Are Shopify stores missing AI search traffic? Curious what others are seeing

SEO
23

most shopify stores are invisible to AI search because they’re using the same supplier images and generic descriptions as everyone else. if chatgpt or perplexity can’t tell your store apart from 50 others selling the same product, it has no reason to recommend yours specifically.

the two things that actually help are unique product descriptions written in natural conversational language and original product imagery. we use Prodofoto to generate unique lifestyle and editorial shots from our base product photos so our listings look completely different from competitors using the same supplier images. when AI search tools cite specific sources, standing out visually and textually is how you actually get mentioned.

24

This thread lines up with a pattern I’m seeing in a small category test.

The failure mode wasn’t simply “the store is invisible.” It was more specific: direct product/attribute searches worked, but buyer-context searches failed.

For example, product/category data could surface results for direct electrolyte terms, but searches like “hydration powder for mountain biking,” “electrolyte drink mix for trail running,” or “zero sugar electrolyte powder for hot weather” returned nothing, even though relevant products existed.

That makes me think the issue is not just schema, /agents.md, UCP discovery, or bot access. Those seem necessary, but not sufficient. The missing layer seems to be whether the product data, collection copy, PDP copy, and visible rendered HTML actually express use case, audience, constraints, ingredients, price band, and comparison language in a way catalog/search systems can retrieve.

Curious if others are seeing this same pattern:

  1. When a store is technically crawlable and has product schema, are the misses usually thin PDP copy, weak collection pages, missing product attributes, or client-rendered content that doesn’t survive crawling?
  2. Who normally owns those fixes in practice: SEO, merchandising, dev, feed specialist, or founder/operator?
  3. Would anonymized failed prompts + a fix roadmap be a useful diagnostic format, or is this still too early/noisy?
25

@Geoffy is right, and it is worth stating clearly. There are three separate layers, and you need all three on a given surface.

Crawl access. The bot can physically fetch the page. The Cloudflare citation bot block is exactly this, and it makes everything else moot.

Eligibility. Clean Product and Organization schema. This makes you a candidate.

Citability. The same facts appear as visible rendered text on the page, not only inside a structured data script block. At answer time the model is mostly reading the rendered HTML, so a fact that lives only in schema often has nothing it can quote.

Most stores have one of the three and assume they are covered.

OperatOS, your pattern is the important one, and your read is correct. Direct attribute search works because the catalog answers what is this product. Buyer context search fails because nothing on the page answers what is this product for, and for whom. Hydration powder for mountain biking is a use case and a constraint, not an attribute. If the use case, the audience, the constraints, and the comparison language do not exist as retrievable text on the PDP, and especially the collection page, the system has nothing to match the query against.

To your specific questions.

When a store is crawlable and has schema, the misses I see most are, in order: thin or generic PDP copy that lists attributes but never states a use case, weak or empty collection pages that should be the answer to a buyer context query and are usually just a product grid with no words, and client rendered content that does not survive the fetch. Missing attributes matter too, but the use case gap is the biggest.

Who owns it. In practice it spans dev, merchandising, and content, which is exactly why it falls through the cracks. The founder or operator has to own the outcome and pull the three together, because no single function sees the whole failure.

And yes, anonymized failed prompts plus a fix roadmap is a genuinely useful format. Failed buyer context prompts are the cleanest signal there is of where the retrievable content is missing. It is the difference between guessing and knowing.

The frame I keep coming back to. Schema makes you a clean entity. Buyer context is what makes that entity findable for a real question. You need both.

26

Not really, many seo experts ( like me) know how to rank for AI overview and generative engines. One tip from my side is to add the llm.txt file, same like robot.txt file.

27

Hey @Geoffy .

Really interesting, and you are ahead of most merchants on this.

AI search traffic is real and growing. A few merchants are already seeing visits from Perplexity and ChatGPT in their analytics. Still small but increasing fast.

Your point about comparison and category content mattering more than product pages is spot on. AI answers questions, it does not rank pages. Plain, clear content that directly answers a question gets cited the most.

Most merchants are just treating it as normal SEO right now, so there is a real early advantage for anyone who starts optimising for AI citations now.

The free audit offer is a great idea. I would love to see what patterns you find!

29

Cloudflare allow-lists, robots.txt, schema audits: all configuration. They tell you what’s permitted, not what happened. And zero citations has two completely different causes that look identical from the config side. Either the bot never fetched you, or it fetched you and picked someone else. One is an access problem, the other is a content problem, and the fixes are opposite.
The observable version is cheap: which agents actually hit your store, which URLs, how often. Server logs will show it, or anything that reads user agents.
It also settles Anthony’s check three without guesswork. If ClaudeBot or ChatGPT-User is fetching a PDP repeatedly and you’re still not cited, access is fine and the page is the problem. If they never show up at all, no amount of comparison content will help yet.
So maybe a check four after ryan-bowne’s two: not “am I allowed”, but “did they come, and where did they land”.
I build a Shopify app in this space, Logs work fine for this.

30

Anthony — there’s one thing I haven’t seen mentioned in any of these threads, and in the stores I’ve looked at it costs more than schema ever will: AI visibility is language-scoped.

Ask “best Korean skincare in Mexico” in English, then ask it in Spanish. You get two different brand lists. Most merchants only ever test in English.

I went through a Mexican store selling Korean skincare recently — good catalogue, physical shop in Mérida, real trust signals. But nearly every product title was still the supplier’s English: “anua pdrn hyaluronic acid capsule 100 serum 30ml”. Their customer searches “serum con ácido hialurónico”. To a model answering in Spanish, that store has almost nothing usable.

Same store had a Cyrillic С instead of a Latin C in several collection handles. Identical to the eye, indexed as words that don’t exist in Spanish. Copy-paste artifacts from supplier feeds — I’d bet it’s far more common than anyone realises.

To your actual questions: the pattern I keep hitting is brand-aware queries work fine, discovery queries return nothing. Which matches what a few others have said. What closes that gap isn’t more markup — it’s whether the page says anything specific enough to be worth citing. Supplier descriptions are identical across hundreds of stores, so a model has no reason to pick one over another.

Free check, two minutes: run your category query in your customers’ language, not in English. If your market isn’t English-speaking, you might be testing the wrong thing entirely.

Disclosure: I build a tool in this space (Optaru), so weigh that accordingly. Happy to go deeper on any of it.

31

Excellent point. Very, very well made.

We’re based in Europe, and we tend to work in a single markets, But in countries where there are multiple languages, you are incredibly correct in what you say.

I forget how much Spanish is actually spoken in the US!

32

I’ve been looking at this too, and I’m still treating it as an extension of SEO rather than a completely separate thing. If the product pages aren’t crawlable, the site structure is messy, or Google can’t make sense of which pages are important, I’d fix those before trying to build a whole new layer of AI-specific content.

The intent part is interesting though. A product page can rank fine for its main terms and still not be the page an AI system wants to reference for something more specific. I’ve been paying more attention to how product and collection pages connect for that reason.

I use SiteGuru for the regular audits, mainly because it helps me prioritize which pages actually need work instead of just giving me another huge list of things to fix. I haven’t seen enough AI referral data yet to say what’s actually moving the needle there.