Edit AI product FAQs before you publish them, because a draft can sound sure about a measurement the description never stated. Product FAQs for AI Crawler keeps generated FAQs fully editable before you save. Description mode can ask clarification questions before it produces the final set. Publish only after a person has checked the claims against the product. There is no accuracy score in the product that can do that check for you.
Bulk generation does not skip the review. Selecting many products loops the same generation call once per product and returns drafts. Each draft is still yours to correct, cut, or abandon. Saving writes the group into Shopify metafields, which is what the storefront shows. A sentence you did not want becomes the sentence on the product page, and the FAQPage JSON-LD in the light DOM is built from that same saved data.
What to check on every draft
Read the draft beside the product, not in a separate window with the description closed. Compare every number with the description, the variant data you trust, or the spec sheet. Compare every material with the product, not with a similar SKU. Compare every policy sentence with the policy page: days, postage, and exceptions. If the draft added a compatible device, a care step, or a ship promise, and you cannot point to the source, delete that sentence.
Check the question as well as the answer. A question that repeats the title wastes the row. A question the inbox never asks can stay off the page. The first sentence of the answer should be the fact, with the unit and the size or condition. If the fact is buried under a slogan, move it up or delete the slogan. You are editing for a shopper who will quote one sentence, not for a brand paragraph.
- Numbers: unit, size, and a source you can point to.
- Materials: the fiber or surface of this product.
- Policies: the same window and postage as the policy page.
- Scope: product assignment for a product fact, empty assignment for a store-wide rule.
Use clarification mode when the source is thin
Description mode sends a free-text description and can ask follow-up questions before the final FAQs. Answer those questions with specifics, or stop. "It's a quality tee" is how a draft invents a fit. "The size medium chest, measured flat, is 20 inches" is how the draft gets a fact you meant. Product mode does not ask you to retype the product. It reads title, description, vendor, and type. If those fields are thin, clarification will not appear to rescue them. Enrich the description, or type the FAQ yourself.
Manual creation is the honest path when you already know the five answers and do not want a draft in the way. A group still has a name, category, tags, status, template, pagination, and the question pairs you add. Hand entry and AI creation share the monthly creation limit. The catalog does not publish the number. A review habit does not raise that limit. It does stop a bad group from being the thing that consumed it.
A reject rule for unsupported sentences
Use one rule: if the source text does not contain the claim, the claim does not get saved. That includes helpful-sounding additions, such as a waterproof rating, a shrinkage percentage, or a "true to size" line. You can rewrite the claim into something the source does support. "The description does not list a waterproof rating" is a valid FAQ if shoppers ask, and it is safer than a guessed rating.
Apply the rule to a bulk run the same way, product by product. A shared mistake, such as every draft saying the store ships worldwide, is a reason to fix the input or to move that sentence into one shop-wide group you have checked. Do not save twenty copies of an unchecked policy. Assign product facts to the products they describe. Leave the global group for rules that survived a comparison with the policy page.
After publish, and what the dashboard can tell you
Editing a saved group re-saves its fields, items, and product links. That is how a correction reaches the metafield. The admin list has status tabs, name search, category filters, and delete, including a multi-select delete that asks you to confirm. If a group should not stay on the store, delete it or change its status and confirm the storefront no longer shows it. The catalog names status as a field. It does not publish the label of a separate deactivate button, so verify on the list that the group is gone from the product before you move on.
Analytics can show which published groups people see and open. An impression is once per group per page load. An interaction is an accordion open or a Load More reveal. The dashboard includes an interaction rate, a daily graph, and a top-10 table by impressions. Use a group people open, if the answer is weak, as the next edit. The rate is not an accuracy score, and an open is not an order. A popular wrong measurement is the one to correct first, because the JSON-LD and the widget are both serving it.
- Reject any measurement the description or spec does not contain.
- Answer clarification questions with units, or cancel the question.
- Save only after the first sentence matches the product.
- Delete a saved group that should not have gone live, then reload the product page.
Questions about editing generated FAQs
Can I edit the draft before it is public?
Yes. Generated FAQs are fully editable before saving. The storefront copy is the metafield written at save time.
Does the app score accuracy?
No. There is no accuracy score. A person compares the draft with the product.
Does a bulk run review itself?
No. Bulk mode loops generation per product. Review is still per draft, before you save.
The listing is Product FAQs for AI Crawler.
Keep the next bulk run in the wizard until the sentences match the products, and read the editing flow on Product FAQs for AI Crawler. Save the groups that pass the source check, and delete the ones that do not.