Your Customer Asked ChatGPT Before They Asked You: The AI Shelf Test
Here is a conversation you will never see in your analytics. A shopper in Singapore opens ChatGPT and types: “What’s a good local skincare brand for sensitive skin?” The AI names three brands. Yours is not one of them. The shopper never visits your site.
This is not rare. 72% of Singaporeans have used AI assistants to help with shopping, according to Adyen’s 2026 survey of 1,041 Singaporean adults. Some of that shows up in your analytics as visits from chatgpt.com or perplexity.ai. Most of the decision happens before anyone clicks at all.
You cannot fix what you cannot see. So start by looking.
The AI shelf test
Think of an AI answer as a shelf. When a shopper asks a question, the AI puts a few brands on that shelf. This test tells you whether you are on it, who is next to you, and what the label says. It takes about 30 minutes.
1. Write five real customer questions
Not keywords — questions, the way a real person would ask them. Mix the types. Include “in Singapore” in at least a few, because that is how local shoppers ask. Swap your own category into these:
- What’s the best place to buy [product] in Singapore?
- Which [product] brands are made or based in Singapore?
- Is [your brand] worth it compared to [competitor]?
- Where can I get [product] delivered in Singapore this week?
- What should I look for when buying [product]?
The third question uses your brand name. That one matters. It tells you what AI says when someone is already considering you.
2. Ask each one in three AI tools
Use ChatGPT, Gemini and Perplexity. Open a private or logged-out window, so your own history does not shape the answer. Ask each question twice — AI answers change from one try to the next, and one run can mislead you.
3. Write down what you see
Were you named? Who was named instead? What did it say about you — and is any of it wrong or out of date? When it linked somewhere, was it your store, or a marketplace listing?
4. Look at where the answer came from
Perplexity shows the pages it drew on, and the others sometimes do. Those sources are the most useful part of the test. They show you what the AI trusts when it talks about your category.
What the results usually mean
Once you have your notes, the gaps tend to fall into a few patterns.
You are not named at all
Often your site does not say plainly what makes you different. If your homepage says “quality products at great prices”, the AI has nothing specific to repeat. Your competitor who says “hand-poured in Singapore, delivered in two days” does.
You are named, but described wrongly
The AI is repeating something old or wrong — a past price, a product you stopped selling, a delivery promise you changed. Check the sources it cited. The fix usually starts there, not on your own site.
Other sites describe you better than you do
If the sources are review sites, forums and marketplace pages rather than your own store, the AI is learning about you second-hand. Your own site should be the clearest, most complete answer about you anywhere online.
The AI sends people to your marketplace listing
This one is easy to miss, because you still get the sale. But you get it on someone else’s platform, in a S$25 average basket on Shopee, without the customer’s email. If AI points to your listing instead of your store, your store is not giving it a better answer.
You are not trying to trick the AI. You are trying to be the clearest, most accurate answer about your own business.
How to show your boss
If you manage a store for a founder, this test gives you something rare: a clear picture of a problem nobody else on the team can see. Put it in one table — one row per question, per tool.
| Question | Tool | Named? | Named instead | What it said / fix |
|---|---|---|---|---|
| Best place to buy [product] in Singapore? | Perplexity | No | Competitor A, Competitor B | Cited a 2024 listicle we are not in. Fix: say what makes us different on the homepage. |
| Is [our brand] worth it? | ChatGPT | Yes | — | Quoted an old price. Fix: update the product page and our marketplace listing. |
Three things make this land with a founder:
- Lead with the competitors who were named. Nothing gets attention faster.
- Call it a baseline, not a verdict. Run the same questions every month, and show the change.
- Pair every gap with a fix. A problem with a next step reads as good management. A problem alone reads as bad news.
What to fix first
- Say plainly what makes you different, on your homepage and your key product pages.
- Put your key facts where AI can find them: delivery times in Singapore, prices, where you are based, what you actually sell.
- Correct anything wrong at its source — especially old marketplace listings and review profiles.
- Make your own store the best answer about you, so AI has a reason to link there instead of a listing.
The technical side — product schema, reviews, and the files AI tools read — is covered step by step in our Shopify AI search optimisation guide. For the bigger picture of how Singaporeans shop across AI and marketplaces, start with how Singapore shoppers actually buy in 2026.
Want a second pair of eyes on your results? Book an introductory call and bring your table.
Sources
- Adyen Index 2026 Singapore Retail Report. Conducted by YouGov, March–April 2026. 1,041 Singaporean consumers aged 18+.
- Cube, “E-commerce in Singapore”, June 2026, Tradewinds Q4 2025 dataset. Shopee average order value, FY2025, physical goods only.
Jessica Bong
Jessica Bong is the founder of Soodo, a Singapore-based Shopify development and CRO agency. She built and scaled her own eCommerce brand before starting Soodo, and has since audited 60+ Shopify stores. Jessica also teaches eCommerce at Equinet Academy. Her hands-on experience running a live brand gives Soodo an edge most agencies lack.
