Hairitage by Mindy McKnight. Ground truth at catalog scale.

Measured every fact AI said about every SKU.

The team compared 8,407 product fields between AI-quoted facts and the live catalog. By far the largest GT pass run on the platform. 4,160 came back exact-match, 2,482 mismatched, 1,753 fuzzy. Mismatches concentrated on price (205) and availability (53) for the primary catalog.

• 8,407ground-truth field comparisons
• 30%mismatch rate across catalog
• 38SKUs covered, every variant audited

Simulated 8 rewrites in shadow. Nothing went live until tested.

The team drafted 8 shadow PDPs (S.O.S. Deep Moisture, Outta My Hair, So Over Your Oil Control Shampoo, and 5 more) each pre-scored for AI-citation likelihood across ChatGPT, Perplexity, and Gemini. The brand sees the projected lift before approving the ship.

• 8 shadow product pages drafted, each split-tested against the live PDP
• 2,710 source recommendations generated (Ulta, Reddit, Walmart reviews)
• Every shadow page carries an EEAT score and a citation-likelihood projection

Shipped the queue: 2,595 action items, prioritised, drillable to the cell.

CONTENT_FIX and SOURCE_OUTREACH lead the queue, focused on category-cohort visibility ("sulfate-free cleansers", "frizz control coastal California"). Every item links back to the exact LLM execution that surfaced it, so the merchant team can verify before approving.

Compounded: the shadow-page primitive is now standard for any catalog over 20 SKUs.

Drafting in shadow before shipping live was a Hairitage-scale necessity that became the platform default. Every multi-SKU catalog now gets shadow rewrites with the same pre-scoring model. The merchant chooses what to push. The team already knows what will move the needle.