Case study: we pointed SEO Joe at our own website

Before asking anyone to trust SEO Joe with their website, we ran him on our own: seo-joe.com. This is what he found, what we changed, and what happened. All numbers come straight from Joe's checks in October 2026.

The starting point

Joe's website check looks at 21 things, from page titles and speed to whether AI assistants are allowed to read the site. Our first score was 73 out of 100. The site loaded, worked on phones and had good titles — but six checks failed:

• No meta description that fitted the guidelines
• No canonical link (the page didn't say which address is the “real” one)
• No sitemap.xml
• No structured data about the business
• No FAQ
• No llms.txt

Then came the part that stung. Joe asked ChatGPT and Claude the questions a potential customer would ask — things like “Is there an SEO service that helps me get mentioned by AI platforms?” and “Best tool for getting recommended by AI”, 13 questions in all.

Zero. Not once. Instead, the assistants named established tools — Ahrefs, Semrush, Surfer SEO, Clearscope — review sites like G2, and general advice such as setting up a Google Business Profile. (Gemini wasn't part of this first run; it is now.)

What we fixed

We worked through Joe's fix list, quickest wins first:

1. Canonical link on every page, so search engines don't split our ranking between address variants.
2. A sitemap.xml, linked from robots.txt, listing every public page.
3. An llms.txt file — a short, plain-text summary of what SEO Joe is and where the important pages are.
4. Structured data: an Organization block and an FAQPage block in JSON-LD.
5. An FAQ on the homepage, answering the questions we wanted to be named for.
6. More text on the homepage explaining plainly what Joe does — up from 368 to 526 words.

All of it took less than a day.

The result

The live site now scores 88 out of 100. Three checks still fail: the first answer from the server takes about 1.9 seconds (we aim for under 1.5), only 9 of 16 images on the homepage have alt text, and one page shares its title with another. Those are on our list.

The AI answers, however, haven't moved yet. That isn't a surprise, and it's the most important lesson in this case study.

What we learned

A good website is necessary, but not enough

The technical fixes make sure the assistants can read and understand us. They don't make the assistants want to name us. When ChatGPT and Claude search for “best tool for AI SEO”, they read comparison articles, review sites and lists — and we aren't on them yet. The established tools are.

The competition in AI search is different

On Google, we compete with whoever ranks for the same keywords. In AI answers we competed with Ahrefs and Semrush — and with free advice. Knowing exactly who gets named is the most useful thing the AI check gave us.

Measure the trend, not one answer

Answers vary between runs and assistants. We'll keep asking the same 13 questions over the coming months and follow how often we're named.

What we're doing next

• Publishing articles that answer our customers' questions directly — like the ones on this blog.
• Getting listed in AI and SEO tool directories and review sites, where the assistants already look.
• Fixing the remaining three checks.
• Re-running the AI check every month, and updating this case study with the numbers.