The 40-Point Agent-Readiness Audit for Your Catalogue and Checkout

Are we ready for AI agents is a question with no useful answer until you make it concrete. This is the concrete version: forty checks across five areas, each scored pass, partial or fail, producing a prioritised work list rather than a feeling.

Most stores score badly on their first run, and the failures cluster in the same places every time. None of the remediation is exotic — it is data hygiene, structured markup and access policy, all of which pay off in conventional search too.

Part 1: Product Data (Checks 1–12)

The foundation. An agent cannot recommend what it cannot describe or compare.

1. Every product has a distinct, descriptive title — not a SKU, not a truncated supplier string.
2. Every product has substantive description text that states what it is and who it is for.
3. Filterable attributes are above 90% populated on the attributes customers actually filter by.
4. Attribute values use a controlled vocabulary, not free text.
5. Units are explicit on every dimensional or quantitative attribute.
6. GTIN is present on products that have one.
7. MPN and brand are present where GTIN is not applicable.
8. Variants are modelled as variants, with their differences expressed as attributes rather than baked into titles.
9. Category taxonomy is consistent and each product sits in the right place in it.
10. Images have descriptive alt text that a machine can read as a description.
11. Compatibility or fitment data is structured where the category needs it.
12. The catalogue has no duplicate or near-duplicate products competing to represent the same item.

Part 2: Structured Markup (Checks 13–22)

How the data reaches a machine reading your storefront.

1. Product schema is present on every product page and validates without errors.
2. Offer schema carries price and currency and reflects the price actually shown.
3. Availability is expressed in schema and matches reality at the moment of rendering.
4. MerchantReturnPolicy is published as structured data rather than only as a prose page.
5. Shipping details are structured, including thresholds and destinations.
6. AggregateRating and Review markup are present where you have genuine reviews — and absent where you do not.
7. Breadcrumb schema reflects the real category path.
8. Organization schema identifies the retailer clearly, including contact routes.
9. Canonical URLs are correct so an agent does not encounter the same product at three addresses.
10. No structured data contradicts the visible page. This is the fastest way to be distrusted.

Part 3: Access and Crawler Policy (Checks 23–29)

You cannot be cited by something you have blocked, and you should not be scraped by everything you have not.

1. You have an explicit, documented position on each major AI crawler, rather than a default nobody chose.
2. Crawlers that cite sources are permitted unless there is a specific reason otherwise.
3. Product pages render their content without requiring JavaScript execution, or you have verified that the agents you care about execute it.
4. No aggressive bot protection blocks legitimate assistant traffic — check your logs rather than assuming.
5. Rate limiting distinguishes crawlers from attacks and does not return errors that look like outages.
6. Your sitemap is current and includes every indexable product.
7. Server responses are fast and consistent for crawler traffic, which is frequently deprioritised by mistake.

Part 4: Policies and Trust Signals (Checks 30–35)

An agent choosing between two products will favour the one whose terms it can reason about.

1. Returns policy states a window in days, explicitly, in a form a machine can extract.
2. Who pays return shipping is stated, not implied.
3. Delivery expectations are specific, per destination, rather than fast dispatch.
4. Warranty terms are stated per product category where they differ.
5. Contact routes are machine-readable and actually monitored.
6. Business identity is verifiable — registered name, address, registration number where applicable.

Part 5: Transaction Surface (Checks 36–40)

The newest area, the least settled, and the one where doing nothing yet is a defensible position.

1. You know whether your platform already exposes an agent interface, and whether it is switched on. Several major platforms enabled this by default.
2. Availability exposed to any agent channel is real-time, not cached beyond a few minutes.
3. Pricing exposed to agents matches your storefront, including promotions and customer-specific rules where relevant.
4. You have a documented position on agent-initiated fraud and promo abuse — velocity limits, agent identity, chargeback handling.
5. You have decided, deliberately, whether you want agents to be able to purchase, and the answer is recorded rather than assumed.

Check 40 is a commercial decision, not a technical one. There are legitimate reasons to say no — margin protection, channel conflict, brand control — and having consciously decided is what matters.

Reading Your Score

• Below 40: the work is in Part 1. Do not touch anything else until attribute coverage and controlled vocabularies are fixed — every other improvement is built on that.
• 40–55: you have data but not markup. Part 2 is a comparatively quick engineering project with immediate benefits in conventional search as well.
• 55–70: solid. Focus on Part 3, where a single misconfigured bot rule can be undoing everything else, and on the policy structuring in Part 4.
• Above 70: you are ahead of your category. Part 5 becomes worth planning.

How to Actually Measure Visibility

The audit tells you whether you can be used. Whether you are needs measurement, and you can do it without buying a tool:

1. Build a prompt panel — 50 queries a customer might realistically ask an assistant in your category, written down and fixed.
2. Run them monthly across the assistants that matter to you, and record whether you appear, how you are described, and who appears instead.
3. Track referral traffic from AI sources separately in analytics, and its conversion rate, which is frequently higher than non-branded organic.
4. Watch your logs for agent user-agents, which tells you what is reaching you regardless of what any dashboard says.
5. Re-run the audit quarterly. Attribute coverage decays as new products arrive.

The remediation work overlaps almost entirely with things worth doing anyway. Attribute enrichment is covered in cleaning a catalogue with AI; the visibility side is in getting your products recommended by ChatGPT and Perplexity; and the transaction surface, when you get there, is in your product needs an MCP server.

We run this audit for clients through our AI integration practice, with the data and engineering work handled by our AI and machine learning team and the SEO-adjacent side by our digital marketing team.

Get Your Score

We will run the forty checks against your store and hand you the scored list with the remediation ordered by return. Most stores are surprised by Part 1. Talk to our team; we reply within a business day.