Sample report. "Fernhill Goods" is a fictional store we made up to show the format. Every page, finding and number below is invented. It isn't a client and it isn't a result. Your report follows the same structure, built from your own public pages.

Free audit, sample

Fernhill Goods: public page audit

A fictional bedding and home textiles store on Shopify, with 24 products. Below: what our bot found on its public pages, what an AI assistant reading those pages would likely get wrong, and what to fix first.

Pages read

38 public pages

Conflicts

5 found, 2 high

Unanswered questions

7 of 30 checked

Feed readiness

142 of 192 fields

Agent risk flags

4 found

Category

Home textiles, we take this

The short version

1. Two pages disagree on the return window (60 days versus 30). An AI reading both can tell a shopper either one. Fix this first.
2. The About page promises a "lifetime warranty" that the product pages limit to one year. An agent must not repeat a guarantee the store doesn't honor.
3. Two products make health claims ("helps with allergies"). An agent shouldn't repeat those, so they need rewording before any agent goes live.
4. The product data is about three-quarters ready for a Business Hub feed. Variant images and structured dimensions are the main gaps.

1. What we read

Public pages only, fetched by ArchitechturaAuditBot at one request per second, with robots.txt respected. We didn't log in, add to cart or touch checkout.

2. Conflicts between pages

Places where two of the store's own pages give different answers. For each one: the exact wording, why it matters, and the fix.

High

Return window: 60 days versus 30 days

FAQ (/pages/faq): "Returns accepted within 60 days." Refund policy (/policies/refund-policy): "30 days, unwashed, in original packaging."

Why it matters: a shopper who's told 60 days and then held to 30 has a fair complaint. A wrong return answer turns into a dispute after the sale.

Fix: confirm which is current, then make the FAQ quote the policy word for word and link to it.

High

Warranty: "lifetime" versus one year

About (/pages/about): "Everything we make comes with a lifetime warranty." Duvet insert product page: "1-year warranty against manufacturing defects."

Why it matters: OpenAI's Business Agents policy rules out invented guarantees. If the About page is wrong, an agent quoting it would make a promise the store doesn't keep.

Fix: decide the real warranty, state it once on a warranty page, and point both pages to it.

Medium

Free shipping threshold: $75 versus $100

The site-wide banner says free shipping over $75. The shipping policy says orders over $100.

Fix: one number, set in the policy, and the banner references the policy.

Medium

Queen duvet dimensions: 90 by 92 versus 88 by 92 inches

The product page and the size guide disagree.

Fix: measure, correct the wrong page, and add dimensions to the product data as structured values, not just text.

Medium

Washing temperature: cold versus warm

The care guide says "cold wash only". Three linen product pages say "machine wash warm".

Fix: confirm with the supplier, then use one care block across all linen products.

3. Questions the pages don't answer

We check a standard set of 30 pre-purchase questions for the category. These seven had no clear answer anywhere on the public site, so an AI would have to guess, or say it doesn't know.

4. Product data and feed readiness

A Business Hub product feed is only as good as the data behind it. We checked eight fields across all 24 products.

Total: 142 of 192 field checks pass, about three-quarters ready.

5. Structured data

• Product markup is present on all product pages (the theme default), but material and size values are missing.
• No return policy markup. The refund policy exists only as page text.
• No shipping details markup. Regions and delivery times exist only as text.
• No Organization markup with the business's legal name, which helps systems identify the store accurately.

6. Agent risk check

What an AI assistant reading these pages today would be likely to say, compared with what a policy-aligned agent should say. This shows why the data has to be fixed before any agent is written.

The first two need changes to the store itself: reword the health claims, and replace the countdown with a real end date or remove it. We'd recommend these whether or not you ever run an agent.

7. Category fit

Home textiles isn't in a category we decline. The two health-adjacent product descriptions above need rewording before an agent goes live. Whether OpenAI approves a Business Agent is OpenAI's decision, reviewed separately from verification.

8. Prioritized fix list

You can do every item on this list yourself; the report is yours to keep either way. A Launch Sprint does all of it, writes the agent, and proves the result with a 100-question test.

9. Method and limits

• This reflects the public pages on the day of the audit. Pages change, and so would the findings.
• We didn't check anything behind a login, checkout, order emails or the store's admin.
• We don't query ChatGPT to see whether or how often it recommends the store. Nobody can measure that reliably from outside, and it isn't what this audit is for.
• The "likely answer" column is our judgment of what a system reading these pages could say. It isn't a recording of an actual AI response.

What a Launch Sprint would do here

Everything in section 8, with the owner approving each wording change; a Business Hub feed; agent instructions written to OpenAI's policy; and the 100-question test, run, fixed and run again. $750, done 72 hours from access. What the price covers.