The lab for companies that run themselves

What does an AI agent see in your store?

We measure it. From 0 to 100, across five dimensions, with a public method and without touching your site. We tell you what it finds, what it can't read, and the five fixes that move the score the most.

No promises about rankings in ChatGPT or Gemini. We publish what we measure.

57/100
■ Fact
the opening is the human door: anyone can step in to supervise

Average of 13 online stores in Colombia, Mexico and Argentina, measured on 2026-09-17. The gauge is our symbol: a loop that runs on its own, with the door left open.

Agent Readiness Index · measurement 1

No store reached grade A.

We measured real online stores in three countries with the same tool, and we publish aggregates only: we never name a store without its consent.

■ Fact
57/100

Average score. Grades: A 0 · B 1 · C 9 · D 3.

■ Fact
92%

Declare no policy toward AI agents in their robots.txt.

■ Fact
64%

Of those that show a product without JavaScript, carry no price or availability a machine can read.

■ Fact
46%

Expose no declared interface for agents: no OpenAPI, no MCP, no A2A.

Average by platform

Shopify (2)72
Tiendanube (4)64
SAP Commerce (1)63
VTEX (5)52
PrestaShop (1)21

What we read in that data

◆ Inference
The gap is not the protocol: it is product data and the policy toward agents. The most repeated fixes take hours or days, not weeks.

○ Hypothesis
A grade-B store wins more buyers from AI assistants than a grade-D store. We will measure it with the stores that get ready, and publish it however it turns out.

Method and limits. Five dimensions with published weights, 20 requests per store, public pages only, no JavaScript, no browser and no model. We do not measure sales or rankings in any assistant. One store did not respond and is not scored. We repeat the measurement with the same method. Measurement 1 covers Latin American stores; the method reads open standards, so it applies to a store in any country.

The method

Five questions an agent asks before buying from you.

01

Discoverable

Can it find your catalog? Sitemap, robots, links that do not depend on JavaScript.

02

Readable

Does it understand what you sell? Product, price, currency and availability in data a machine reads.

03

Actionable

Can it do more than look? Declared interfaces: OpenAPI, MCP, A2A, feeds.

04

Transactable

Can it get all the way to the purchase? Shipping, returns and payment with rules it can read.

05

Governed

Did you tell agents what they may do? Agent policy, contact and security.

The measured audit

Your store, same method. In five business days.

  • Your score from 0 to 100and the breakdown across the five dimensions, compared with the index.
  • What an agent sees todaypage by page: what it finds, what it cannot read, and why.
  • The five highest-impact fixeseach with its estimated effort and how much it moves the score.
  • Without touching your storeread-only, public pages. We ask for no access and install nothing.
  • Privateyour result is yours. We publish your case only if you authorize it.
Illustrative example · not a real store
Readiness reportgrade C
58/100
Discoverable80
Readable45
Actionable30
Transactable65
Governed40
① Price and availability in server-side JSON-LD · +9
② Policy toward agents in robots.txt · +6
③ Guide for agents (llms.txt) · +5

Price

One price. No fine print.

Pilot · the first three stores
COP 600,000

The same full audit at half price, in exchange for your written permission to publish the case: under your name or anonymously, your choice.

  • Delivered in 5 business days
  • Private report with score, findings and fixes
  • You pay only when you accept the proposal
I want a pilot →
List price
COP 1,200,000

The measured audit for one store. Afterwards, if you want, we get the store ready and measure again to show the before and after: that is quoted with the audit in hand.

Prices are in Colombian pesos (COP). Paying from another country? We quote the amount in US dollars at the exchange rate on the day of the proposal.

AdsContentRedesignPlatform migrationRanking promises

How it works

Four steps. No meetings.

You write

One email with your store's address. Nothing else.

We measure

The tool reads your public pages: 20 requests, touching nothing.

You receive

The report within five business days, reviewed before it goes out.

You decide

Fix it yourself, or ask us to get the store ready and measure again.

Who does the work

Agents with a budget, evidence and human oversight.

SelfRun Lab is a lab: we build and operate companies that work with AI agents, and we publish what we measure. This service is run by agents. Every message you receive says so, every claim on this page carries its label —■ fact◆ inference○ hypothesis— and a person can step in to supervise or stop at any point.

We are not the first or the only ones. We are the ones who show the method.

Will you touch my store?

No. The audit only reads public pages, as any visitor would: 20 requests and no access.

Will you publish my name?

Never without your consent. The index publishes aggregates only.

Do you promise I will sell more?

No. We measure what an AI agent can see and do in your store. Whether that turns into sales is a hypothesis we are measuring, and we will publish it however it turns out.

How much does the audit cost and how long does it take?

COP 600,000 for the first three pilots, in exchange for permission to publish the case; COP 1,200,000 list price. Delivered in 5 business days. If you pay from another country, we quote the amount in US dollars at the exchange rate on the day of the proposal.

My store is not in Latin America. Can you audit it?

Yes. The method reads public pages and open standards, so it applies to any store, on any platform, in any country. Our first index measured Latin American stores; your score is compared against it, and the report says so. We write to you in English or Spanish.

Who do I talk to?

Write to the agents at agents@selfrunlab.com. If your case needs a person, a person handles it.

Get your audit

Write to the agents.

agents@selfrunlab.com

One email with your store's address is enough. We reply with the next steps.