Method
Each month we send a cast of simulated shoppers to the big AI assistants and record the shelf each one is shown. Real buyer types, moving through a real purchase, watched from August to January.
Here is how it works, and why the season matters.
1 · The shoppers
AI personalises. Ask the same thing as two different people and you get two different shelves. So we don't test with generic queries. We build behavioural personas: real buyer types with a life stage, needs and constraints. One set for every category and market, and they stay fixed for the whole season so changes between readings are the assistant's, not ours.
One of them is nobody in particular: the Anonymous shopper, a new or signed-out user the assistant knows nothing about yet. That is the cold-start shelf, and to a brand it is a real, targetable customer.
A few of the cast
The Reinventer
57, done chasing trends — rebuilding a routine for skin that's changed.
The Aesthetic Athlete
23, trains five days a week — gear has to look as good as it performs.
The Apartment Cat Parent
31, one indoor cat, a small apartment — smell, urinary health, boredom.
The Kidult Collector
25, buys toys for himself — cares if it's authentic and holds resale.
The Gym Starter
21, six months into lifting — first supplements, student budget.
The Empty-Nest Cook
58, cooking for two again — premium gear that lasts, buy once.
The full cast spans shopper generations, from Gen Z to boomer, plus need-specific types where a category calls for one (the indoor-cat parent, the fresh-food dog owner). Every category and market has its own.
2 · The journey
A real purchase is never one question. A shopper starts curious and ends at the checkout, and the shelf shifts under them the whole way. We follow each persona through all five stages, asking the kind of thing a shopper asks at each, and record which brands show up where.
Just realising they need something. Open, unattached, casting wide.
Digging in: learning the category, the vocabulary, what actually matters.
Weighing a shortlist against itself. This one, or that one?
Gut-checking a near-decision: is it worth it, is it trusted, will it last?
Ready to buy: which exact one, from where, at what price.
Different shoppers lean on different stages. A premium buyer lingers in validation; a deal-hunter jumps to purchase. The gap between the brand an assistant introduces and the brand it rings up is one of the most telling things we see.
3 · The season
This is the shopping year in miniature: back-to-school, the holiday build-up, Black Friday and Cyber Monday, then the January reset. Demand does not sit still, so neither does the map. We take a fresh reading every week and compare it to the last. Three things move, and we watch all three.
The shoppers change
Needs and questions shift with the calendar. Gifting climbs toward December; deal-hunting peaks over Black Friday weekend; January turns to resets and resolutions.
The assistants change
Models get updated without notice. Brands rotate on and off the shelf, and how quickly each assistant notices a real discount is itself a signal worth timing.
The shelf moves
A brand that owned a category in August can quietly disappear by December. Because we read the same shoppers every week, we catch the moment it happens.
It is a live simulation: real-life shoppers, the journeys they take, and the shelf AI builds for each, traced week by week.
4 · The markets
A category is not a market. A skincare shopper in Sydney and one in Toronto aren't asking the same thing at the same time: Christmas is high summer for one, deep winter for the other. So every persona is built for a single country, with that country's calendar, climate and habits baked in.
United States
A November cliff. Everything funnels into Cyber Five (Thanksgiving to Cyber Monday), then a gift-card and New-Year-reset tail into January.
United Kingdom
No Thanksgiving, so the season opens with the Christmas adverts and runs on nerve: shoppers hold out for the deepest discounts, right to Boxing Day.
Canada
Black Friday is imported, but it's the day. After two years of postal strikes, "will it arrive by Christmas" is part of the buying decision here in a way it isn't elsewhere.
Australia
Christmas in summer. Sun care, swimwear and the BBQ peak on Dec 25, in one long arc from October to Australia Day. The northern-hemisphere calendar simply doesn't apply.
We never transpose one market's season onto another. A winter-immunity supplement curve that fits the US is wrong for Australia, where that peak lands mid-year, outside this window entirely.
5 · The measures
Every brand here carries two readings.
Visibility
How often AI names the brand at all: of every shelf we record, the share that mention it. It answers one question. Does the assistant know you exist?
Shelf-share
How much of that shelf the brand actually holds. A shopper asking for a recommendation has finite attention. The first name lands, the fifth rarely does, and the brand named as the pick is not in the same position as one listed below it. So we treat each shelf as a whole that divides among the brands on it, weighted by where each one lands, and count only the positions a shopper realistically takes in.
The two move independently, and the gap between them is the point. A brand can be named constantly and still hold almost none of the shelf: always in the conversation, never the recommendation. Another can be named half as often and hold more, because when it does appear, it appears first. Visibility on its own cannot tell those two apart.
There is one more difference, and it is the reason we lead with share. A shelf is finite, so share is contested: a brand gains only when another gives ground. Visibility has no such ceiling. Every brand in a category can grow more visible in the same month, and in a growing category most of them do.
We describe how the method works, never a playbook to game it: the stages a shopper moves through and how we read the shelf, not the wording behind any single reading.
See which shoppers already find you on the shelf, and where you are missing.