Methodology

How we score 100 brands on AI agent readiness

We send real AI agents to shop on every site we track — browser agents that click, data agents that parse, accessibility agents that navigate by ARIA tree. Then we score what they find across 7 weighted categories. No surveys, no self-reporting. Just what actually happens when an AI tries to buy something.

Scoring philosophy

Weighted, not averaged

Cart & Checkout is worth 25% because that's where revenue happens. Performance is 5% because fast-but-broken doesn't help. Weights reflect what matters to real agent commerce, not equal treatment.

!

Failures can't hide

A perfect product page means nothing if agents can't check out. Scoring 100 on data standards doesn't compensate for a 10 on cart flow. Catastrophic weaknesses drag the overall score down — by design.

5 agents, 10 lenses

We send 5 specialized agents to test every site. Then we score the results through 10 AI shopping agent lenses — ChatGPT Shopping, Amazon Buy For Me, Claude Computer Use and more. A site that's great for feed agents might fail browser agents.

🔍

Observable, not declared

We don't ask brands to self-report. Our agents visit the live site, try to shop, and record what happens — screenshots, DOM state, structured data, response times. The score is based on what we observe, not what brands claim.

7 scoring categories

Every site is scored across these dimensions. Weights sum to 100% and reflect the relative importance to agent-driven commerce.

15% weight

Discoverability

Can AI agents find products from the homepage? Measures navigation, search, and product listing quality.

20% weight

Product Understanding

Can AI agents read and understand product details like price, size, color, and availability?

20% weight

Navigation & Interaction

Can AI agents click buttons, use filters, select options, and navigate between pages?

25% weight

Cart & Checkout

Can AI agents add items to cart and complete a purchase — the most critical step in the shopping journey.

5% weight

Performance & Resilience

Does the site load fast enough and stay stable when AI agents interact with it?

5% weight

Data Standards & Feeds

Does the site provide machine-readable product data (like Schema.org markup and XML sitemaps) that AI agents can parse?

10% weight

Agentic Commerce

Does this site support programmatic checkout APIs (like ACP) that let AI agents buy without browser automation?

Grading scale

Scores map to letter grades. Each grade has a named readiness level that tells you what the number means in practice.

AAgent-Ready85+
BMostly Ready70–84
CNeeds Work50–69
DPoor30–49
FNot Ready0–29

10 AI shopping agent lenses

We score every site through 10 real-world AI shopping agent lenses. Each has a different weight profile based on how it actually works. Feed agents care most about structured data. Browser agents care about clickable UI. A site can score 90 for one agent and 40 for another.

Feed & API Agents

Read structured data, product feeds, and APIs. Never open a browser.

chatgpt
ChatGPT ShoppingOpenAI
Mainstream

Reads product feeds and structured data (Schema.org, JSON-LD). Uses ACP with Stripe for programmatic checkout sessions — no browser needed.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout
google
Google AI ModeGoogle
Early Mainstream

Leverages Google's existing product index, Shopping Graph, and merchant feeds. Reads structured data and product listings to generate AI-powered shopping responses.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout
perplexity
Perplexity ShoppingPerplexity
Early Adopter

Crawls product pages for structured data, compares across retailers, and presents curated results. Supports one-click checkout for Pro subscribers.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout
microsoft
Microsoft CopilotMicrosoft
Early Mainstream

Uses Bing's product index and Microsoft Shopping graph. Reads structured data from pages and merchant feeds to surface product recommendations in conversation.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout
klarna
Klarna AIKlarna
Early Mainstream

Leverages Klarna's merchant product feeds and the Klarna App Protocol (APP). Focuses heavily on pricing accuracy, availability, and comparison shopping.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout

Browser Automation Agents

Open a real browser, click buttons, fill forms, navigate visually.

chatgpt
ChatGPT OperatorOpenAI
Early Adopter

Launches a real browser session and visually navigates the site like a human. Clicks buttons, fills forms, handles popups, and attempts the full checkout flow. Requires clear UI and accessible interactive elements.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout
amazon
Amazon Buy For MeAmazon
Early Adopter

Uses browser automation to navigate external retailer sites. Fills in shipping and payment details, handles variant selection, and attempts to complete checkout on non-Amazon sites.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout
perplexity
Perplexity CometPerplexity
Experimental

Automates a browser to navigate sites, interact with UI elements, compare products, and attempt the checkout flow. Combines Perplexity's search intelligence with browser automation.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout
claude
Claude Computer UseAnthropic
Experimental

Takes screenshots and uses visual understanding to navigate. Moves cursor, clicks elements, and types into forms. Relies heavily on visual clarity, performance, and accessible UI patterns.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout
openclaw
OpenClawOpenClaw
Experimental

Open-source browser automation agent that navigates sites, interacts with product pages, and attempts the full shopping flow. Community-driven and extensible.

Discoverability
Product Understanding
Navigation & Interaction
Cart & Checkout

How a scan works

Every scan follows the same pipeline. No shortcuts, no sampling — every brand gets the full journey.

1

Land on the homepage

Browser agent loads the site with realistic headers. Data agent fetches robots.txt, sitemap, and structured data. We check if the site blocks automated visitors.

2

Find a product

Browser agent navigates from homepage to a product page — via search, category links, or menu. We record every click and measure how many steps it takes.

3

Understand the product

Can the agent read the price, size options, availability, and images? We check both DOM elements and structured data markup.

4

Add to cart

The agent tries to select options and add to cart. This is where most sites start failing — unlabeled buttons, variant selectors that require visual understanding, dynamic overlays.

5

Attempt checkout

Navigate to cart, begin checkout flow. We don't complete purchases but test whether the agent can reach each step without getting stuck.

6

Score everything

Screenshots, DOM snapshots, structured data, timing, and accessibility tree state are analyzed. Each category gets a 0–100 score. Weights are applied. Grade is assigned.

Coverage & frequency

100brands tracked
Dailyscan frequency
30dtrend history

Want your brand added? Submit it here. We review submissions daily and add qualifying e-commerce sites to the index.

Common questions

Why not just average all categories equally?

Because a site that loads in 200ms but can't complete checkout is useless to an AI agent trying to buy something. Cart & Checkout gets 25% weight because that's where transactions happen. Equal weighting would let a fast, broken site score the same as a slower site that actually works.

How is the per-agent score different from the overall score?

The overall score uses our default category weights. Each AI agent has its own weight profile based on how it actually works. ChatGPT Shopping (a feed agent) cares heavily about structured data. Amazon Buy For Me (a browser agent) cares about clickable buttons and form fields. Same site, same categories, different weights = different scores.

Do you actually visit the site, or just check metadata?

Both. The browser agent loads the site in a real headless browser and tries to shop. The data agent checks structured markup, APIs, and feeds. The accessibility agent navigates via the ARIA tree. We test what agents actually do, not just what's technically present.

Can a brand game the score?

The best way to improve your score is to make your site work better for AI agents — which means better structured data, accessible UI, and functional checkout flows. These are the same things that improve SEO, accessibility, and conversion rates. There's no trick; just build a better site.

How often do you update the methodology?

As new AI shopping agents launch and existing ones evolve, we update our agent profiles and may adjust category weights. The core categories are stable. Weight changes are announced and applied going forward — historical scores aren't retroactively changed.