Search is splitting in two: the ten blue links, and the answer an AI assistant gives before anyone clicks. I work on both, in a category where every health claim has to be defensible.
llms.txt, named AI-crawler access, citation-style notes, and a Lighthouse SEO score of 100.The brand sells a supplement for people on GLP-1 medications, one of the most searched and most regulated health topics right now. People increasingly ask ChatGPT or Perplexity things like "what helps with GLP-1 side effects?" and trust the answer.
Two constraints shape everything. The content has to be genuinely useful enough to be quoted, and it has to stay inside supplement-claim rules: structure/function language only, never a treatment claim. An SEO win that creates a compliance problem is a loss.
This was during the period I owned the Shopify site end to end, before it moved to a dedicated software team.
SEMrush research split demand into buying intent ("GLP-1 support capsules"), comparison intent, and question intent ("GLP-1 nausea supplements"). Each got a home: product pages, comparison sections, or FAQ answers.
So: No page competes with another page for the same query, and every question a buyer asks has one clear answer on the site.
I rewrote product titles in a "product | key benefit" pattern with meta descriptions written as the search snippet. For example: GLP-1 Total Support Powder | Fiber, Protein, & Collagen. I also cleaned up headings and checked sitemaps and indexing in Search Console.
So: These titles and descriptions are still live on the store today.
I added product and organisation markup so Google and AI systems read name, brand, price, and availability the same way everywhere, plus FAQ markup for the question content.
So: Structured data is how an AI assistant gets your facts right without having to guess from the page layout.
Question pages were rewritten answer-first: the direct answer in the first two sentences, then the detail, all inside claim rules. AI crawlers were allowed in, and I split AI referrals (chatgpt.com, perplexity.ai) into their own GA4 channel so the traffic stopped hiding in "referral".
So: You can’t improve what you can’t see. Separating AI referrals was the first step to treating AI answers as a channel.
A lot of my Summit work is now maintained by another team. So I built my own portfolio as a working GEO example that anyone can inspect:
Person, WebSite, FAQPage (mirrors the visible FAQ), Article + BreadcrumbList on every case study, and BlogPosting on every note.sameAs.What I believe works, and what I'd bring to your team:
llms.txt, and explicit AI-crawler permissions.I'd build structured data into the theme itself, not individual pages, so it survives redesigns and handoffs. Some of the markup I added didn't make it through the move to a new team, and theme-level markup would have.