For twenty years, winning in eCommerce meant winning positions: the first Google result, the first row on Mercado Libre, the listing with the Buy Box. That game is still on, but a new shelf has appeared that almost no brand is measuring.
NielsenIQ sums it up with a simple comparison. On the physical shelf, shoppers see more than 50 SKUs. On the digital shelf, 5 or more. On mobile, 4 or more. On the shelf AI controls, just 1 or 2 (NIQ, E-commerce Unlocked Mexico, June 2026).
When someone asks ChatGPT “what supplement should I give my kid who’s a picky eater?” or “what’s the best sugar-free juice for a school lunch?”, the answer isn’t a list of ten links. It’s two or three brands, with an explanation of why. If yours isn’t there, you didn’t lose a position: you lost consideration.
We call that the AI Shelf.
What the AI Shelf is (and what it isn’t)
The AI Shelf is a brand’s presence and preference inside AI assistant answers (ChatGPT, Gemini, Google AI Overviews and AI Mode, Perplexity, Copilot, Claude, Meta AI) when the user has purchase intent in your category.
It’s not the same as “showing up in ChatGPT”. A brand can show up and not be chosen. It can be chosen and described with outdated information. Or it can be recommended while the link sends the shopper to a reseller that isn’t yours.
That’s why we separate four questions:
- Does it name me? Presence: the brand appears in the answer.
- Does it pick me? Preference: the brand is the first recommendation, or the only one.
- Does it describe me correctly? Accuracy: price, pack size, benefits and claims are right.
- Does it send people to buy somewhere that works for me? Path to purchase: official store, the right retailer or a third party.
Why it matters now, not in two years
In Mexico, 77.2 million people already shop online and the digital channel accounts for 17.7% of retail sales, according to AMVO’s 2026 Online Sales Study (Marketing4eCommerce, March 2026). That shopper already uses AI assistants to research before buying. In the US the shift is already measurable: traffic to US retail sites from generative AI tools grew 693.4% year over year during the 2025 holiday season, according to Adobe Analytics (Adobe, January 2026).
And the shift toward AI that not only recommends but also buys has already started. Google, Shopify and a group of retailers launched the Universal Commerce Protocol (UCP) in January 2026, and OpenAI and Stripe are pushing the Agentic Commerce Protocol (ACP) (Findskill, April 2026). When the assistant can close the sale, the AI shelf will also be the point of sale.
Brands that measure their AI Shelf today will have 12 to 18 months of history when that happens. Everyone else will start from zero.
The 6 AI Shelf metrics
These are the metrics we use at Balloon. They’re measured the same way on every assistant, with the same prompt set, on the same day, logging the model version.
| Metric | What it answers | How it’s calculated |
|---|---|---|
| AI Share of Voice | What share of answers mention my brand? | Answers with the brand ÷ total category answers |
| First-recommendation rate | How often am I the number-one option? | Answers where the brand is first ÷ answers where it appears |
| Accuracy | Does AI say true things about me? | Correct claims ÷ verifiable claims |
| Sentiment and rationale | Why does it recommend me (or not)? | Classification of the reasons cited: price, efficacy, taste, trust, availability |
| Cited sources | Where does it get its information? | Domains the assistant cites or links when discussing the category |
| Path to purchase | Where does it send people to buy? | Official store, retailer, official marketplace store or third parties |
The metric that surprises teams most is the last one. It’s common to find that AI recommends the brand and sends the shopper to an unofficial marketplace seller, with a different price and no warranty.
How to run your first measurement in one week
Step 1. Define purchase intents, not keywords. Write 20 to 30 questions the way your shopper would ask them, split by moment: discovery (“what can I give my kid for…?”), comparison (“X or Y?”), decision (“where can I buy X cheapest?”) and problem (“why doesn’t X agree with me?”).
Step 2. Pick your real competitors. At least three: the category leader, the fastest-growing challenger and the strongest retailer private label.
Step 3. Run the prompts under clean conditions. Use fresh accounts, with memory turned off and a new chat per question. Log the date, the assistant, whether web search was on and the full answer. Without this, results can’t be compared month over month.
Step 4. Classify the answers. For each answer, log which brands appear, in what order, with what rationale, which sources are cited and where it links.
Step 5. Calculate and compare. Compute the 6 metrics for your brand and your competitors. The comparison matters more than the absolute number.
Step 6. Turn findings into actions. Every gap has a likely cause: an incomplete listing at the retailer AI cites, too few reviews, content that doesn’t answer the question, outdated information on your own site.
What usually explains why AI prefers your competitor
In the measurements we run for consumer brands, the pattern repeats: AI doesn’t recommend by market share or by physical shelf presence. It recommends based on what it can read and verify. This is what weighs most:
- Content that answers the exact question. If your site says “advanced nutrition” and your competitor’s explains for what age, at what dose and why, AI cites the second one.
- Reviews and answered questions on marketplaces. They’re the favorite source for “users say that…” arguments.
- Consistency across your site, retailers and marketplaces. If price, pack size or claims change depending on the source, AI distrusts it or mixes up the data.
- Third-party sources. Press coverage, comparisons, forums and specialized media that talk about the brand.
- Structured data and complete listings. Product schema, filled-in attributes, images with alt text.
What to do after measuring
A measurement without a plan is a curiosity. The 90-day plan we recommend has three fronts:
- Fix what AI gets wrong. Update the sources it cites: your site, your retailer listings and the third-party pages you can influence.
- Produce the missing content. One page or listing for each intent where your brand doesn’t show up, written to answer the question, not to rank for a keyword.
- Measure again every month. The AI Shelf moves with every model version and with every move your competitors make.
In short
The AI shelf has room for one or two brands. Almost nobody measures it today, which is why it can still be won with an advantage. The first step is knowing where you stand.
At Balloon we put together a free assessment that compares your AI Shelf against three competitors and cross-references it with your Digital Shelf at your top retailers. We deliver it in less than a week and present it in a 30-minute meeting. Request it here.
FAQ
What is the AI Shelf?
It's the space your brand takes up in AI assistant answers when a shopper asks for a recommendation in your category. It works like a shelf: it has room for very few products, and someone else is already taking that space.
How is it different from SEO?
SEO measures your position in a list of links. AI Shelf measures whether AI names you, picks you, how it describes you and where it sends people to buy. They share foundations (content, authority, structured data), but they're measured differently.
How often should I measure it?
At least once a month, plus before and after every peak season (Hot Sale, Buen Fin, Black Friday, holidays). Answers change with every model version.