every fact in it is visible on the page. Markup that claims reviews or prices the page does not show is a trust problem, not a boost.
- 7. Entity consistency. Business name, address, phone, founding year, people and services should match across your site, Google Business Profile, LinkedIn and the major directories. Record every mismatch; each one is a fix.
Quick check for item 4: paste a section into any assistant and ask it to answer the heading using only that text. If it hedges or asks for more, the section needs rewriting.
Content: is there a reason to cite you?
- 8. Original evidence. List what on each page exists nowhere else: your own price ranges, your own before-and-after numbers, your own photos, a process you actually follow. Pages with none are rarely cited, because the engine can get the same generic advice from a thousand sources.
- 9. Authorship and dates. A named author with a real role, a short bio, and a modified date that reflects a real change. Undated or anonymous pages lose to dated, attributed ones on time-sensitive questions.
- 10. Question coverage. Take the twenty questions your customers actually ask, including the awkward ones about price and problems, and check whether a page answers each plainly. Gaps here are the content plan.
Local: do the sources AI answers use for Melbourne queries check out?
- 11. Google Business Profile and reviews. For local questions the engines lean on the profile and on review text. Confirm categories, services, hours and photos are complete and current, and that recent reviews mention the services and suburbs you want to be known for. Thin or stale profiles are the most common local failure we find.
Suburb pages belong here too: a page for a suburb you genuinely serve, with real detail, is citable. A city name swapped into a template is not. See our local SEO service for how we build these.
Measurement: will you know if it worked?
- 12. The measurement set-up. Confirm four things exist: a GA4 view that isolates referrals from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com; a written list of twenty to thirty customer prompts with a monthly citation check per engine; access to server logs filtered to AI crawlers; and a Search Console view of informational queries where impressions rise while clicks fall, the usual sign of an AI Overview answering above you.
Without this item, nobody can say whether the other eleven did anything. It is the first thing we set up on a new engagement and the last thing most audits mention.
How to score it and what to fix first
Score each of the twelve as pass, partial or fail. Then fix in this order, because each layer depends on the one before it: access, then measurement, then structure, then content, then local. A site that fails access or rendering should do nothing else until that is solved. A site that passes everything but has no original evidence has a content job, not a technical one.
Most Melbourne businesses we audit fail items 1, 4, 8 and 12 and pass the rest. The fixes for those four are a few hours of technical work, a rewrite of ten openings, one page of genuine data, and an afternoon in GA4.
If you want the audit done for you, our SEO audit includes these twelve checks, and the AI SEO service implements the fixes. For the background on how AI answers choose sources, read our guide to generative engine optimisation in Melbourne.
