What Should an SEO Audit Checklist Check First?

An SEO audit checklist for a contractor website starts with six foundation checks, in this order: one canonical URL, nothing blocked from indexing, an honest sitemap, a homepage that loads in under 2.5 seconds on a phone, HTTPS enforced everywhere, and a clean domain. If any of those fail, nothing you do to titles, content, or links will matter. Most contractor sites we audit fail at least one.
Last updated: October 2026. This is day two of a five-part series on auditing your own website with AI. Day one covers how to run the full three-pass audit prompt.
Can an AI agent actually use your website?
AI agents are starting to visit sites on a customer's behalf — comparing companies and requesting quotes. We run the same agentic browsing checks Google built into Chrome and show you what an agent hits when it lands on your site. No email to run it.
Why the foundation comes before everything else
Most SEO audit checklists you find online start with title tags. That is backwards. A title tag on a page Google cannot reach, or will not trust, is wasted work. It happens more than you would think: a developer leaves a single noindex tag in place after launch, and the site is invisible for months while the owner keeps paying for content.
So the first pass of our audit prompt ignores copy entirely. It asks one question: can Google and the AI crawlers reach this site, read it, and trust it? That question breaks into six checks. Here is what each one looks for, why it matters to a contractor specifically, and how to read the result.
Check 1: Does your site resolve to one address?
Your site can be reached four ways: http://, https://, with www, and without. Three of those should redirect to the fourth in one hop. When they don't, Google sees two or four copies of your site and splits credit between them. We see this constantly on contractor sites that were moved from one host to another, or had an SSL certificate added after launch by someone who never set up the redirect.
The prompt fetches all four versions and reports where each one lands and how many hops it took. Two hops is a warning. A version that serves your homepage without redirecting at all is a failure.
On top of that, every page needs a canonical tag that points at itself. A canonical that points at a different page, a redirecting URL, or a page marked noindex is a quiet way to tell Google "ignore this page." Those mistakes usually come from a plugin or theme setting nobody has looked at in years.
Check 2: Is anything blocked from indexing?
This is the one that keeps us up at night, because it is the easiest to miss and the most expensive. Three places can block a page:
- robots.txt. A single
Disallow: /line, usually left over from a staging site, hides everything. Less obvious: rules that block the/services/folder, or the CSS and JavaScript folders, so Google renders a blank page. - A noindex meta tag. A checkbox in WordPress called "Discourage search engines from indexing this site." Developers check it on staging. Sometimes it stays checked.
- An X-Robots-Tag response header. This one is invisible unless you look at the raw server response. Hosting companies sometimes add it at the server level. The prompt checks for it specifically because almost no other checklist does.
There is a fourth blocker that matters more every month: rules that block AI crawlers. GPTBot, ClaudeBot, PerplexityBot, and Google-Extended are the crawlers behind ChatGPT, Claude, Perplexity, and Google's AI features. Some security plugins block them by default. Block them and you are removed from every AI answer a homeowner gets. We wrote a full post on which AI crawlers a contractor site should allow, and the short version is: all of them.
Check 3: Is your sitemap telling the truth?
A sitemap is a list of pages you want Google to index. Two things go wrong with it on contractor sites.
First, it includes junk: thank-you pages, tag archives, the admin login, pages marked noindex. Google learns the sitemap is unreliable and leans on it less.
Second, every page shows the same lastmod date, usually today or the last deploy date. The site is claiming that every page changed today. Google figured this out years ago and ignores dates it does not believe. When you later make a real update to a service page, the date carries no signal. We found this exact problem on our own site in July 2026, so nobody is immune.
The prompt reads the sitemap, counts the URLs, checks whether the dates vary, and flags any noindex or junk pages it finds listed.
Is your own website costing you leads?
Free scan — speed, mobile, local visibility, and lead capture. Results in about 30 seconds.
Check 4: Is the site fast enough to rank on a phone?
Most homeowners find a roofer, plumber, or painter on a phone, often standing in the room with the problem. Google measures three things about that experience: how fast the biggest thing on screen appears (LCP), how quickly the page responds to a tap (INP), and whether things jump around while loading (CLS).
An AI model cannot run a browser profiler, so the prompt is honest about that. It reports what it can see in the HTML:
- Whether the hero image has explicit dimensions, uses a modern format like WebP, and is marked as high priority.
- Whether images further down the page are set to load lazily.
- How many third-party scripts load in the head of the page. Chat widgets, call tracking, pixels, tag managers, review widgets, fonts. More than five is a warning. Ten or more is common on contractor sites.
- Whether the site sits behind a CDN.
Then it tells you to run the homepage and one service page through PageSpeed Insights yourself and record the mobile scores. Mobile LCP over 2.5 seconds, or CLS over 0.1, is a problem worth a developer's time.
Check 5: Is HTTPS actually enforced?
Having an SSL certificate and enforcing HTTPS are different things. The prompt checks that every fetched page loads over HTTPS, that nothing on the page (an image, a script, an embedded map) loads over plain HTTP, and that the security headers browsers expect are present. The one that matters most is Strict-Transport-Security, which tells browsers never to try the insecure version again.
Mixed content is common on older contractor sites where a gallery of job photos was uploaded years ago with hard-coded http:// links. Browsers show a warning. Homeowners leave.
Check 6: Is the domain itself carrying baggage?
Two things here. First, the top-level domain. .com is standard. A .xyz or .info commercial site starts with less trust and has to earn it back.
Second, exact-match domains. If your domain is literally the keyword, something like best-roofer-anaheim.com, Google has treated that with suspicion since 2012. It is not fatal, but the content has to be clearly better than the competition to overcome it.
The prompt also tells you to check the domain's history on the Wayback Machine. A domain that was something else before you bought it may carry a penalty you never earned.
What the prompt returns
Pass one returns four blocks: a scorecard with pass, warn, or fail for each of the six checks and a one-sentence "what it costs you" for each failure; a short list of stop-everything issues; a numbered fix list ordered by impact with a rough developer time estimate; and a list of things the model could not verify, with the exact Search Console report or tool where you can.
That last block is the part we are proudest of. The prompt is built so the model cannot claim to see your Search Console data. Where a check needs it, the model has to say so. If you run this and get a confident number for something the model could not possibly have fetched, throw the whole answer out and run it again.
Run it yourself
The full foundation pass is a single prompt file. Copy it, replace the site URL and trade, paste it into Claude, ChatGPT, or Gemini with web browsing turned on, and read the fix list.
Get it here: Pass 1: Foundation on GitHub. The repo has all three passes and a README that explains how we built it.
Tomorrow, day three: why your service pages don't rank, and the pass-two prompt that finds out.
If you would rather we ran it
The version we run for clients pulls live Search Console, Analytics, and Ahrefs data, scores fifteen categories, and produces a 30/60/90 day plan. It is the first thing we do inside the Foundation Engine, and the findings drive everything in our contractor SEO service after that. If you want it run on your site, request a free audit and we will send the scorecard.
Free website audit — see what's costing you leads
No email required. Takes 2 minutes. Completely free.

Ryan Goering
CEO & Founder, BaaDigi
U.S. military veteran and digital marketing strategist who built BaaDigi to help contractors generate predictable leads and revenue. 15+ years in SEO, PPC, and AI-powered marketing automation.
Frequently Asked Questions
How long does a foundation SEO audit take with the prompt?▼
About fifteen minutes. Paste the pass-one prompt into Claude, ChatGPT, or Gemini with web browsing enabled, replace the site URL and trade, and let it fetch the homepage, robots.txt, sitemap, and a few service pages. Reading the fix list and running PageSpeed Insights yourself adds another ten. The fixes themselves range from five minutes (a robots.txt line) to a day (redirect cleanup across an old site).
What is the single most damaging foundation problem on contractor websites?▼
A noindex tag or a robots.txt block left over from staging. It removes pages from Google entirely, it is invisible to the owner because the site looks fine in a browser, and it is often discovered only after months of paying for content or ads that point to pages Google will not show. The prompt checks all three places a block can live, including the X-Robots-Tag header most checklists skip.
Can an AI model really check page speed?▼
Not fully. A language model cannot run a browser profiler, so it cannot measure LCP, INP, or CLS directly. The prompt is honest about that: it reports what the HTML reveals (image dimensions, lazy loading, third-party script count, CDN headers) and then tells you to run PageSpeed Insights yourself and record the mobile scores. Any prompt that returns a confident Core Web Vitals number without you running the tool is making it up.
Should a contractor website block AI crawlers like GPTBot?▼
No. GPTBot, ClaudeBot, PerplexityBot, and Google-Extended are the crawlers behind ChatGPT, Claude, Perplexity, and Google's AI features. Blocking them removes your business from every AI-generated answer a homeowner gets. Some security plugins block them by default, so the prompt checks robots.txt for these specific user agents and flags any Disallow rule under them.
What should I do if the prompt reports something it could not have seen?▼
Discard the whole answer and run it again. The prompt is written so the model must mark anything requiring Search Console, Analytics, or Google Business Profile as 'owner must check' and tell you where to look. If it returns a confident indexing count, a manual action status, or a Core Web Vitals score it did not fetch, the model ignored that instruction and nothing else in the answer can be trusted either.
Popular Contractor Marketing Guides
Roofing SEO by City
We serve contractors nationwide — based in Orange County, CA
Explore More Topics
Ready to Get More Leads?
Get a free marketing audit for your contracting business. We'll show you exactly where you're losing leads.
Get Your Free AuditRelated Articles
We Audited 10 Orange County Roofing Websites. What Broke?
Ten real roofing company websites in Orange County, two passes of our AI audit prompt, no guessing. Here is what failed most often, how bad it was, and what to fix first if your site looks like these.
What Does a Local SEO Audit Actually Check?
A website audit misses half the picture. Pass three of our AI audit prompt leaves your site and checks the map pack, your listings, your reviews, and whether ChatGPT names you. Here is what it finds.
SEO for Contractors: Why Don't Your Service Pages Rank?
Most contractor service pages fail for the same three reasons. Pass two of our AI audit prompt reads every page on your site and finds them. Here is what it checks and what we see on real sites.