How the AI Shopping Check uses AI
How this tool uses AI
Most of the tool results are created using fixed rules, not AI. The results come from rules applied to:
The user’s answers
Information the tool finds on the product pages the user provides
AI affects only one answer: the product details question, and the user can answer it manually, if preferred.
What the code checks
For each product page the user shares, the tool uses code (not AI) to read the page’s structured data and tags. These checks automatically fill in two answers in the assessment. Users can edit or replace either answer manually.
What AI does
AI has two limited jobs:
Check product details. For each page submitted to the tool, AI checks whether the page states specific types of product information. It does not copy text or write content from the user’s page.
Write the report summary. AI writes three short paragraphs using the labels already in the user’s result. It does not see the user’s pages, links, answers, or email address.
What AI can see
For the product-details check, AI sees the text of one product page at a time.
Before sending page text to AI, the tool removes:
Menus, headers, and footers
Scripts
Marked-up reviews and customer questions
Email addresses and phone numbers
Links containing tracking codes
The text is used for one check. It is not stored, logged, or included in the report.
Which AI is used
The tool uses Claude Sonnet 5.5 from Anthropic through Anthropic’s API. Anthropic says it does not use API data to train its models and deletes API data within 30 days, except in limited cases described in its policy.
What AI does not do
AI does not:
Read retailer or marketplace listings
Decide whether the user’s page content is true, safe, compliant, or legally allowed
Run JavaScript on the user’s site, so information added that way may not be detected
If the AI product-details check is unavailable or fails, the user answers that question manually. Everything else in the result stays the same.
How AI is checked
Every AI update is tested against a fixed set of test cases before it goes live. Results are reviewed every week.
Questions? Email kara.lineal@gmail.com.