This week I asked Perplexity where to buy commercial catering equipment online in the UK.
Nine sources came back. All nine were sellers. No magazine, no review site, no forum.
Then I asked about the same kind of oven, but with a problem: why does a combi oven keep showing a descale error? The sellers disappeared. What came back was manuals, repair companies and a Q&A site.
Same engine. Same product. Two different questions, two completely different cast lists.
I repeated this across three product niches and two AI engines. 24 answers in total. The pattern held more often than I expected, and in one place my own hypothesis broke.
What most store owners believe about AI visibility
Here is what most of us assume: AI answers cite whoever looks most authoritative, so the way in is more magazine mentions, more Reddit threads, more links.
That is incomplete. In this sample, whether a shop got named depended less on the product and more on what the buyer was trying to do.
When the question was where to buy, the answers were lists of sellers. When it was which one is best, brands, models and sellers got mixed together. When it was why something is broken, the sellers mostly dropped out.
The cost of the old assumption is simple. You can spend months writing for a moment when shops are rarely named, and stay invisible in the moment when the buyer has a card in hand.
To be fair to the old view: magazines and forums did show up. A home improvement magazine appeared once, Reddit appeared once. They just were not the main story.
What I tested, and what this is not
In August I promised, in a comment under a Growth Memo study on AI citations, that I would run a similar analysis on ecommerce. This is the first look, not the full study.
3 UK product niches: commercial combi ovens, safety and workwear, designer radiators. They come from the kinds of stores I looked at this summer.
4 questions per niche, one for each buyer intent: best X, A vs B, where to buy, and a problem question.
2 engines: Perplexity and ChatGPT with search on.
12 prompts, 24 answers, one run each, on 5 October 2026.
Here are the four prompts I used for combi ovens, so you can see the pattern:
Best X: best combination oven for a small restaurant UK
A vs B: Rational vs Unox combi oven
Where to buy: where to buy commercial catering equipment online UK
Problem: why does my combi oven keep showing a descale error
The other 8 prompts follow the same pattern. Want the full list? Reply to this email and I will send it.
The limits, because you will ask:
24 answers is a pattern to test, not proof.
One run per prompt. The same prompt can give a different answer tomorrow.
I was logged in. ChatGPT ran in Temporary Chat, and one answer still came back in Croatian, so some personalisation leaks through.
I sorted the sources by hand. ChatGPT hides some of its sources behind a “+2” label, so what I counted for it is a floor. For that reason I do not compare the two engines as percentages.
I am not comparing my numbers with the SaaS numbers from that study. Different method, much smaller sample.
What came back, by buyer intent
Buyer intent changed who got named more than the product did.
Where to buy
Perplexity: sellers in all 3 niches. In catering equipment, 9 of 9 sources were sellers.
ChatGPT: shop lists in all 3 niches. 10 names for catering equipment, 6 for radiators (with a local map).
Best X
Perplexity: mostly sellers, some brands, one government page. A home improvement magazine in 1 of 3 niches.
ChatGPT: led with brands and models: Rational, Unox, Safety Jogger, uvex, Terma.
A vs B
Perplexity: a mix of blogs, retailers, manufacturers, one Reddit thread and one trade magazine.
ChatGPT: leaned on manufacturer sites: Rational, Unox, Dickies, Snickers, Fondital.
Problem (”why does...”)
Perplexity: manuals, repair companies, health sites, YouTube and a Q&A site.
ChatGPT: the NHS in the safety boots answer. The rest I could not read.
In plain words: sellers get named when the buyer wants to buy. Brands get named when the buyer wants to choose. Sellers mostly step aside when the buyer has a problem.
One exception is worth remembering. A radiator retailer did show up in the “why is my radiator cold at the top” answer on Perplexity. A good help page from a shop can win a problem question. It happened once in 3 problem answers, so I would not build a strategy on it yet.
Where my own hypothesis did not hold
In that August comment I guessed that ecommerce would look like SaaS on one side and different on the other. Sellers and manufacturers would keep a big share, but the community layer would shift: marketplace Q&A and YouTube would take the place that Reddit holds in SaaS.
The first half held. Seller and manufacturer sites made up most of the sources in the buy and best-of answers, and a large part of the comparisons.
The second half is not confirmed. Reddit appeared in 1 of the 12 Perplexity answers. YouTube appeared in 2. I saw no marketplace Q&A page at all. With numbers this small, that is a non-result, not a refutation.
My honest reading: either community sources matter less in physical product niches than I guessed, or they only show up in prompts I did not run. The next step is a bigger sample with repeated runs, so I can tell the difference.
What this means if you run an online store
If sellers get named when the buyer is ready to buy, your question gets very short: when someone asks an AI where to buy what you sell, is your shop on the list?
I should be upfront. I am the founder of Lobit, an agency that does ecommerce SEO, so I have an interest in you caring about this. But the 24 answers above hold whether or not you ever hire anyone. And you can check your own store in five minutes.
The 5 minute test: is your store on the list?
Open Perplexity and ChatGPT (use Temporary Chat in ChatGPT). Ask: “where to buy [your main product] online UK”. Swap in your own market if it is not the UK.
Write down who is on the list. Is your shop there? Which sellers are ahead of you?
Ask two more: “best [your main product] UK” and “why does my [product] [most common problem]”.
Score yourself. In how many of those 3 prompts does your shop appear, on each engine? That gives you 0 to 6. It is a crude version of what I track as AI Visibility Rate.
If your site is verified in Bing Webmaster Tools, open the AI Performance report and look at Citation Share and Intents.
Open yoursite.com/robots.txt and check that you are not blocking AI crawlers such as GPTBot or ClaudeBot by accident. Some shops do it through a CDN setting without knowing.
It takes five minutes, and for most shops it ends with a very specific list of who is beating them.
Questions you are probably asking
Is 24 answers enough to trust this?
No. It is enough to see a pattern worth testing, and not enough to quote as a percentage. That is why I call this a first look.
Will I get the same answers if I run the prompts myself?
Probably not exactly. AI answers shift between runs, and between logged-in and logged-out use. I ran each prompt once, so a repeat run is the first thing I want to add.
Does this mean helpful content and reviews are wasted effort?
No. Problem questions got answers too, from manuals, repair companies, health sites and one shop’s help page. Helpful content is a way into answers. It is just a different set of answers than the ones where people are ready to buy.
What should I do first?
Run the 5 minute test above. If you are missing from the where-to-buy list, start with the basics: AI crawlers are not blocked, and your category pages say clearly what you sell.
Want me to run it on your store?
Send me your store URL and your main product category. I will run these same prompts for your niche on both engines, and tell you in plain language who gets named instead of you and what looks fixable.
That is the free mini-analysis. If I find nothing worth fixing, I will tell you so, and you keep the notes anyway.
Ready now?
Not ready yet? Stay subscribed. Every Tuesday you get one practical tactic for being visible in Google and in AI assistants, with real audits and real numbers behind it.
Neven Lovreković
P.S. The one thing to take away: AI does not name shops equally everywhere. It names them when the buyer is ready to buy. Run the 5 minute test, and if your shop is not on the list, reply with your URL and I will look at why.

