6 min read
This is an illustrative story. Maren and her shop are a composite built from common situations small online sellers describe, not a real business, and the numbers are examples rather than reported results. The Shopify features mentioned are real.
The email arrived on a Tuesday morning in early November. “Hi, I asked ChatGPT for a caffeine-free tea for my mum and it suggested your Winter Ember blend. She was up until 2 AM. Is it actually caffeine free?” Maren read it twice, then winced. Winter Ember is a smoky black tea with orange peel and clove. It has plenty of caffeine. Nowhere on her site did she say otherwise. But nowhere did she clearly say it, either.
A one-person tea shop that was already inside ChatGPT
Maren runs a small loose-leaf tea shop on Shopify from a converted spare room: about 40 blends, a handful of gift sets, and a holiday season that brings in close to a third of her yearly sales. She had never set up anything for AI shopping assistants, so she assumed they did not know she existed.
She was wrong. When Shopify launched Agentic Storefronts in March 2026, it said merchants’ products become discoverable in ChatGPT by default, with no separate apps, and the program now reaches ChatGPT, Microsoft Copilot, AI Mode in Google Search and the Gemini app, all managed from the Shopify Admin. Modern Retail covered the shift as purchases moving inside the chat itself. In other words, AI assistants had been reading Maren’s product listings for months. They were just reading listings that had been written for humans who would look at the photo, not for software that only reads the words.
What the AI assistants were getting wrong
Maren spent an evening asking ChatGPT, Gemini and Copilot the questions her customers might ask: “best caffeine-free tea gift under $30,” “a smoky black tea for someone who likes Lapsang,” “loose-leaf tea that ships before Christmas.” She kept notes on every answer that mentioned her shop. Three problems showed up again and again:
- Missing facts got filled in with guesses. Her descriptions were poetic (“a fireside evening in a cup”) but rarely stated caffeine level, so assistants inferred it, sometimes badly.
- Inconsistent names confused the matching. The same blend appeared as “Winter Ember,” “Winter Ember Blend” and “Ember (Winter Edition)” across her product title, a collection page and an old blog post.
- Gift sets had no contents list. The “Cozy Sampler” description said “four of our favorites” without naming them, so assistants could not recommend it for specific tastes.
Shopify’s own announcement says AI surfaces favor structured data with clean attributes and real-time accuracy. Maren’s listings had personality, but very little structure.
The fix: one weekend of product data cleanup
Maren did not rewrite her brand voice. She added a factual layer underneath it. Here is the routine she followed, which any small product seller can copy.
1. A fact sheet for every product, drafted with AI
She exported her products to a spreadsheet and gave an AI assistant one row at a time with this prompt: “From this product description and ingredient list, fill in these fields: tea type, caffeine level (none, low, medium, high), main ingredients, flavor notes, steep time, water temperature, net weight, best for. If a field is not stated in my text, write UNKNOWN instead of guessing.” That last sentence mattered. Every UNKNOWN became a question only she could answer, and she answered all of them from her own supplier sheets.
2. One name per product, everywhere
She chose a single official name for each blend and updated product titles, collection pages and old blog posts to match. Winter Ember is now just “Winter Ember Black Tea,” and the caffeine level sits in the first line of the description.
3. Plain facts first, poetry second
Each description now opens with a short factual sentence (“Smoky black tea with orange peel and clove. High caffeine. 50 g, about 20 cups.”) followed by her original, more lyrical copy. Humans still get the story. Software gets the facts first.
4. Gift sets that list their contents
Every bundle now names each tea inside it, with its caffeine level, so an assistant can confidently recommend “the sampler with two caffeine-free blends” to someone shopping for a relative.
5. Shipping cutoffs stated in words
She added a clear holiday shipping cutoff to her shipping policy and product pages, because “ships before Christmas” was one of the questions assistants could not answer about her shop.
Testing it the way a customer would
Two weeks later, Maren reran her list of questions in all three assistants. Winter Ember was described as a high-caffeine black tea every time. Her two genuinely caffeine-free blends started appearing in “caffeine-free gift” answers. The Cozy Sampler was suggested by name for “a tea lover who likes variety.” Not every answer was perfect, and AI results shift from day to day, but the wrong claims about her products stopped appearing in her tests.
She also started checking her Shopify Admin for orders arriving with AI channel referral attribution, which Shopify says these orders carry. In this illustrative example, a handful of holiday orders a week now trace back to AI assistants, a small number for now but one that costs her nothing extra to keep.
What Maren would tell another solo seller
- Assume the AI assistants are already reading your store. If you sell on Shopify, the company says ChatGPT discovery is on by default.
- Write down the facts a customer would ask about, such as size, materials, allergens, caffeine or compatibility, in plain words near the top of each product page.
- Tell your AI helper to say UNKNOWN instead of guessing when it drafts product data. Then fill the gaps yourself.
- Use one name per product across every page that mentions it.
- Test monthly by asking the assistants the questions your customers ask, and keep a simple log of what they get wrong.
Her cleanup took one weekend, and it also improved her own site search and her email product blocks, the same kind of tidy data that makes a clean customer list pay off at holiday time. If you want the bigger picture on why this matters right now, our look at the forecast jump in AI holiday traffic covers where to spend the weeks before Thanksgiving.
Your turn: one product, ten minutes
Pick your best-selling product tonight. Ask ChatGPT or Gemini one question a real customer might ask about it, and see whether the answer is right. If it is not, fix the first line of that product’s description so the fact is stated plainly. The AI can only recommend you accurately if your listings tell it the truth in plain words. What did the assistant get wrong about your products? Share it in the comments.
