The AI shelf changes with the question
By Malin Gaertig · October 9, 2026
Originally published in AdNewsA piece I wrote for AdNews: we tested three established Australian appliance retailers in AI recommendations. The most striking result came from a retailer that wasn't in the competitive set at all — Appliances Online was named in 99% of answers. The AI shelf is query-led, which makes the competitive set fluid.
- —Appliances Online was named in 99% of AI answers we tested — a competitor that wasn't in the competitive set at all.
- —AI recommendations aren't fixed: the competitive set changes with the model and the wording of the question.
- —On availability questions, retailers outside the original set appeared in 75% of answers, including specialists and marketplaces.
- —The AI shelf is query-led: the competitor is whoever the model believes can best answer this particular question.
- —Like Auction Insights before it, AI reveals the competitive set actually competing for the customer — earlier in the journey.
| Dimension | Traditional competitor analysis | AI recommendations |
|---|---|---|
| Competitive set | Stable, defined internally and tracked over time | Fluid, reshaped by the model and the customer's question |
| Who the competitor is | The business selling the same products in the same channels | The business the model believes can best answer the question |
| Measurement problem | Track share against known rivals | Track whoever the model puts in front of the customer |
Most retailers know their competitors and track them. But when a customer asks AI where to buy a fridge, the competitive set can look very different from the one in the reporting dashboard.
We tested three established Australian appliance retailers — Harvey Norman, Bing Lee and The Good Guys — to see how they appeared in AI recommendations.
The results were strong for The Good Guys and Harvey Norman, which were mentioned in 82% and 80% of answers respectively. Bing Lee, on the other hand, only appeared in 28%.
But the most striking result came from a retailer we weren't tracking as part of the competitive set: Appliances Online was named in 99% of the AI answers we tested.
It wasn't an occasional appearance. It was consistently being put in front of the customer alongside, and often ahead of, the retailers we had defined as competitors.
That is the more important finding than any individual ranking. The AI shelf was showing us a competitor that wasn't in our competitive set.
Why is the AI shelf bigger than the competitor list?
Being named isn't the same as being cited. The models differed on who they put first, with OpenAI heavily favouring Appliances Online, Gemini shifting towards it as the wording changed, while Claude more evenly split between Appliances Online, Harvey Norman and The Good Guys.
That variation tells us that AI recommendations aren't fixed, but that the competitive set changes according to the model and the question, with some brands consistently present and others moving in and out.
That creates a different measurement problem. A retailer can know exactly how it performs against Harvey Norman, Bing Lee and The Good Guys and still miss the business AI is repeatedly putting in front of the customer.
In other words, the competitor set a retailer measures isn't necessarily the brand set its customer sees.
How does the question change the competition?
The effect becomes even more pronounced when the customer's question changes. On a generic 'best retailer' question, retailers outside the original competitive set were almost absent.
Ask about real-time availability, however, and they appeared in 75% of answers. Interestingly, they weren't just the usual national retailers. The interceptors included Signature Appliances, Kogan, JB Hi-Fi, BIG W, Bi-Rite, Lucky White Goods and MyHouse.
These businesses don't necessarily compete with the established retailers in the same way. Some are specialists. Some have a particular service proposition. Some operate very different business models. But they become relevant when they have information that helps answer the customer's question.
Claude, for example, put Harvey Norman first in every availability answer, but on the generic 'best retailer' question, it put Harvey Norman first in none of the 25 answers.
Gemini behaved almost the opposite way, putting Appliances Online first in every availability answer in both runs.
A different question on a different model reveals different competitive landscape.
Why isn't this just a visibility problem?
Traditional competitor analysis assumes the competitive set is relatively stable. You decide who matters, measure performance against them and track whether you're gaining or losing ground.
AI, on the other hand, starts with the customer's question and looks for evidence that can help answer it. That means the competitor isn't necessarily the business selling the same products, operating in the same channels or sitting on the internal competitor list, but the business the model believes can best answer this particular question.
For a fridge retailer, that could mean competing against Appliances Online when the customer asks where to buy. A specialist when they ask about availability. A marketplace when they ask about price. The conclusion, therefore, is that the AI shelf is query-led, which makes the competitive set fluid.
A retailer can protect its position against the competitors it knows while is exposed this blind spot. This, of course has commercial implications when the prospect has more relevant options to shop from while only seeing a select few options.
Have marketers seen this problem before?
Marketers have seen a version of this before. Google Ads' Auction Insights showed retailers which other advertisers were appearing in the same auctions. It often exposed brands, marketplaces and specialists that weren't considered direct competitors but were still competing for the same customer.
Search revealed the competitive set that was actually competing for attention. AI recommendations are creating a similar moment, but earlier in the journey. The competition has moved upstream and now the game is about making it onto the curated shortlist. This leaves brands wondering who does AI add to the shortlist when their customers ask what to buy?