We analyzed how ChatGPT and DeepSeek recommend gold jewellery brands in Singapore

AI has a shortlist of Singapore jewellers, and it's short

We tracked nine jewellers. Three of them, POH HENG, SK Jewellery and Goldheart, are in almost every answer. TAKA is a coin flip. Tomei turned up twice.

Who AI names when you ask where to buy jewellery in Singapore

POH HENG100%

SK Jewellery97%

Goldheart96%

TAKA Jewellery55%

Tomei1%

PKPoh Kongnever named

CTClock Towernever named

PAPandoranever named

SWSwarovskinever named

Who AI names when you ask where to buy jewellery in Singapore

Jeweller	Answers naming it	Share
POH HENG	139 of 139	100.0%
SK Jewellery	135 of 139	97.1%
Goldheart	134 of 139	96.4%
TAKA Jewellery	76 of 139	54.7%
Tomei	2 of 139	1.4%
Poh Kong	0 of 139	0.0%
Clock Tower	0 of 139	0.0%
Pandora	0 of 139	0.0%
Swarovski	0 of 139	0.0%

Then there are the zeros. Poh Kong, Clock Tower, Pandora and Swarovski didn't appear in a single answer. These aren't obscure names. Pandora and Swarovski have stores in the same malls. But when a shopper asks an AI where to buy jewellery in Singapore, they simply aren't part of the conversation.

The two models mostly agreed, too. DeepSeek named SK every time; the OpenAI model did in 94% of its answers. This isn't one model's quirk.

What this means for you. Before you worry about what AI says about you, check whether it says your name at all. For four well-known brands here, the answer is no, and no amount of clever messaging fixes absence.

Being named isn't the same as being picked

Here's where SK's good news gets complicated. Named in 97% of answers. Actively recommended in 72%. And when we asked for a ranked top five, 140 shortlists in all, SK made the list about six times in ten and was never the first pick.

When SK does appear, it's usually near the bottom of the list, described as a sensible place to compare: mall stores, modern designs, 999 gold. All true, all pleasant, none of it a reason to go there first.

What this means for you. Track three numbers, not one: how often you're named, how often you're recommended, and how often you're first. Most brand dashboards stop at the first.

AI sends ring buyers and gold savers somewhere else

So who gets the top spot? It depends entirely on what the shopper wants, and in each case AI has a clear favourite.

For engagement rings it's Love & Co. or JannPaul, 49 times out of 50. Love & Co. is SK's own sister brand, so the group isn't losing that shopper, but SK Jewellery itself isn't in the running. JannPaul wins on something specific: the AI keeps praising its diamond education and cut-quality detail.

For people buying gold as savings, AI barely recommends jewellers at all. BullionStar and UOB take four of every five first picks, because the model thinks a saver wants live prices, tight spreads and buyback, and it says so. For wedding sets and Peranakan gold, POH HENG leads comfortably.

What this means for you. AI doesn't rank brands in general. It ranks them for a need. Find the two or three needs you should own, then look at exactly what AI praises the current winner for. That's your brief.

AI knows what SK sells. It doesn't know why you'd choose it

We looked at what AI actually says when it mentions SK. The picture is consistent: contemporary designs, 999 pure gold, mall stores, good for gifting. Accurate, and completely generic.

Now look at what it almost never says. Buyback came up in about one in ten mentions. Trade-in and upgrade value, 7%. Certified diamonds, once. Daily gold price, twice. Trusted family jeweller, never. These are the things shoppers in this category ask about before they hand over money, and AI doesn't connect any of them to SK.

What this means for you. Write down the five reasons someone should choose you. Then check whether AI ever gives those reasons. The gap between your list and AI's list is your content plan.

We tried one fix. It barely moved anything, and that's useful

SK's website says its gemstones are ethically sourced. AI never mentions it. So we ran an experiment: take AI's own description of SK, add a short ethical-sourcing line, hand each version to the model as background, and ask for shortlists again.

The shortlist didn't budge: SK made the top five 63% of the time before and 61% after. Still never first.

We also wanted to know how shoppers might take it. So we had the model role-play six kinds of jewellery buyer, from a first-time ring shopper to a retiree buying gifts, and react to each version in their own words. Then we scored how interested each reaction sounded, out of five. Overall interest went from 3.20 to 3.24. A nudge, not a shift.

The breakdown is more interesting than the average. The gold saver and the anniversary-upgrade buyer warmed up the most. The parent buying wedding gold cooled slightly. But read what they said and one thing jumps out: nobody talked about sourcing. They asked for today's gold price, workmanship charges, buyback terms, grading certificates.

What this means for you. A values message on its own won't move you up an AI shortlist. Pair it with the specifics that buyer is already asking for, and test the pairing before you rewrite the website.

Ask the same question two ways and you get two answers

One last thing, and it matters for anyone pasting an AI answer into a slide. We asked about engagement rings in several slightly different ways. With “places to consider”, SK appeared in two of five shortlists before our change and none after. With “rank the leading spots”, it went the other way: none before, two after.

Same shopper, same need, same brand. The wording of the question moved the result more than our message did. That's why we ask every question five times and ask it several ways. A single AI answer is an anecdote.

What this means for you. Never judge your AI visibility from one prompt. If an agency shows you a single screenshot, ask how many times they ran it.

If I ran marketing at SK Jewellery

You already have the hard part: AI knows who you are. The job now is giving it reasons to put you first. Three things I'd do, in this order.

1. Own one need before you chase them all

   Pick the shopper you most want, say the gold saver, and build the page AI is looking for: purity, today's price, workmanship charges and buyback terms, all in one place and kept current. Right now AI sends that shopper to a bullion dealer because the dealer answers those questions and you don't.

2. Turn your proof points into things AI can repeat

   Buyback, trade-in, certified diamonds: if they're real strengths, write them down plainly, with specifics, where AI models can find them. “Trusted family jeweller” showing up zero times tells you the story you tell in-store isn't making it online.

3. Use sourcing as backup, not the headline

   Ethical sourcing is worth saying. But our test suggests it reassures rather than persuades. Put it beside the certificate and the price, not instead of them, and test the pairing before you roll it out.