SI Store Ops

Troubleshooting guide · updated 2026-10-12

Six ways bulk AI product copy goes wrong, and how to find each before you import

Wrong-product text, cut-off sentences, made-up specs, overwritten fields, broken characters and feed drift, with a check for each pattern.

The six patterns

Check for each pattern separately, because a random sample tends to miss them. The checks below are authored, and most can be done in a spreadsheet.

  • Different item: compare the product type and colour in the text with the title and the source row.
  • Cut-off sentences: filter for descriptions that do not end in a full stop.
  • Made-up specs: look up every number in the copy in the source sheet.
  • Overwritten fields: compare the before and after export, column by column.
  • Broken characters: search for odd symbols and leftover tags.
  • Drift: compare page text with the feed and with the structured data for changed products.

Why a sample is not enough

One poster in the thread says random spot-checking tends to hide the failures, because the bad ones cluster by pattern. Group by supplier, template or product type and check each group. Keep the pre-update export as your rollback.

Search and shopping rules for AI text

Google asks merchants to disclose text created with generative AI in product data, with separate attributes for titles and descriptions. Its spam policy names generating many pages with AI tools without adding value for users as scaled content abuse. Neither rule is a reason to avoid checking; both are reasons to add real, checked facts.

Which paid job fits

The copy check traces up to 150 products to your source data and plants known defects to prove the audit finds them. It does not import anything or approve claims.

Sources and limits

  • Shopify Community: checking AI-generated content after a bulk update Checked 2026-10-12.
    • A thread of 25 to 28 Sep 2026 names these failure modes after bulk AI updates: copy describing a different item than the title, sentences cut off by length limits, fabricated specs or claims, fields changed that were meant to stay untouched, garbled characters or leftover markup, page text drifting from the Merchant Center feed, and structured data keeping old copy.
    • A poster says random spot-checking tends to hide the failures because the bad ones cluster by pattern, and to keep the pre-update export as your rollback.
  • Google Merchant Center Help: disclose content created with generative AI Checked 2026-10-12.
    • Google asks merchants to disclose text created with generative AI, using the structured_title and structured_description attributes.
  • Google Search Central: spam policies Checked 2026-10-12.
    • Google names using generative AI tools to generate many pages without adding value for users as scaled content abuse.