PDP optimization for AI search
Turn the PDPs AI engines skip into the ones they cite.
When a product loses the answer, the sale routes to a competitor before your analytics log a visit. eCommerce Insights finds the PDPs losing those answers and proposes the exact fix — title, description, bullets, Product schema, metafields — as a diff a human approves. Never a silent rewrite.
What gets recommended
Five fields do the work. Each one ships as a diff.
AI engines synthesize a handful of sources per answer — Perplexity typically cites 3–7 — and they read data before prose. The diffs target the fields that decide whether your product is parseable, comparable, and citable.
Title
Rewritten to resolve as an entity: brand, model or series, category, primary differentiator. "Merino Base Layer — Forest" becomes a title an engine can cite without guessing what the product is.
Description & bullets
The facts engines extract — use case, material, fit, care, compatibility — move to the front. Adjective chains get deprioritized, not deleted; your voice guidelines condition every line.
Product JSON-LD
Complete, valid schema.org Product markup: name, brand, sku, gtin where it exists, offers with price and availability, material, and aggregateRating only when real review data exists.
Metafields & attributes
Shopify metafields populated against the category's attribute set — material, weight, dimensions, care, GTIN, MPN, origin — written as standard definitions so theme reads and PIM round-trips stay intact.
Image alt text
Duplicates and generics ("product photo") flagged across the catalog; proposed alts name the SKU, the variant attribute in frame, and the use context. The cheapest accessibility win on most stores, too.
Tied to the two scores
Every diff maps to the citation-score or agent-readability factor it improves, so the queue is ordered by expected impact. Methodology in the PDP Score docs and Listing Rewrite docs.
One product, end to end
A sample PDP rewrite, exactly as the diff queue shows it.
Illustrative SKU, not a real customer. Before the diff: citation score 52, cited by two of six engines, three engines returning a competitor instead.
SKU · MERINO-BL-MD · citation score 52/100 · cited 2 of 6 engines
title
− Merino Base Layer — Forest
+ Merino 200 Base Layer, Men's, Forest Green — 200 gsm Cold-Weather
description (lede)
− Our softest merino yet. You'll feel the difference from mile one.
+ A 200 gsm, 18.5-micron merino base layer for cold-weather running,
+ hiking, and travel. Flatlock seams, offset shoulder, machine-washable.
bullets
− Feels amazing on the trail
− Designed to last
+ 200 gsm, 18.5-micron merino wool, ethically sourced
+ Flatlock seams and offset shoulder reduce pack rub
+ Machine wash cool, tumble dry low
metafield · custom.material
+ 100% merino wool, 18.5 micron, 200 gsm
Product JSON-LD
+ "sku": "MERINO-BL-MD"
+ "gtin13": "0860009123456"
+ "material": "Merino wool"
+ "offers": { "price": "95.00", "priceCurrency": "USD",
+ "availability": "https://schema.org/InStock" }
image alt
− product photo
+ Merino 200 base layer in forest green, men's medium, worn on trailAfter approval, the next scans record the measured change in citation score and per-engine citations against the diff's timestamp — a delta, not a promise. Illustrative example; eCommerce Insights never publishes customer catalogs.
The approval workflow
Scan, triage, review, approve, ship. A human holds the pen.
Channel-aware output
The same engine, vendor-specific recommendations.
A Shopify PDP and an Amazon listing fail differently. The diff format adapts to where the product lives.
Metafields, variants, push
Title, description, standard metafield definitions, variant-accurate Product JSON-LD — pushed via the admin API after approval (Early access) or exported.
AMAZONRufus & COSMO scoring
Title, bullets, A+ Content, and backend keyword recommendations scored the way Rufus reads listings, across COSMO's commonsense intent relations. CSV for Seller Central.
OTHER CHANNELSAdobe, BigCommerce, Woo, headless
Schema and content diffs mapped to each platform's field model — Adobe Commerce, BigCommerce, WooCommerce, headless — exported as CSV.
Software you drive, or a service that drives for you.
Enterprise done-for-you suites run catalog optimization as a managed service — writing, uploading, and monitoring content across the marketplaces and AI engines on an engagement model. If you want a partner with a procurement cycle to handle the work end to end, that shape may serve you well.
eCommerce Insights is the other shape: self-serve software with public pricing, built for the brand whose catalog lives in Shopify admin and whose team is four people. Your merchandiser reviews the diffs; you keep the editorial pen. The managed-service head-to-head is eCommerce Insights vs Gushwork; the full field sits in the compare section.
Audits without fixes are noise. The diff is the deliverable.
Questions buyers ask
Does it rewrite my pages automatically?
No. eCommerce Insights produces reviewable diffs — proposed before/after text for each field it flags — and a human approves every line before anything changes. The Shopify push (Early access, Growth plan and up) only writes approved changes, each one versioned and reversible. Recommendation-only mode and CSV export are on every plan.
Will the recommendations match my brand voice?
You set the guardrails: voice notes, banned words, capitalization rules, tone examples. Every diff is conditioned against them, and anything that cannot be recommended without breaking the rules gets flagged instead of suggested. The diff always shows your original next to the proposal, so you can edit before approving.
What if I'm not on Shopify?
Tracking, scoring, and recommendations work on Amazon, Adobe Commerce, BigCommerce, WooCommerce, and headless stores, with channel-specific output — Amazon SKUs get title, bullet, A+ Content, and backend keyword recommendations scored against Rufus and COSMO. Write-back is Shopify-only for now; every other channel exports the approved set as CSV.
How do I measure whether a fix worked?
Every approved diff is timestamped, and the next scans record the SKU's citation score, agent-readability score, and per-engine citations against that timestamp. You see the before/after delta per fix over the following refreshes — shown as a measured change, never a projected promise.
Who approves changes before they go live?
Whoever you designate — typically a merchandiser or the ecommerce lead. Every approval is logged and attributed, every push is versioned, and any line can be reverted field-by-field later. Nothing reaches the store without an explicit human approval; there is no silent auto-write mode.
Will optimizing for AI engines hurt my Google rankings?
The recommended changes — specific entity-resolving titles, populated attributes, complete Product JSON-LD, surfaced review signal — are the same disciplines Google's own product structured-data guidelines reward. Existing keyword targeting is preserved; the diffs add machine-readable specificity rather than trading one channel for another.
See it on your own catalog
The first five diffs are free.
Run the grader on five PDPs, no signup. Or start the 14-day trial — no credit card — and see the full queue. The tracking layer behind the queue: SKU-level tracking.