AI SEO agent case study · our own site
Every skill our AI SEO agent has runs on alchemyleads.com before it touches a client site. Six months of that work moved our Ahrefs health score from 74 in April to 94 in September and 96 in October. These are the problems it caught along the way.
The short answerThe agent audits our site from Ahrefs, Search Console, GA4 and AI visibility data, finds what’s broken and ranks it by impact. Our team turns its findings into detailed instructions, approves each change and ships it. The agent found most of the errors below. Some were invisible in a normal site audit, like a lead form whose submissions GA4 hadn’t recorded for 28 days.
Inputs
What the agent works from
An agent is only as good as the data it can see. On our own site it sees nearly everything.
Technical
Ahrefs and our own crawls
Site Audit exports, plus the agent’s own crawls and manual checks on the live site after every fix.
Search
Search Console and SEOGets
Every query and page, including the AI Mode prompts that now show up in Search Console, with trends over time.
Outcomes
GA4 and PostHog
Traffic, AI-assistant visits and key events like form leads, so fixes are judged on revenue signals, not rankings alone.
AI visibility
Searchable, GeoToolbox and QueryBurst
Which buyer prompts mention us or competitors in AI answers, near-misses where we’re almost cited, and how conversations fan out.
What it found
Seven things the agent caught on our site
- Form leads GA4 never recordedGA4 showed 0 lead events in 28 days, even though the contact form was live. The agent traced it to HubSpot’s newer form embed, which announces a submission with a different browser event than the one our tracking listened for. We added a listener for the new event, so each lead now reaches GA4, PostHog and our tag manager with its first-touch source. We tested it with simulated events, never with real form submissions.
- 195 pages linking through redirectsOld service URLs had been redirected, but 195 pages still linked to them, wasting a redirect hop on every click and crawl. One site-wide fix rewrites those links to their final URLs as each page is served.
- 48 legacy service pagesOld service pages competed with their newer versions. They now redirect, with canonical tags pointing at the current pages.
- A service page with no schemaOur service schema comes from a map of URLs to descriptions. One service page had moved and its new path wasn’t in the map, so it quietly had no Service markup at all. The agent caught the missing path.
- Dead links baked into the themeFour links in a theme-level widget pointed at pages that no longer exist. Theme files are locked on our host, so they’re corrected on output instead.
- 311 AI conversations in Search ConsoleThe agent separated 311 AI Mode and AI Overview prompts from 10,486 normal queries, including one prompt with 866 impressions that no page on our site answered. That’s now our public AI Query Classifier.
- Pages with impressions but no clicksEach month the agent ranks pages with high impressions and almost no clicks, then pulls every query and AI prompt reaching them. In October that led us to rewrite four pages: CRO services, what an SEO platform is, content mapping tools and the Apple search engine.
The lead-tracking fix: a few lines that put every form lead back in GA4.
Agent plus team
How the work splits
The agent
Finds and ranks
Runs the audits, reads the data, finds most of the errors, groups them by cause and drafts the fix plan. It does much of the strategy work.
The team
Directs and approves
Gives the agent detailed instructions for each implementation, approves the changes, edits the drafts and decides priorities.
Together
Ships and checks
Every change is checked on the live site, and each lesson becomes a reusable skill for client sites.
One judgment call shows why the team stays in charge. Ahrefs reports our Cloudflare email protection links as 404s. The agent flagged them, and we kept the protection on anyway, because blocking email-harvesting bots matters more to us than a perfect report. The agent now leaves those links out of its audits.
Same agent, client site
Now running on client sites
The skills proven here now run for clients. On a specialty coffee equipment retailer’s Shopify store, the agent traced 1,186 structured data errors, 12 404s and 155 links to broken pages to four root causes, and our team cleared all of them in one day. Read the Shopify case study.
FAQs
Straight answers
What does the AI SEO agent do on your own site?
It audits alchemyleads.com from Ahrefs, Search Console, GA4 and AI visibility data, finds technical and tracking problems, ranks them by impact and drafts fixes. New skills run on our site before any client’s.
Did the agent make the changes itself?
Partly. The agent found most of the errors and drafted the plans. Our team gave it detailed instructions for each implementation, approved every change and checked it on the live site.
How did the Ahrefs health score go from 74 to 96?
By fixing the causes behind errors rather than single URLs: redirected legacy pages, links rewritten to their final URLs, schema gaps closed and dead theme links corrected. The score went from 74 in April to 94 in September and 96 in October 2026.
Why weren’t form leads showing up in GA4?
HubSpot’s newer form embed signals a submission with a different browser event than older embeds. Our tracking listened for the old one, so GA4 recorded 0 leads in 28 days. Listening for the new event fixed it.
Can the same agent work on my site?
Yes. Every skill proven on our site runs for clients on WordPress, Shopify and Webflow, with the same approvals and undo history.
Your site next
See what the agent finds on your site
We’ll run the same audit on your Search Console, GA4 and crawl data and show you what to fix first.
Suggested
The Cost of Inaction in Manufacturing Lead Generation: What Five Missed RFQs a Month Cost in 15 Industries
What five missed RFQs a month cost in 15 manufacturing industries, with customer lifetime values, a conservative 15 to 20% close rate, and a cost of inaction calculator.
October 27, 2026
Build an AI Crawler Monitor with Workers, Analytics Engine and Cloudflare AI Search
Log every AI crawler with Workers and Analytics Engine, then use Cloudflare AI Search to see which of your pages a RAG engine actually retrieves for buyer prompts.
October 27, 2026
How to Get Cited by ChatGPT, Perplexity and Google AI Mode
How ChatGPT, Perplexity and Google AI Mode choose sources, which crawlers to allow, and a five-step plan to get your company cited, based on official docs.
October 26, 2026
How to Set Up AI Search Attribution in GA4: A Step-by-Step Guide for 2026
Step-by-step GA4 setup for AI search attribution: an AI Assistants channel, a copy-paste regex, ChatGPT UTM fixes and a self-reported AI source field.
October 26, 2026
Voice Search in 2026: How AI Assistants Pick Their Sources
We first wrote about voice search in 2022, when the question was whether “Hey Siri” and “OK Google” would change how people shop. Four years later, voice is back in a bigger form. People now talk to ChatGPT, Gemini, Siri, Alexa+ and Copilot the way they talk to a knowledgeable friend, and those assistants answer by reading the web in
October 17, 2026
Contact us
We value your privacy and won't share your email with others.
We'll only contact you with curated content.
