Mars published a report on where shopping and retail are heading, made with Kantar and Synthesis. It expects AI shopping assistants to take over more of the weekly grocery run, but not to remove in-store checkout. Neil Reynolds, Mars’s global chief customer officer, told Digiday on 11 November 2026 that the company is working on two tracks at once. One is making its brands discoverable by LLMs and making sure they convert once discovered. The other is building immersive in-store destinations with retailers.

Mars owns Dove chocolate, M&Ms and Pop-Tarts, so the worry is plain. Many snack sales happen at the checkout, when a shopper sees something and grabs it. If an agent does the shopping, who gets tempted?

The need is real, and so is the assumption underneath it

Mars is asking a fair question. A person tempted by a shelf is one kind of buyer. An agent filling a basket is another kind. It has no eyes, no mood and no queue to stand in.

The excitement around this is that brands can still win impulse sales in an agent’s world. I would test that claim before repeating it. What changed is who does the choosing. What did not change is that something still has to make a product look like the right choice.

For a person, that something was the shelf, the colour and the placement. For an agent, it is whatever the agent can read about the product. An impulse buy happens when the agent finds a product that fits a request so well that it gets added. That only works if the facts are there to be read.

Mars’s two tracks fit this. The immersive store is for the shoppers who still walk in. The LLM work is for the ones who don’t. The article gives me no numbers on what either costs or returns, so I won’t guess at them.

Mars can afford both tracks. A small brand can’t

Mars has data partnerships, new squads that put brand managers, insights staff and category managers in one team, and an agency hired for connected commerce. It is also working on how its brands appear in LLMs and where that data comes from. A giant can afford to run experiments on both fronts and wait to see which one pays off.

Agentic commerce is still too early to call. I’d put it at 50/50. Some platforms will open up to agents and others will block them to keep their hold on shoppers. The giants can wait that out or shape it. A small food or grocery brand can’t.

For smaller businesses the shift is already decided. Get data-centric. Your website is one place your product shows up. In agentic commerce, wherever your data appears, it needs more precision and more detail, and it has to describe accurately what you are.

What a small grocery brand should write down

Take a shop that sells a small-batch chilli crisp, or a bakery that sells kaya buns. A shopper tells an agent, “something spicy to bring to a potluck, no peanuts, under S$15.” The agent has to decide which product fits. It can’t taste anything. It can only compare the attributes it finds.

An agent will recommend a product when enough of its attributes are materially different from the competitors’ and the agent can read them. “Delicious, handcrafted, premium” gives it nothing to compare. These do:

  • Ingredients, including allergens, listed in full.
  • Heat level and flavour described in concrete terms.
  • Pack size and price.
  • Where it is made.
  • What it goes well with.
  • Whether it is halal, vegan or gluten free, if it is.

Numbers survive and adjectives don’t. Engines quote specific facts and drop marketing language.

People ask for the same thing in many different ways

Everyone asks their agent the same question a little differently. One person wants a gift for a coworker who likes coffee. Another wants something for a office pantry. Another wants a snack for a long flight with no nuts. This is the super long tail, and a brand that wants to catch these requests has to cover a much wider spread of questions than before.

So one product page isn’t enough. Write down the use cases, the dietary needs, the occasions and the comparisons. Answer each in its own words, with its own detail, so an agent matching a specific request finds a specific answer.

Where this connects to what I do

This is the same logic behind my GEO work. AI search doesn’t pick the biggest brand. It picks the one it understands most clearly and can quote most precisely. AeroChat follows the same practice. We publish many pages that each target one integration with a specific piece of software, and each page catches one long-tail question.

What to do

Don’t copy the giants’ spending on immersive stores. Your money is better spent on the plain work of writing your product facts down.

  1. List every attribute of each product that is true and measurable, and put it on a page an agent can read.
  2. Find what makes you different from the three nearest competitors, and state those differences as facts.
  3. Write down the different ways a customer might ask for what you sell, and give each one an answer.
  4. Date your pages and check them regularly, because AI answers change from week to week.
  5. Ask a few AI tools the questions your customers would ask. Look for whether you are named, and fix what is missing.

Source: Digiday: Mars wants to future proof impulse buys as AI is poised to take over more grocery runs