ChatGPT SEO: I Built a Test Store, and 80 of Its 91 Orders Came From AI Search

Everyone in marketing is asking the same question this year: does ChatGPT actually send customers, or is AI search just another hype cycle?

I got tired of reading opinions. So I ran the experiment myself.

Three months ago I set up a small crystal store on Shopify. New brand, no audience, no domain history. About one day to set up the store and the brand, another two or three days on products and pages. Then I promoted it the way any small store would, and watched where the orders actually came from.

The store has taken 91 orders. More than 80 of them came from ChatGPT and Perplexity.

Not from the Instagram ads I paid for. Not from organic Google. From AI answers, tracked order by order through utm_source and referrer data.

And buried inside those orders is a finding I have not seen published anywhere: the way AI-referred buyers pay tells you something about AI search that no traffic report can. This is the full case study, with real screenshots, the playbook, and the honest limits.

The experiment: one small store, one question

The setup was deliberately ordinary, because the point was to test what a normal small brand can expect.

A niche product people genuinely ask AI assistants about, crystals, where buyers want guidance more than brands. A clean Shopify store with proper product pages. A consistent brand across Instagram, Facebook and the store. Some Meta ads, mostly aimed at brand building. Prices are in INR because the store sells in India; I will convert everything for context.

One thing worth being upfront about: the money here is small. Ninety-one orders and about ₹50,100 (~$600) in sales, at an average order of roughly ₹550 (~$6.6). This was never a get-rich story.

It is a behaviour study with real stakes. What matters is not the revenue, it is who sent the buyers, and how those buyers behaved when they arrived.

What happened in three months

Here is the Shopify overview, three months from launch.

Ninety-one orders. And when I break them down by source, the picture stops being ordinary.

The ads bought sessions. The AI answers bought customers.

The run rate has kept climbing too. The store now does 40+ orders in a month, and the pattern holds day to day. On the day I wrote this, four orders came in; two of them carried utm_source=chatgpt.com.

That UTM tag is worth pausing on. ChatGPT stamps its outbound clicks, so this is not guesswork or modelled attribution. Each of those orders is traceable to an AI answer, which is more than most channels can honestly say. If you have read my piece on AI dark traffic, you know how rare clean attribution is in this space.

The stat nobody talks about: AI buyers prepay

Here is where the data stopped being a traffic story and became a trust story.

Some context for readers outside India. Cash on delivery, COD, is the default in Indian ecommerce, roughly 60 to 65% of all orders. And COD is brutal on small stores: 25 to 30% of COD orders return undelivered, called RTO, return to origin, and the seller pays shipping both ways. Prepaid orders RTO at just 2 to 3%.

COD is, bluntly, what buyers choose when they do not fully trust you.

Now the number. Out of our 80+ ChatGPT and Perplexity orders, exactly one was COD.

One. In a market where two out of three orders normally refuse to pay upfront.

Sit with that for a second, because I did.

A buyer who lands from an AI answer has already had their questions answered. The assistant compared options, explained the product, and named the store. By the time they click, the trust decision is made. They pay online, upfront, to a brand that is three months old.

An AI recommendation arrives pre-trusted. That is the real value of ranking in AI search, and it never shows up in a sessions report.

Why did the ads lose to a chatbot?

I want to be fair to Meta here, because the ads were not useless. They were doing a different job.

Most of our Instagram spend was brand building: making the store look real, alive and worth trusting. That matters, and I will come back to why. But as a direct sales channel, the ads were poor. Four or five genuine sales across the whole test. The rest of the ad-attributed orders were COD, and most of those never completed delivery.

The difference is intent. An ad interrupts someone who was not looking. An AI answer responds to someone who asked.

The macro data says my little store is not weird on this. Salesforce found holiday-season AI search traffic converting nine times more often than social media referrals. My ledger just watched that stat happen at street level, with the COD column as the tell.

So no, I have not stopped the brand work on social. I have stopped expecting social ads to be the checkout channel. In this store, they set the stage; the AI answer closes.

Is my store a fluke? What the big data says

One store, one niche, three months. I am the first to say that proves nothing on its own. So I checked the ledger against every serious dataset published on AI-referred commerce, and it lands right in the middle of a real fight.

The bull case is loud. Adobe has tracked AI-referred traffic to US retail growing 14x since October 2024, and its conversion quality flipped from 43% worse than other channels to 54% better. Shopify's own platform data shows AI-referred orders up ~13x year on year, converting about 50% higher on product pages. Similarweb puts ChatGPT referral conversion at 11.4% against 5.3% for organic search on ecommerce sites.

And the honest counterweight: a peer-reviewed study of 973 sites found ChatGPT still under 0.2% of ecommerce traffic, converting below organic search across 2024-25 data.

Both can be true. The stream is small. The stream is also getting better fast, and my store's version of that story is 80+ orders that almost all paid upfront.

Do not bet the company on AI search. Do refuse to ignore it.

How does ChatGPT decide what to recommend?

Everything above is the what. Here is the verified how, because most articles on this are guesswork.

ChatGPT's shopping results launched in April 2025, and OpenAI is unusually clear about the rules: results are organic, selected independently, and not ads. You cannot buy the slot. The only fee in the system is a reported ~4% when a sale completes through Instant Checkout, and normal recommendations carry no fee at all.

On your side of the fence, the machine needs to be able to read you. Shopify and Etsy catalogs are already integrated into ChatGPT shopping, and since late March 2026, eligible Shopify stores appear inside ChatGPT by default through agentic storefronts, with checkout on your own store. Non-Shopify merchants can submit OpenAI's product feed, refreshable every 15 minutes. And whatever you do, do not block OAI-SearchBot; blocked stores are invisible regardless of quality.

On the web's side, the engines look for proof you are real. Ahrefs studied 75,000 brands and found web mentions correlate with AI visibility roughly 3x more strongly than backlinks. For buying-intent queries, citations concentrate in review media, review platforms and YouTube, while forums like Reddit are under 2%, whatever the LinkedIn gurus say.

Being mentioned beats being linked. That is the doctrine shift, and it is measurable now.

The playbook I actually ran

Nothing here is exotic. That is rather the point: this is a few days of setup and a discipline, not a growth hack.

What was not on the list matters too. No paid AI placement exists, so nobody can sell you one. And no, an llms.txt file is not the trick here either; the levers are the catalog and the mentions.

What I cannot claim from this

A case study without limits is an advert, so here are mine.

It is one store, in one niche, in one market, over three months. Crystals are a category people naturally ask assistants about; a cement dealer will not replicate this curve. The revenue is small, and I have deliberately shown it.

I also cannot show you the ChatGPT answer that recommends us, because it usually does not show it to me. AI answers vary by user, session and phrasing, so a store owner searching for themselves proves nothing either way. That is exactly why I trust the order-level UTM data over screenshots, mine or anyone's.

And attribution has edges. A buyer who saw our Instagram first and asked ChatGPT later shows up as ChatGPT. The brand work and the AI visibility feed each other; I can separate the orders, not the influence.

Small store, loud signal, honestly labelled. That is the deal.

Should you optimise for ChatGPT in 2026?

If you sell online, yes, and the case is stronger than most of the industry realises.

The scale is already there: ChatGPT crossed 900 million weekly users this year, and Adobe's surveys have 39% of US consumers using generative AI for shopping. The traffic is still a sliver of totals, but it is the fastest-improving sliver in commerce, and Google is racing the same direction with AI Mode and AI Overviews.

More importantly, the buyers are different. Every dataset now agrees with what my COD column screamed: AI-referred buyers convert better because they arrive with the decision already made. For a small brand in a trust-poor market, that is not an incremental channel. It is the first channel where being genuinely good at answering questions beats having the biggest ad budget.

The work is unglamorous: a readable catalog, a consistent entity, mentions you earn rather than buy. Old fundamentals, new surface. Which is precisely why it favours honest small brands.

Final take

I built a store to settle an argument with myself, and the store settled it.

Ninety-one orders in three months. More than 80 from ChatGPT and Perplexity. One COD out of the lot, in a country where COD is the norm. Ads that bought attention, and AI answers that quietly closed the sales.

AI search is no longer a prediction. On my Shopify dashboard, it is a line item, and it pays upfront.