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

SEO
1

Hi everyone,

I’ve been digging into how Shopify stores show up in AI search (ChatGPT, Perplexity, Gemini), and I’m starting to think there’s a gap most of us aren’t really accounting for yet.

When someone asks something like:
“What’s a good [product] under £X?”
or
“Best option for [specific use case]?”

AI engines don’t behave like Google. They’re not really “ranking pages” in the traditional sense — they’re pulling structured, intent-specific answers.

From what I’ve seen so far:

  • Most product pages aren’t written in a way these systems can easily use

  • Even strong SEO stores don’t consistently show up in AI responses

  • Category and comparison-style intent seems to matter more than individual product pages

I’ve been testing a few approaches around this, mainly:

  • creating different layers of content based on intent (broad → comparison → very specific queries)

  • structuring it so it’s easier for AI systems to interpret and cite

  • publishing it in a way that sits alongside the main store rather than replacing anything

Results are still early, but interesting enough that I wanted to sanity check it with others here.

A couple of questions:

  • Has anyone actually seen traffic or sales coming from AI tools yet?

  • Are you doing anything deliberately to influence that, or just treating it as SEO for now?

  • If you’ve tested anything, what’s worked (or not)?

If it’s useful, I’m happy to run a quick audit on a few stores and share what I’m seeing — no pitch, just curious to compare notes.

Cheers,
Anthony

2

Hi @Geoffy, great questions and your framing is exactly right. Yes, there is a real gap, and it is not being addressed by standard SEO tools yet.

To answer your specific questions:

Has anyone seen traffic/sales from AI tools?

Yes, but it is highly query-dependent. I have seen direct referral traffic from Perplexity (it passes referrer headers) and occasional ChatGPT-attributed sessions via UTM when using Shopify Agentic Storefronts. The volume is small right now but the conversion rate on those sessions tends to be higher because the buyer has already been pre-qualified by the AI before arriving.

What actually works right now:

  1. Product Schema quality is the foundation. AI systems like ChatGPT and Perplexity crawl your store and try to understand product-level details. If your JSON-LD is clean (name, description, price, availability, brand, reviews), you are giving the model something structured to work with rather than forcing it to guess from paragraph text.
  2. Your layered content approach is correct. The intent pyramid (broad to specific) maps well to how AI systems chunk and retrieve content. A page answering “best collagen supplement for joints vs skin” will outperform a generic category page for that kind of query every time.
  3. llms.txt is the underused lever. It is essentially a structured brief for LLMs about your store, what you sell, who you sell to, and what makes you different. Instead of the model guessing, you give it a spec sheet. Most Shopify stores do not have one yet, which means there is still first-mover advantage.
  4. Perplexity Collections and Merchant Center for ChatGPT Shopping. Submitting a clean product feed to Google Merchant Center (which feeds ChatGPT Shopping) and getting into Perplexity’s shopping index are the two highest-leverage distribution actions right now.

What has not worked:

Rewriting product descriptions without changing the information density does nothing. Adding FAQ sections that just rephrase existing content also does not move the needle. The model needs new, citable, specific information it cannot find elsewhere.

Happy to compare notes on what you are seeing. The intent-layered content approach you are testing is the right direction.

3

We consider it as an inevitable part of SEO now, but we don’t overcomplicate it by calling AEO and GEO something totally different.

The reason is that many measures to improve your visibility correlate with what you normally do for SEO. It is still vital to have your content well-organized, useful, and reasonably structured. Probably, short formats and prompts that help your goods be included in listicles predominate for the mission. The prompts can be taken from Ahrefs/Semrush if you already use those for SEO, or you can keep an eye on your AI visibility in a tracker like Beamtrace for AI search only.

You already know that DR and backlinks are vital for SEO - so they are for AI visibility, but the focus is made on natural mentions (not necessarily links) in discussions, threads, and social media posts.

4

Nobody really knows how to optimize for AI chatbots at this time, except that now the product information can be served over Shopify’s global catalogue.

Most conversations with AI are still heavily influenced by current SEO rankings.

5

I believe AI traffic does exist but it’s still small due to the lack of input. The interesting part is that customers coming from AI search have really specific intent. Thus conversion tends to be higher. And you’re right about the product pages where most of them are too salesy or not designed to answer. That’s why brands are now more focused on providing comparisions, use cases to offer anwsers which customers really look for.

Having said that, what could work might include:

  • clear structured answers to directly provide information to customers via summaries, FAQs
  • comparision content
  • clear data

Basically, you’re making your pages easier to read and extract information, making it more “understandable”

Hope this helps!!

6

Good to see this thread get fresh discussion. Wanted to add some data points to what Ugurcan and Khanh-Linh2 have shared.

On Ugurcan’s point that “nobody really knows how to optimize for AI chatbots at this time” - that was true 12 months ago but the picture is clearer now. The mechanism is knowable: AI systems retrieve from two layers. The first is external citations (what other sites say about you). The second is direct crawl data from your store. Most SEO advice only addresses the external layer, but the on-store layer is where Shopify merchants specifically have the most control and the most gap.

The concrete pattern I keep seeing across stores: the ones getting AI-attributed sessions tend to share three things. Their product schema is complete and clean. Their product descriptions answer “who is this for and why is it better than X” rather than just “what is it.” And they have either an llms.txt file or enough structured content on their About page for the model to confidently describe them.

Khanh-Linh2 is right about comparison content. The specific format that gets cited most consistently is not FAQs but comparison framing: “[your product] vs [category alternative]” pages written in plain language. These answer the exact queries AI tools get asked and the model can pull a clear citation from them.

The volume is still small but the conversion rate gap is real. A session from someone who arrived after ChatGPT said “this store sells X that would work for your use case” converts differently from a cold PPC click. That quality gap is going to matter more as the volume grows.

@Geoffy are you still tracking attribution? Would be useful to see how the numbers have moved over the last month.

7

Thanks all — useful thread, and a few of these replies have shifted how I’m thinking about it.

Quick update from the testing side: nine days on, the pattern that’s holding up most consistently is what a couple of you flagged — AI-sourced sessions are smaller in volume but materially better in intent. The visitors arrive already past the “is this a thing I want?” stage and are mostly trying to confirm fit. Conversion-rate skew has been the most interesting signal, not raw traffic.

A few things I’ve changed my mind on since posting:

  1. Schema is necessary but not sufficient. Clean Product/Offer/AggregateRating JSON-LD gets you eligible to be cited, but it doesn’t get you cited. The pages that actually show up in answers are the ones written around a specific buyer question rather than a product spec. Lots of stores have the schema and still don’t surface.

  2. Comparison and use-case pages punch above their weight. Echoing what was said earlier in the thread — “best [X] for [use case]” pages outperform individual PDPs by a wide margin in citations. I think it’s because they map cleanly to how people phrase prompts.

  3. llms.txt is mostly performative right now. Worth shipping for hygiene, but I haven’t seen evidence any of the major engines are pulling from it in a way that changes outcomes. Happy to be proved wrong if anyone has data.

On the “is this knowable?” debate — I’d lean towards “knowable, but the inputs are different.” It’s not classical SEO with new keywords; it’s structuring content around intent layers (broad question → comparison → specific configuration) so an AI can stitch a coherent answer that cites you. That structuring is doable; what’s still genuinely uncertain is which engines weight which signals.

Anyone here seen meaningful citation lift from feed distribution specifically (Shopify → Google Merchant → Gemini)? That’s the one I’m still trying to get clean data on.

Cheers, Anthony

8

Hey Anthony — one specific thing we just ran into that’s relevant to the citation question:

If your store is on Cloudflare and you have “Managed robots.txt” enabled (DNS → AI Crawl Control), the default block list includes ClaudeBot, Google-Extended, and Applebot-Extended alongside the training-only crawlers (GPTBot, CCBot, Bytespider). A lot of stores leave this default on without realising those three are *citation* bots rather than training crawlers — they’re how Claude search, Gemini AI Overviews, and Apple Intelligence reach out to fetch the page when they want to quote it with attribution. So you can be doing all the intent-layered content work and still be invisible to those surfaces because the bot can’t read the page.

You can keep an `ai-train=no` posture by leaving the training-only ones blocked but unblock the citation/inference bots. We turned the managed setting off and wrote a custom robots.txt that blocks GPTBot/CCBot/Bytespider/Amazonbot/meta-externalagent but explicitly Allows ClaudeBot/Google-Extended/Applebot-Extended. Sample size of one — but the first cited hit from Claude search showed up about 48 hours later, on a comparison-style post.

On Shopify → Merchant → Gemini feed-citation specifically — haven’t seen a clean signal there either. What keeps tripping us up is that Merchant Center listings surface through Google Shopping/SERP rather than Gemini’s chat-citation flow, so the route you’d hope for (“Gemini answer mentions our store with a link”) isn’t what the feed actually feeds. Would also be curious if anyone has clean data on that — it’d change how I’d prioritise feed-quality work.

One observation on content shape: the biggest delta we saw was from comparison-style posts (“X vs Y vs Z, when to pick each”) rather than list articles. They answer the exact question shape AI search rephrases, so the model can cite a single paragraph cleanly without summarising five.

9

This is the most useful single comment I’ve had on the thread, thank you.

The training-vs-citation bot distinction is exactly the lens most stores aren’t applying. ClaudeBot, Google-Extended and Applebot-Extended are the inference-time fetchers — they’re how the answer surface confirms the page exists and pulls a quotable chunk at the moment of the query. Blocking them while leaving “I want AI traffic” as the goal is self-defeating, and the Cloudflare managed default makes it a one-click own-goal. I’d add PerplexityBot and OAI-SearchBot to your allow list for the same reason — Perplexity’s answer engine and ChatGPT’s browsing/search mode both rely on them, and they’re distinct from PerplexityBot-Train and GPTBot respectively.

The split posture you’ve described — block the training-only crawlers, allow the citation/inference ones — is the right shape. We’ve been recommending the same logic and the 48-hour cited-hit timeline matches what we’ve observed: it’s quick once the door is open.

On the Merchant feed → Gemini point: agreed, and I think you’ve put your finger on why the signal is muddy. The feed flows into the Shopping graph, which surfaces through SERP/Shopping/AI Overviews-shopping-modules, not through the chat-citation pathway. Two largely separate retrieval flows that both happen to be Google. So feed quality lifts your odds in the shopping carousel inside an AI Overview, but doesn’t necessarily get you cited in a conversational Gemini answer. Worth investing in for the former, not a substitute for the latter.

On comparison vs list content — strong agreement, and I think the mechanism is exactly what you said: the question shape matches. List articles (“10 best X”) force the model to either summarise the whole page or pick one item and lose context. Comparison posts (“X vs Y vs Z, when to pick each”) give it a single paragraph that’s already structured as a defensible answer, so it can cite cleanly. The corollary is that the granularity of the comparison matters — “X vs Y” beats “10 alternatives to X” because the former is paragraph-citable.

One thing I’d add: head terms get a different treatment from long-tail. For specific configurations (“best [product] for [niche use case under £X]”), comparison structure wins. For broader queries, the engines tend to lean on category-level pages or Reddit/forum content. Worth structuring across both layers if you can.

Thanks, Anthony

10

Well, the classic SEO that used for Google will not work in that case. In order to appear in the search results of LLM’s, then mainly you need to focus on LLM SEO instead the Google SEO.

And yes, LLM SEO is possible and via this SEO many of the shopify stores getting traffic and that traffic converts into buyers.

11

Thanks — interested to hear you’re seeing converting traffic.

I’d push back gently on the “classic SEO won’t work” framing though. From what I’ve seen the overlap is bigger than people give it credit for: clean information architecture, crawlable HTML, fast pages, internal linking, structured data — these all still matter, because the citation/inference bots are essentially crawlers with a different ranking objective on top. The bit that is genuinely new is content shape (intent-layered, comparison-format, paragraph-citable) and bot access policy (allowing the inference fetchers, which I posted about above).

So I’d frame it less as “LLM SEO instead of Google SEO” and more as “Google SEO foundations + a new layer on top.” The foundations still do work for you.

If you’ve got specific tactics that are driving the converting traffic you mention, would be useful to hear what they are — that’s the part of this thread I’d most like to see grounded in actual data.

12

@Geoffy the “Google SEO foundations + a new layer on top” framing is exactly right and worth repeating more loudly because most merchants think they have to choose.

On specific tactics driving converting traffic across stores we work with: the pattern that comes up most consistently is intent-matched FAQ content on collection pages. Not generic FAQs about shipping or returns, but questions that are literally the prompt a buyer types into ChatGPT. “What is the best [material] for [use case] under [price]” answered in two sentences on the relevant collection page. The model can lift that paragraph verbatim and it answers the query completely, which is the citation trigger.

The bot access point you raised earlier in the thread is the prerequisite that most people skip. We have seen stores with well-structured content get zero citations because ClaudeBot or OAI-SearchBot was sitting in a default Cloudflare block. Unblocking those before worrying about content shape is the right order of operations.

On the Merchant feed to Gemini question: we have not seen a clean path from feed quality to chat-citation either. The shopping carousel in AI Overviews is a separate retrieval path and worth keeping clean, but it is not how you get named in a conversational answer. The conversational citation comes from the page content itself, not the feed.

13

@Geoffy this is the best thread on the topic I have read here, and I think you all landed in the right place: citation-bot access first, then comparison and paragraph-citable content. John_Moore1’s Cloudflare point and your two-retrieval-flows breakdown are the parts most stores miss entirely.

You asked twice for data over mechanism, so here is an offer of data rather than more theory. The two things this thread agreed on, bot access and whether your schema is actually readable, are both binary and both measurable in a couple of minutes. The piece most checks get wrong is the one John_Moore1 flagged: the real blocks usually sit at the Cloudflare edge in AI Crawl Control, not in the Shopify robots.txt, so you have to look at both or you get a false all-clear.

So, no theory attached. Reply back with your store URL and I will run those two checks and post exactly what I find right here, so it is data and not another opinion. Start with yours and we can hold it against what you have been seeing in your testing. Same for anyone else on the thread who wants it.

14

The other interesting aspect here is the role of the Shopify Catalog and agentic commerce. It is still early days but it will be interesting to see how protocols like UCP get adopted.

I have been running some tests using the Shopify Catalog to see how results are returned. If a store has agentic storefronts enabled then you can get some insights into what Shopify is returning to the agents on the discoverability side.

You can also add /agents.md to the end of your store and that will give you some hints too.

15

@ryan-bowne the methodology is right and the double-check matters. One thing worth adding as a third failure mode: stores that pass both (Cloudflare edge + robots.txt) can still have a retrieval problem where the bot reaches the page but the content doesn’t survive. All the product detail lives in JavaScript that never renders at crawl time. Bot access confirmed, schema present, still zero citations—because the page the model sees is effectively empty. Worth adding that as check three before drawing conclusions.

@adam-cpd the /agents.md point is underrated. The pattern worth noting is the difference between auto-generated and manually curated versions. Auto-generated ones tend to be capability descriptions that don’t tell an agent what to recommend or why. Manually curated reads more like a positioning brief—use cases, differentiators, the specific questions a buyer might ask. That’s the version that actually gets lifted.

On UCP: Shopify now ships sitemap_agentic_discovery.xml and UCP discovery natively, so for Shopify stores the infrastructure question is largely answered. What’s not answered is content quality within that infrastructure—whether what an agent retrieves is actually good enough to produce a confident citation. That’s the gap most stores discover last.

16

Sharing what has worked for me.

I’m not just a Shopify app developer. I also own multiple Shopify stores, so everything below comes from testing on my own sites.

First, write something that’s actually useful. Don’t mass-produce AI slop. Use AI to research your product, your competitors, and, most importantly, the subreddits where your target audience hangs out.

Figure out:

* What are they talking about?

* What problems do they have?

* What language do they use?

* What are they worried about?

Then write content that genuinely answers those questions. AI can help with research, but your own judgment is what makes the article valuable.

Now here’s the part that surprised me.

Most people just add a text link to their product page. In my testing, ChatGPT may cite the article, but it almost never recommends the product itself.

What worked much better was embedding a rich product card directly inside the article.

Include the product image, name, summary, price, and other structured information. Give ChatGPT something obvious to recognize and extract.

In our experiments, this dramatically increased the chances of the product being surfaced as a shopping card or product card in ChatGPT responses.

That’s exactly why we built our product, clickfrom.ai

17

@Geoffy that third one is the real killer, and it is the one every upstream check sails past. Bot in, schema present, content rendered client-side, model lands on an empty page. Worth adding: Shopify ships the discovery surface natively now, but the crawlers that actually cite are mostly still reading served HTML, not the new agent files. Infrastructure is ahead of consumption, which is exactly why your layer three matters more than the files do right now.
Offer stands for anyone who wants the first two run on a real store.

18

I’d separate this from classic ranking for now. For a Shopify store, the practical checks I’d start with are whether product names, variants, prices, availability, shipping, returns, and policy pages are easy to find in served HTML, and whether Product/Offer/Review schema matches what shoppers can actually see. AI traffic may still be small, but those same basics help AI tools understand the store without requiring any special claim about rankings.

21

Hi Anthony,

You hit the nail on the head. AI search (often called GEO) is definitely the next big thing, and most Shopify stores are falling behind because they still treat it like old-school Google SEO.

You are right that AI agents don’t just “rank” you; they “read” you. They are looking for specific facts and technical details to answer user questions. If your product data is messy, the AI can’t verify your information and will likely skip your store for a competitor that has clearer data.

To get recommended by tools like ChatGPT or Perplexity, your store needs a perfect “cheat sheet” that the bots can understand. This means having very strong technical schema (JSON-LD) that identifies exactly what your product is, its price, and its unique features.

For anyone who wants to automate this part, I highly recommend using SearchPie: SEO, Speed & Schema. It automatically builds the advanced schema and structured data that these AI engines rely on to find and cite your products. It basically translates your store into a language that AI bots can read perfectly.

Hope this helps, and I’d love to hear more about the results of your audits!

22

Thanks PieLab — agreed on the framing: AI engines read you rather than rank you, and stores with messy product data just get skipped. Clean JSON-LD is genuinely part of the answer.

One thing I’d add from our own testing, because it caught us out early: schema gets you eligible, but it isn’t reliably what gets cited. When ChatGPT, Perplexity or Claude fetch a page at the moment they answer, they’re mostly reading the visible rendered HTML — the JSON-LD in a <script> block tends to get ignored at inference time. So if your rating, spec and FAQ data only live in the structured-data block and never appear as visible text on the page, the model often has nothing it can actually quote. The fix is unglamorous but it works: mirror the same facts into the on-page HTML, not just the schema.

That also lines up with what Rahul and John said earlier — the engines need new, specific, citable information they can’t find elsewhere, and they need to physically be able to fetch the page (the Cloudflare citation-bot point). Schema is one layer; visible citable content and crawl access are the other two. Miss any one and you’re invisible on that surface regardless of how clean the JSON-LD is.

Happy to keep comparing notes — the visible-vs-structured gap is the single most common thing I’m seeing across audits.