| Read and reconcile | Every product, policy, FAQ and guide page, cross-checked claim by claim. Each conflict is logged with the page that should win. | The conflict log |
| Fix the data | Exact replacement wording for each page and product, a Shopify import file for structured product fields, and a product feed formatted for Business Hub. You publish them in about 30 minutes using our checklist. We never edit your live store. | A store whose pages agree with each other |
| Write the agent | Instructions written to OpenAI's Business Agents policy: clearly AI, answers only from your data, and a list of what it must decline and where it hands off to your team. | Agent instructions you own |
| Build the test | 100 questions tuned to your catalog, about one in five of them traps (invented discounts, medical questions, pressure, requests for personal data). | A reusable test set |
| Grade, fix, regrade | Every answer checked against a page you own and graded correct, incomplete, wrong or should-have-declined. Failures are fixed at the source, then the whole test runs again. | First-run and final-run test sheets |
| Business Hub walkthrough | Step-by-step help with the claim, DNS verification and agent submission. You do the ID step, since OpenAI requires the owner. | A claimed profile, or a ready submission |
| Handoff | What changed, what's still open, and how to keep it accurate as you add products. | A short handoff report |