You understand the challenges of agentic commerce and know that protocols overlap. That leaves the operational question: What, specifically, do we check on an existing store? Here is the list we go through during an audit, in order of priority.

1. Can agents simply access your site?

This is the first point, and it already eliminates one in three stores. Many merchants unknowingly block the very traffic they’re trying to capture.

Check your robots.txt: user agents from the major conversational engines must be allowed access to product pages. A blanket block rule inherited from an old anti-scraping measure excludes them all.
Check your application firewall and anti-bot protection. This is the most common cause: the rule may be allowed in the configuration file, but the protection service returns a challenge or a 403 error. Test from outside your network, not from within it.
Check your server logs for the past thirty days and count the visits by user agent. You’ll immediately see which ones are getting through and which ones are being blocked.
Watch out for blocking consent pop-ups: if the content isn’t rendered until consent is given, a bot will only see a blank page.

2. Is the content readable without executing JavaScript?

Some bots execute JavaScript, others do not, and those that do have limited resources. A product page where the price and stock status only appear after three asynchronous requests will, at best, be partially read.

The test: retrieve the page via the command line, without a browser, and look for the product name, price, and availability in the returned source code. If they’re not there, server-side rendering or pre-rendering is needed on the product pages.

3. Is the structured data complete and accurate?

This is the format that crawlers prioritize because it is unambiguous. The minimum required information on a product page:

Identity: name, brand, global identifier (GTIN or EAN), internal reference.
Offer: price, currency, availability, price validity date, shipping and return conditions.
Features: the key attributes for your category, expressed in explicit units.
Social proof: aggregate rating and number of reviews, if you have genuine ones.

Two common mistakes: structured data that contradicts the visible content—an agent that detects the discrepancy will downgrade the source—and availability coded as “in stock” by default across the entire catalog, which results in orders that cannot be fulfilled.

4. Are the price and inventory accurate in real time?

An agent that offers an unavailable product or one at an outdated price creates a poor customer experience for which you are responsible. Key points to address:

Maximum stock update latency: Aim for less than five minutes between the management system and what customers see.
Safety buffer: Do not publish the raw physical quantity; subtract pending orders and a safety margin.
Total price: shipping and taxes included, calculable before confirmation. This is the total the agent uses for comparison.

5. Is the catalog searchable, not just exportable?

A daily file used to be sufficient for a price comparison site. An agent asks questions such as: “Which of these products are compatible with this model?” or “Which one has the longest warranty?” This requires a search interface capable of filtering by attributes, ideally exposed via MCP—see our article on the subject.

Without a central product repository, this step is out of reach: it’s impossible to properly expose data that exists in four divergent copies across four systems.

6. Does the checkout flow accept non-human customers?

Idempotent order creation: two identical calls produce a single order. This is non-negotiable; by design, an agent will retry.
No CAPTCHA on the order path for authenticated channels.
Explicit and actionable errors: “quantity unavailable, 3 remaining” allows the agent to correct the issue; “error 500” causes it to abandon the request.
Price consistency between the offer and confirmation, with a clear policy in case of a discrepancy between the two.

7. Do you measure this channel separately?

Segment your analyses by agent and referral source. Track four metrics: agent visits, products viewed, orders attributed, and error rate encountered by agents. The latter is the most useful—it directly highlights what needs to be fixed.

The one-page checklist

Priority	Checkpoint	Common Failure
1	Authorized agent access	Anti-bot protection returning a 403 error
2	Content readable without JavaScript	Price injected after loading
3	Complete structured data	Generic availability, no identifiers
4	Up-to-date inventory and prices	Overnight synchronization only
5	Searchable catalog	File export only
6	Idempotent order creation	Duplicates created at the first sign of a network issue
7	Dedicated measurement	Agent traffic embedded in live traffic

Frequently Asked Questions

Where should I start if I only have two days?

Points 1 and 3. Unblocking agents and correcting structured data produce the bulk of the immediate impact, at a very low cost.

Do I need a file? `llms.txt` ?

It’s a useful convention for directing agents to your main resources, at virtually no cost. It’s not a ranking factor for traditional search, and it’s no substitute for structured data.

Does opening up to search engines mean opening up to competitors?

Your public prices are already being tracked by your competitors, with or without your consent. The relevant question isn’t whether to block access, but how to distinguish useful traffic from mass data scraping—which is addressed through bandwidth throttling, not a blanket block.

How long does it take to achieve full compliance?

Two to three weeks for the first six points on a standard online store. If the product catalog is in poor condition, that project will dictate the timeline, and it will take months.

Conclusion

Nothing on this list is out of the ordinary. These are the fundamentals of technical best practices and data quality—ones that agentic commerce simply makes impossible to circumvent: a human can compensate for an imperfect site, but an agent cannot.

Next up: a platform designed for this constraint, and why it all comes down to product data. Would you like this audit for your store? Contact us.

Making Your Store Readable by AI Agents: The Technical Checklist