A site audit has one job: find what stops search engines and AI assistants from reaching, reading and trusting your pages, and turn it into a list of fixes in the right order. Most audit checklists stop at Google. This one adds the layer that decides AI answers, and it is ordered so that the checks that block everything else come first.
At the end you will find what happened when we ran it on our own site, attensira.com, using our own product.
How do you run a site audit, step by step?
Gather four sources before you start. Each answers a different question, and an audit built on only one of them misses things.
- A crawler (any desktop SEO crawler): what your site links to, and what each URL returns.
- Google Search Console: what Google indexed, which queries show your pages, and the errors it found.
- Server or CDN logs: what crawlers actually fetched, including AI crawlers that analytics scripts never see.
- The answers themselves: which of your pages ChatGPT, Perplexity, Claude and Google's AI cite for the questions your buyers ask.
Then work down the checklist in order. The order matters: a perfect title tag on a page no crawler can reach is wasted work.
What is on the 12-point site audit checklist?
Twelve checks, in the order to run them. Each names the source that answers it best: using the right one avoids the most common audit mistake, reading a crawler's guess as if it were Google's or an AI engine's record.
1. Can crawlers reach the site?
Check robots.txt for rules that block whole sections by accident, including rules for AI crawlers such as GPTBot, ClaudeBot and PerplexityBot. Check that your firewall or bot protection is not challenging well-behaved crawlers.
- Passes when
the pages you want found return 200 to a crawler with no login and no challenge page.
- Best source
- robots.txt, firewall rules, server logs
2. Is what you want indexed actually indexed?
In Search Console, compare indexed pages with the pages you care about. Look at the reasons pages are excluded.
- Passes when
every page you want to rank is indexed, and what is excluded is excluded on purpose.
- Best source
- Search Console
3. Does the main text exist without JavaScript?
Load key pages with JavaScript turned off, or fetch them with a plain HTTP request. Most AI crawlers do not execute JavaScript, so text that appears only after scripts run is invisible to them.
- Passes when
the heading, the main copy and the links are in the HTML the server sends.
- Best source
- A plain HTTP fetch, or a browser with scripts off
4. Is each page reachable at exactly one address?
Check www versus bare domain, http versus https, trailing slashes and parameters. Every variant should redirect once to the canonical address, and every page should declare a self-referencing canonical. Google says duplicate content is not a spam violation, but it wastes crawling and splits signals.
- Passes when
one address per page, one redirect hop at most.
- Best source
- A desktop crawler
5. Do redirects and removed pages behave?
Find redirect chains, redirect loops and internal links that point at redirects or at pages that return 404.
- Passes when
internal links point straight at live pages, and every removal was a deliberate decision.
- Best source
- A desktop crawler
6. Does the sitemap tell the truth?
Your XML sitemap should list the canonical, indexable pages, with real last-modified dates, and nothing that redirects or returns an error.
- Passes when
sitemap URLs and indexable URLs match.
- Best source
- A desktop crawler
7. Are the on-page basics in place?
One clear title per page that says what it is, a description written for a person, one H1, headings that describe sections, and descriptive alt text on meaningful images. Skip the keywords meta tag: Google does not use it.
- Passes when
no missing or duplicated titles across the site.
- Best source
- A desktop crawler
8. Is the structure shallow and well linked?
Important pages should be a few clicks from the home page and linked from related pages with descriptive anchor text. Orphan pages, which nothing links to, rarely perform. Our site architecture post covers the shape.
- Passes when
no orphans among the pages that matter.
- Best source
- A crawler's link graph
9. Is the site fast and stable on a phone?
Check Core Web Vitals in Search Console for real-user data, then test the slowest templates.
- Passes when
the templates that carry most of your traffic meet the thresholds on mobile.
- Best source
- Search Console Core Web Vitals
10. Does each important page answer its question?
Read each page that should win a query and ask whether its first paragraph answers the question a buyer typed, in a sentence that stands alone. AI answers are built from passages, and a page that leads with a slogan gives the engine nothing to lift.
- Passes when
each key page answers its question in its first 40 to 60 words.
- Best source
- Reading the page as a buyer
11. Is structured data valid and honest?
Validate Organization, Product, Article, FAQ and Breadcrumb markup where you use it. The markup must match what the page visibly says.
- Passes when
no errors, and no markup describing content that is not on the page.
- Best source
- A schema validator
12. Which of your pages do AI answers cite?
This is the check classic audits skip. Read the answers to the questions your buyers ask and list which of your URLs are cited, on which engine. Do not assume they are your best-ranking pages; engines often cite pages that rank nowhere for the query.
- Passes when
you know which pages engines rely on, and none of them is broken, outdated or about to be removed.
- Best source
- The answers themselves, read repeatedly
What is a site audit not?
It is not a ranking report. Positions belong in a separate, regular report; an audit is about causes. It is not a content strategy either: it tells you which existing pages fail, not which new pages to write, which is what a content gap analysis is for. And it is not a score. Many tools roll hundreds of checks into one health number, and the number goes up when you fix trivia. Judge an audit by the blocking findings it surfaced and whether they got fixed.
