How to Optimize Your PDP for AI Recommendations (Based on 5,000 Simulations)
Joe Kiernan breaks down 5,000 synthetic PDP simulations to show what makes AI agents like ChatGPT recommend your product, and what quietly hurts your odds.
Keyword-stuffed, 4,000-character text blocks actively decrease an AI agent's likelihood to recommend your product.
Including full-text, real customer reviews on your PDP is the single most impactful way to boost AI recommendations.
Hidden text works for AI, you can embed targeted Q&A for the AI to read without cluttering the visual experience for human shoppers.
JK
Joe Kiernan · Lead of Portfolio Operations and Analytics, Digital Fuel Capital
Joe Kiernan spent a decade at Bain & Company before helping Amazon aggregator Perch scale from 15 to 250 employees. He later served as SVP of AI at Razor Group before bringing his optimization expertise to Digital Fuel Capital.
The mini-framework
The AI Selection Funnel
Narrows at each stage
Influence Training Data
Produce high-quality SEO content across the web so foundational models understand your brand's overarching authority.
Ensure Live Discoverability
Implement clean JSON-LD schema markups and structured product data so AI agents can easily parse your products mid-search.
Optimize PDP Engagement
Embed full-text human reviews and targeted Q&A to convince the AI to actually recommend your product over competitors.
How this builds back to the Amazon Formula
Joe Kiernan’s team ran 5,000 synthetic simulations to find out what makes an AI agent recommend a product. The headline result surprises most sellers: a 4,000-character block of keyword-stuffed text actually lowered the odds of being recommended, and the single biggest lift came from putting the full, unedited text of real customer reviews on the page.
Joe is not guessing at this. He spent a decade at Bain, helped the Amazon aggregator Perch scale from 15 to 250 people, and ran AI at Razor Group, where his team rewrote Amazon product pages automatically every night to catch emerging long-tail demand.
Product pages, rewritten automatically, every single night.
Joe on how that system worked:
Automating PDP Updates Every Night · 1:56 clipJoe on the system that rewrote Amazon PDPs nightly at Razor Group.
Before you optimize for AI, it is worth asking whether it is driving sales in your category yet. The data splits hard. Impulse buys see almost no AI traffic today. High-consideration purchases, the ones people research before buying, are already moving to AI.
Impulse buys see almost no AI traffic. High-consideration research is already moving to AI.
Joe on where AI already drives sales:
Which Brands Get AI Traffic Today? · 1:22 clipWhere AI already drives sales, and where it barely registers.
If your category qualifies, optimize it like a funnel. Joe calls it the AI Selection Funnel, and it narrows in three stages. Influence the model’s training data with real off-site SEO, so the model understands your brand. Make yourself discoverable during live searches with clean, structured JSON-LD data. Then win the page itself, where the 5,000 simulations point in one direction: authentic context beats volume. Full-text human reviews, good and bad, and targeted Q&A do more than any block of SEO copy. You can even embed Q&A the AI reads without cluttering the page for human shoppers.
This is the Impressions lever of the Amazon Formula, specifically SEO and AEO. AI is trying to copy human judgment at scale. Give it the transparent reviews and clear answers real shoppers want, and you train the algorithm to trust you for the same reasons a person would.