AI SEO vs GEO Shopify: SEO Pays, GEO Is Insurance

AI SEO vs GEO Shopify, tested on our own site — the Cloudflare mistake that blocked every AI crawler, and why classic SEO still pays the bills.

Linh Nguyen · Updated 2026-07-04

In June 2026, a few weeks after we launched storefleet.io.vn, I was feeling quietly smug about our AI readiness. We had an llms.txt file at the root. We shipped FAQ and Article structured data on every post. By the checklist standards of every "GEO agency" pitch landing in my inbox, we were done. Then I opened the Cloudflare dashboard for an unrelated reason and noticed its AI-bot blocking feature was switched on — silently returning errors to GPTBot, ClaudeBot, PerplexityBot, every AI crawler we supposedly wanted to attract. We had written a welcome mat for generative engines and locked the door from the inside.

That mistake is the most honest way I know to frame the AI SEO vs GEO question for Shopify, because it captures both truths at once: GEO work is cheap and worth doing, and almost nobody — including us, at first — verifies it's actually switched on. This post is where I've landed after running both disciplines on our own site and thinking about them across the five Shopify stores we operate: classic SEO still pays the bills, GEO is inexpensive insurance, and anyone selling you guaranteed AI citations is selling faith.

The three acronyms, minus the consulting deck

The terms themselves are standard and worth thirty seconds:

• SEO targets rankings and clicks on a search results page — keywords, links, technical speed, content depth. Unchanged in spirit since forever.
• AEO (Answer Engine Optimization) formats content so answer systems — Google's AI Overviews, Perplexity summaries — can extract it and present it directly. In practice this means unambiguous headings, direct answers near the top, and structured data.
• GEO (Generative Engine Optimization) aims at being cited inside AI-generated answers when someone asks ChatGPT or Claude for product research or recommendations.

What I refuse to do is inflate these into some proprietary multi-layer framework, because the most useful data point I've seen argues the opposite: they mostly collapse into each other. BrightEdge's year-one analysis of AI Overviews found them appearing on roughly 48% of monitored queries — and about 38% of the URLs cited also ranked top-10 organically for the same query (their monitored panel, not the whole web). Read that second number again. A large share of "AI visibility" is just regular search visibility wearing a new interface. Do the boring SEO work well and you've already done most of the GEO work by accident.

What classic SEO looks like on a brand-new domain (spoiler: slow)

Here's the unglamorous part nobody's inbox pitch mentions. Our site launched around June 2026 with a sitemap of roughly 265 URLs across English and Vietnamese. We did the technical work properly: prerendered static HTML for every page, self-hosted fonts, structured data validated before launch. Then we met reality: on a new domain, Google indexes almost nothing on its own schedule, and Google Search Console's manual "Request Indexing" button is quota-limited — in our experience a rolling ~10–12 URLs per day before it cuts you off (one site, one operator; your quota may differ).

So indexing became a daily grind: pick the highest-priority URLs, request them, wait, repeat tomorrow. And it's still slow. In early July 2026 I checked a batch of posts we'd published on June 20 — weeks later, several were still not indexed at all. That's one small site, not a study, but it recalibrated my expectations: on a fresh domain, classic SEO is a months-long compounding bet, not a switch.

Here's the thing though — I can see all of it. Search Console shows me exactly which URLs are indexed, which queries show impressions, which pages earn clicks. Slow, but measurable. Keep that word in mind for the next section.

Our GEO afternoon — and the trapdoor underneath it

The entire GEO implementation for our site took about an afternoon, which is roughly what it deserves:

1. llms.txt at the root — a plain-text map pointing language models at our key pages. Our step-by-step guide to optimizing a store for AI with llms.txt covers the exact format we used.
2. Structured data — FAQ and Article JSON-LD on posts, validated with Google's Rich Results Test.
3. Explicit crawler access — we allow GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot, PerplexityBot, and Google-Extended by name in robots.txt.

Step 3 is where the June discovery bit us. All of it was worthless while Cloudflare's "block AI bots" setting sat between the crawlers and the site, returning errors on our behalf. If you take one tactical thing from this post: after any GEO setup, fetch your own pages as an AI user agent (curl with GPTBot's user-agent string is enough) and confirm you get a 200 and real HTML. CDN-level bot blocking defaults have quietly flipped on for a lot of sites, and llms.txt does nothing from behind a locked door.

Now the uncomfortable honesty. I cannot measure whether any of this works. There is no Search Console for ChatGPT. I don't know if Claude has ever cited us. And the best available evidence says my llms.txt file specifically is probably decorative: SE Ranking's analysis of 300,000 domains found no statistically significant relationship between having llms.txt and increased AI citations. I shipped it anyway, because the cost was twenty minutes and the downside is zero. That's the correct frame for most GEO tactics: cheap insurance, not a growth channel — and definitely not something to pay a retainer for on promised outcomes.

Translating this to an actual Shopify store

Everything above was our content site. On a Shopify store, the mechanics shift but the SEO-first logic holds:

• You may already have llms.txt. Shopify has rolled out native llms.txt files for stores at yourstore.com/llms.txt, including store metadata and MCP endpoints. Check yours exists before paying anyone to "implement" it — and then check your CDN and bot-protection apps aren't blocking the same crawlers it invites, which is exactly the trap we fell into.
• Product structured data is the real GEO work. Product, Offer, and Review JSON-LD is what lets any engine — Google or generative — quote your price, availability, and ratings with confidence. Run key templates through the Rich Results Test; broken schema is more common than missing schema in my experience auditing our own stores.
• Clean catalog data feeds AI agents directly. Shopify's Catalog MCP server lets AI agents search products across merchants, which makes product data quality and tagging a discoverability feature, not housekeeping. On our stores, the tag and taxonomy cleanup that helps AI filtering is the same work that helps Google Shopping — one effort, two channels.
• Watch the citation-heavy engines. Perplexity and Google's AI Mode surface sources more aggressively than ChatGPT does, so if AI referral traffic shows up anywhere first, it's likely there. We check referrer logs, not vendor dashboards.

Where I land: SEO pays the bills, GEO is cheap insurance

You've probably seen the Gartner projection that traditional search volume drops 25% by the end of 2026 as AI search grows. Maybe. It's a 2024 projection, and from where I sit in mid-2026, Google is still where our measurable traffic lives — while AI referrals remain a rounding error I can't yet attribute properly. Both things can be true: the shift is real, and it's earlier than the panic suggests.

So my split, as an operator: spend real, recurring effort on classic SEO — content, structured data, indexing, site speed — because it's measurable and compounding. Spend one honest afternoon on GEO — llms.txt, crawler access, schema — then verify it with your own curl requests and move on. Revisit quarterly. And when an agency promises AI citations as a deliverable, ask them how they'll measure it; the silence is informative.

One more thing GEO pitches never mention: discovery is worthless if the store behind it is a mess. As AI agents take on more of the buying journey, the winners will be stores whose data — orders, inventory, shipping, disputes — is clean enough to act on. That's the operational layer: every store in one dashboard, orders, revenue, shipments, and disputes consolidated, so operations keep pace with however customers find you.