# Agent Optimization (GEO for Commerce)

How to make products discoverable and recommendable by AI shopping agents: data completeness, feeds, reviews, crawlability, llms.txt, policies, and testing.

Updated: 2026-10-04

> **In short:** Agent optimization is the practice of making products easy for AI shopping agents to find, understand, trust, and buy. It overlaps with search engine optimization but leans more on structured product data, feeds, and policies than on keywords and links. The basics are complete and accurate data, access for the right crawlers, credible reviews, clear policies, and regular testing of how agents describe your products.

## What it is

Several labels circulate for the same idea: **generative engine optimization (GEO)**, **answer engine optimization (AEO)**, and "AI search optimization". In commerce the goal is narrower than general GEO. A brand wants its products to:

1. appear in agents' candidate sets for relevant shopping requests;
2. be described accurately, with the right price, variants, and availability;
3. be recommended over alternatives when they fit the request;
4. be purchasable through whatever checkout the agent uses.

No AI platform publishes its ranking method for products, and claims of guaranteed placement should be treated with skepticism. Google's guidance for its own AI features states there are "no additional requirements" to appear in AI Overviews or AI Mode beyond normal Search eligibility, and no special schema.org markup or AI text files are needed. What follows are practices grounded in what the major surfaces document about their data sources.

## Why it matters

AI referrals are small but growing quickly and convert well. Adobe reported that AI traffic to US retail sites rose 393% year over year in Q1 2026, and that in March 2026 those visitors converted 42% better than other traffic. Shopify reported that orders from AI-powered searches on its platform grew nearly 13x year over year in Q1 2026. See [market landscape](/wiki/market-landscape).

## The building blocks

### 1. Complete, structured product data

Agents filter on attributes. If a shopper asks for a "machine-washable wool sweater under $100", products without a care-instruction attribute or with the price only in an image are hard to match. Priorities:

- Fill every required and recommended feed field, including brand, GTIN or MPN, and variant data.
- Express key specifications as discrete attributes, not only in prose.
- Write titles that name brand, product type, and distinguishing attributes, without promotional text.
- Keep descriptions factual; agents summarize them, and marketing language adds little.

See [product data](/wiki/product-data) for a full checklist.

### 2. Feeds to every relevant surface

Most AI surfaces build product results from feeds and catalogs rather than only from crawling:

| Surface | Main product data route |
|---|---|
| Google AI Mode and Gemini | Google Merchant Center (Shopping Graph) |
| ChatGPT | OpenAI product feed under ACP; commerce platform integrations |
| Microsoft Copilot | Platform and payment-provider integrations (Shopify, PayPal, Stripe) |
| Perplexity | Merchant integrations, including via PayPal; web crawling |
| Meta | Shopify Catalog and partner integrations |

Merchants on Shopify are syndicated to several of these through Agentic Storefronts by default. Others use feed tools such as Feedonomics or payment-provider programs. See [commerce platforms](/wiki/commerce-platforms).

Google's optional **conversational attributes**, introduced in 2026, let merchants add structured FAQs (`question_and_answer`), manuals and size guides (`document_link`), accessories and substitutes (`related_product`), and variant details. Google describes them as intended for conversational experiences such as AI Mode.

### 3. Crawlability for AI search bots

Agents also read product pages directly. Each operator distinguishes between crawlers:

- **OpenAI** uses OAI-SearchBot to surface sites in ChatGPT search; OpenAI says sites that block it will not appear in ChatGPT search answers, apart from navigational links. GPTBot collects training data and can be blocked separately. ChatGPT-User makes visits a user requested, and OpenAI notes robots.txt rules may not apply to it.
- **Perplexity** uses PerplexityBot for search results, which it says is not used for model training, and Perplexity-User for user-requested fetches, which generally ignores robots.txt.
- **Google** uses Googlebot for Search, including AI features; the Google-Extended token controls use of content for some other Google AI systems.

A common configuration allows search-oriented crawlers while making a separate decision about training crawlers. Check that bot management and firewall rules do not block verified AI crawlers by accident, and that key product information renders without heavy client-side scripting.

### 4. On-page structured data

Schema.org markup (`Product`, `Offer`, `AggregateRating`, `OfferShippingDetails`, `MerchantReturnPolicy`) gives crawlers and browsing agents a machine-readable version of what is on the page. Google says it is not required for AI features specifically, but it supports regular rich results and helps any agent that parses the page. Keep markup consistent with feeds and visible content.

### 5. Reviews and third-party signals

Shoppers ask agents for "the best" or "most reliable" option, and agents draw on reviews, ratings, expert coverage, and forum discussion. Practical steps:

- Collect reviews on your own site and on major retailers, and mark up aggregate ratings.
- Respond to recurring complaints in product updates and FAQs.
- Earn coverage from credible review publications in your category.

Do not create or buy fake reviews. The FTC's Rule on the Use of Consumer Reviews and Testimonials, in effect since October 2024, prohibits them. Research such as Microsoft's Magentic Marketplace study has also shown agents can be manipulated by fake reviews, which is why platforms are likely to treat review manipulation as abuse. See [consumer protection and liability](/wiki/consumer-protection-and-liability).

### 6. Policies agents can read

Shipping costs, delivery times, return windows, and warranties often decide between two similar offers. Publish them in structured form in feeds and markup, not only in a footer link, and keep them consistent across channels.

### 7. llms.txt

**llms.txt** is a proposal, first published by Jeremy Howard in September 2024, for a Markdown file at `/llms.txt` that gives language models a curated overview of a site with links to key pages. It is used mainly for developer documentation and coding tools. As of October 2026, Google says no AI-specific text files are needed for its AI features, and the major shopping surfaces document feeds and crawlers rather than llms.txt as their product data sources. It is cheap to publish, but it should not replace feeds or structured data.

### 8. Agent-ready checkout

Being recommended matters less if the agent cannot complete a purchase. Support for [ACP](/wiki/acp) or [UCP](/wiki/ucp) through your platform, agent-friendly payment options, and checkout pages that work for browser agents all reduce drop-off. Shopify's WebMCP checkout tools, released in September 2026, are one example of exposing checkout to browser agents in a structured way.

## Testing how agents see you

Because placement is not published, testing is the main feedback loop:

1. **Build a query set.** List 30 to 100 realistic shopping requests in your category, including comparisons, constraints ("under $50", "for wide feet"), and brand-plus-category queries.
2. **Run them across surfaces.** ChatGPT, Gemini and AI Mode, Perplexity, Copilot, and Alexa for Shopping, in a logged-out or neutral state where possible. Results vary by session and location, so repeat runs.
3. **Record outcomes.** Whether you appear, position, which product and variant, stated price and availability, sources cited, and errors.
4. **Fix the inputs.** Trace errors to feed data, page content, or third-party sources and correct them.
5. **Track over time.** AI visibility tools such as Profound, Scrunch, and Adobe LLM Optimizer automate parts of this; Merchant Center and platform dashboards show feed health.
6. **Connect to outcomes.** Measure AI referral traffic and conversions (see [attribution and affiliate](/wiki/attribution-and-affiliate)).

## Common mistakes

- Blocking AI search crawlers through a blanket bot rule.
- Prices or stock in feeds that differ from the live page.
- Variants listed as separate, unconnected products.
- Key details only in images or PDFs that are not linked from structured data.
- Relying on llms.txt or keyword-stuffed copy instead of fixing product data.

For an action plan by audience, see [for brands](/wiki/for-brands).

## Sources

- [AI features and your website — Google Search Central](https://developers.google.com/search/docs/appearance/ai-features)
- [How to use conversational attributes — Google Merchant Center Help](https://support.google.com/merchants/answer/17085370?hl=en)
- [Overview of OpenAI crawlers — OpenAI Developers](https://developers.openai.com/api/docs/bots)
- [Perplexity crawlers — Perplexity](https://docs.perplexity.ai/guides/bots)
- [Product feed specification — OpenAI Developers](https://developers.openai.com/commerce/specs/feed)
- [Agentic commerce on Shopify: how it works — Shopify](https://www.shopify.com/blog/how-agentic-commerce-works)
- [AI traffic to US retailers rose 393% in Q1 — TechCrunch](https://techcrunch.com/2026/04/16/ai-traffic-to-us-retailers-rose-393-in-q1-and-its-boosting-their-revenue-too/)
- [The Consumer Reviews and Testimonials Rule: questions and answers — FTC](https://www.ftc.gov/business-guidance/resources/consumer-reviews-testimonials-rule-questions-answers)
- [Magentic Marketplace — Microsoft Research](https://www.microsoft.com/en-us/research/publication/magentic-marketplace-an-open-source-environment-for-studying-agentic-markets/)
- [The /llms.txt file — llmstxt.org](https://llmstxt.org/)
- [Shopify extends WebMCP support to checkout for browser agents — Unite.AI](https://www.unite.ai/shopify-extends-webmcp-support-to-checkout-for-browser-agents/)
- [Supported structured data attributes and values — Google Merchant Center Help](https://support.google.com/merchants/answer/6386198?hl=en)