Can AI Shopping Assistants Read Your Product Catalogue? What Singapore SMEs Must Fix
An AI shopping assistant will only recommend your products if it can read them as text. Agentic assistants - ChatGPT with browsing, Gemini, Perplexity, Copilot and the shopping agents now built into them - do not look at your website the way a customer does. They pull structured facts: one stable page per product, a price in a machine-readable field, stock status, written specifications, delivery terms. If your catalogue is a downloadable PDF, a gallery of product photos, or a WhatsApp chat with your sales staff, you are invisible to that entire channel. The fix is not an AI project and does not need an AI budget. It is product data housekeeping, and a small company can get most of the way there in two to three weeks of ordinary staff time.
What is an AI shopping assistant actually doing when it looks for a supplier?
The pattern is consistent across the major assistants. A buyer asks a shaped question - a 600mm induction hob for a commercial kitchen in Singapore under 2,000 dollars, or a BizSafe certified aircon servicing contractor in the west. The assistant runs several searches, fetches a handful of pages, and tries to extract comparable attributes from each: what it is, what it costs, whether it is available, who sells it, and whether the claim can be cited. It then presents three to five options with a source link.
The important part is what happens to anything it cannot extract cleanly. The assistant does not guess, and it does not call you. A product with no readable price simply drops out of the comparison, because it cannot be placed next to a competitor that has one. Being unreadable is not a lower ranking. It is absence.
Why does your product page look empty to an AI agent?
Most Singapore SME catalogues fail on the same handful of points, and almost none of them are about design quality. A page that looks perfectly good to a human can carry no machine-readable content at all.
- The price exists only inside an image, a brochure, or a line asking the buyer to enquire.
- Specifications sit in an attached spec sheet rather than typed on the page.
- Forty products share one long category page, so there is no individual URL to cite.
- The page renders only after JavaScript runs, and the fetched version is blank.
- Dimensions and capacities appear without units, so 300 by 600 could be millimetres or anything else.
- Variants - size, finish, voltage, pack quantity - are selectable but never described in text.
- Your firewall or robots file blocks the AI crawlers outright, which many default security settings now do.
What does an AI assistant need from a Singapore SME catalogue?
Treat it as a short list of facts that must be present as text on a page the assistant can actually fetch. For each product you genuinely want to sell: a name buyers use rather than your internal code, a single indexable page, a price in Singapore dollars with the GST position stated plainly, availability and a realistic lead time, and a typed specification table. Add schema.org Product and Offer markup so the price and stock fields are unambiguous, which is the same markup Google Shopping already reads, so the work serves two channels at once.
Then make your company details consistent. Business name, address, service areas, opening hours and contact number should match between your website, your Google Business Profile and your invoices. Assistants cross-check these, and a mismatch between your site and your profile reads as a confidence problem. Review volume matters here too, because a cited third-party rating is one of the few trust signals an agent can verify quickly.
Does this matter if you sell B2B and never publish prices?
Yes, and arguably more. Buyers are using assistants to build shortlists, not to check out. A procurement executive asking which Singapore suppliers can fabricate a particular bracket in small volumes is not expecting a price - they are expecting three names and a phone number. If your site cannot answer the shortlisting questions, you are not on the list that gets contacted.
You can stay commercially discreet and still be readable. Publish indicative ranges or a from-price, minimum order quantity, typical lead time, certifications such as BizSafe or relevant SS standards, brands and materials you carry, and the geographies you deliver to. That is enough to be compared and quoted, and it does not expose your margin.
What should you fix before the December buying window?
Do not start with the full catalogue. Start with the twenty products or services that produce most of your revenue, in this order.
- Give each of the twenty its own page with a buyer-facing name.
- Put the price on the page in text, with one line stating whether GST is included.
- Type the specifications out, with units. Keep the PDF, but stop depending on it.
- State stock position and lead time, and assign one person to keep it true.
- Add Product and Offer structured data, which is usually under a day of developer time on WordPress, Shopify or WooCommerce.
- Check that your robots file and your web firewall are not blocking assistant crawlers.
- Test it. Ask two different assistants a buyer question in your category and see who gets named. Then ask about your own product by name and check whether the price and availability it reports are correct.
What does this cost, and what is the risk of leaving it?
The cost is mostly internal time - roughly two to four hours per twenty products for the content, plus a developer day for markup and a product feed. The risk of leaving it has two parts. The first is silent absence: there is no bounce rate, no abandoned cart and no analytics line for a recommendation you never appeared in, so nothing in your reporting will tell you it is happening. The second is worse, which is being read wrongly. An assistant that finds a stale price quotes it confidently to a buyer, who then arrives expecting it, and you either absorb the difference or start the relationship with an argument.
One practical warning for December. If you plan a price revision in January, fix the data now and schedule the change properly, because assistants cache and cite what they found. A price that was accurate in November will keep being repeated into February unless the page it came from is updated.
Frequently asked questions
Do I need to add an AI chatbot to my website for this?
No, and a chatbot does not help. The assistants read your pages, not your widget. A chatbot answers the visitor who already found you; structured product data is what gets you found in the first place. Spend the budget on the catalogue.
Will blocking AI crawlers protect my content?
Blocking stops you being recommended. It does not stop competitors from being recommended instead. If your concern is price scraping, the realistic position is that your published prices are already public, so the trade is between visibility and a protection you mostly do not have. Most SMEs selling standard products should allow the crawlers.
How do I know whether buyers in Singapore are really using this?
Check your own referral traffic for assistant sources, which now appear separately in most analytics tools, and ask your sales team how recent enquiries described finding you. A buyer who arrives already naming a model and a price range, with no record of a site search, has usually been given a shortlist by something.
Ready to Transform Your Business?
Let Digital Perpetual help you automate, streamline, and grow.
Get Started with Digital Perpetual →