Agentic Commerce Optimization: How to Get Picked by AI Shopping Agents

More than half of US shoppers now use AI somewhere in their shopping. NielsenIQ's September tracker put it at 51% for the past month, up from 42% in May. Meta's Muse agent can now check out at Shopify stores with Shop Pay. The shopper asks one question and gets two or three answers.

If your product is not one of them, the shopper never sees your listing.

A new name has started to stick for the work of being one of those answers: agentic commerce optimization, or ACO. This guide covers what it is, how it differs from SEO and GEO, and a weekly routine a small team can actually run.

The short version

	SEO	GEO	ACO
Where the buyer asks	A search box	An AI assistant	An AI shopping agent that can also buy
What wins	A ranked page	A mention or citation	Being one of the few products the agent recommends
What the engine reads	Pages, links	Pages, sources, entities	Structured product data: attributes, price, stock, reviews, policies
What you measure	Rank, clicks	Mentions, share of answer	Share of shelf per question, and orders from agent channels

ACO does not replace SEO or GEO. It adds one layer on top, where the agent is choosing a product, not a page.

What a shopping agent actually looks at

Ask an agent for "a carry-on that works for two kids on a long flight" and watch what it does. It does not skim your homepage. It compares a handful of products on facts it can read and repeat:

Specific attributes. Weight, dimensions, materials, capacity. "Lightweight" loses to "2.4 kg".
Price and availability. A current price that matches across channels, and stock it can trust.
Proof. Review count and rating, and claims that point to something.
Policies. Warranty term, returns window, shipping promise.
Fit to the question. Whether the listing says who it is for, in words close to how buyers ask.

The reason an agent gives for its pick is the most useful signal you will get. "Picked because it lists interior dimensions and a 10-year warranty" tells you exactly which fields to fix.

A weekly ACO routine

1. Write down the questions worth winning

Start with 10 to 20 buyer questions, in the buyer's own words. Pull them from reviews, support tickets, and your search terms. Mark each one high, medium or low priority, and note which SKUs should be the answer.

2. Check the shelf

Ask each question in the agents your buyers use: ChatGPT, Gemini, Meta Muse, Amazon Rufus. Use a fresh session and no brand hint. Record where you appear (or that you do not), who was picked instead, and the reason the agent gave. Keep an excerpt as evidence.

Answers vary from run to run, so treat one check as a sample. The trend over several weeks is what matters.

3. Find the gap in the data, not the copy

For each question where you lose, put your listing next to the winners and name the concrete difference. Most gaps are one of five things: a missing attribute, a claim with no source, a price that differs by channel, stock status, or missing policy information.

4. Propose one fix per field, with a source

"Add packed weight: 3.1 kg (supplier spec sheet, rev C, page 2)" is a fix someone can check. "Improve the description" is not. Never copy a competitor's number because it helped them rank. If you cannot source a value, it waits.

5. Let a person approve what ships

Agents repeat your listing as fact to a buyer who may pay on the spot. Price, warranty, certification and compatibility wording should always wait for a person. Copy tweaks can move faster.

6. Re-check

Next week's check shows whether the fix moved you. Report movement per question and agent. One run is not proof of cause.

What goes wrong

Chasing the answer text instead of the data. Rewriting copy to sound like the winner rarely works. Agents pick on facts.
Inventing specs to fill gaps. A made-up attribute is worse than a missing one. The agent will quote it, and the customer will hold you to it.
Different prices on different channels. An agent that finds your product at $179 on one checkout and $189 on another may pick neither, or buy at the wrong one.
Measuring once. A single screenshot of an AI answer is an anecdote.

Running ACO with an agent team

The routine above is mostly repetitive work: asking the same questions every week, comparing listings, drafting field-level fixes. That is agent work. Deciding which facts are true and which prices are acceptable is not.

In Buda, an AI agent team for ecommerce works from one store workspace. Your product data, supplier sheets and policies live next to it. A research agent tracks competitor listings. A catalog enrichment agent fills attribute gaps and flags listings with stale prices or conflicting stock across channels. A listing agent drafts channel-ready copy. Each output comes back as a draft for a person to approve before it reaches the storefront.

The same workspace can hold the weekly routine as a reusable skill and run it on a schedule. The SEO agent that already maps keyword clusters can keep your search and AI-answer work in one place.

Start small: pick five buyer questions and one product line, run the check for three weeks, and see what the reasons tell you.

FAQ

Is agentic commerce optimization the same as GEO?

No. GEO is about being mentioned or cited in AI answers. ACO is about being the product an AI shopping agent recommends, and increasingly the one it buys. It depends much more on structured product data, price and stock than on articles.

Which AI shopping agents should I check?

The ones your buyers use. For most US sellers that means ChatGPT, Gemini, Meta Muse and Amazon Rufus. Walmart and others run their own shopping assistants inside their marketplaces.

Can I automate the checks?

Check each platform's terms first. Many do not allow scraping. A safer pattern is an agent working in a browser session you control, at human pace, with every answer saved as evidence.

How long before a fix shows up?

It varies by agent and by how often it refreshes product data. Track it weekly and look at the trend over a month.