The good news: it’s still early. The brands that figure this out now will have a real head start. According to Shopify’s commerce data, AI-referred traffic is up 9x and orders from AI searches are up 14x since January 2025. Purchases attributed to AI-powered search have grown 11x, and those orders carry a 30% higher average order value than typical search traffic.

Less traffic. Higher intent. Bigger baskets. That’s the deal on the table.

How is AI shopping different from Google shopping?

When someone Googles “best running shoes,” they get a list of links and they click around. They do the research themselves.

When someone asks ChatGPT “I’m training for my first marathon, I have flat wide feet and I’m getting arch pain. What shoes should I buy?” the AI does the research for them. It breaks the question into multiple sub-searches (a process called query fan-out), reads 10 to 15 sources, and synthesizes a recommendation.

By the time the buyer arrives at a product page, they’ve already been pre-sold. They’re not browsing. They’re buying. That’s why average order values from AI-referred visits are significantly higher.

The catch? If you’re not showing up in those AI recommendations, you’re invisible to a whole new category of high-intent buyers. Not penalized. Just not in the room.

Where does AI actually get its product recommendations from?

This is the part that surprises most brand teams. When AI recommends a product, only about 30% of what it’s drawing from lives on your own website. The other 70% comes from off-site sources.

Your 30% (on-site content) includes:

• Product pages with structured attributes, variants, and images
• Docs and manuals with specs, installation instructions, and materials
• Blog posts, explainers, comparisons, and buyer’s guides
• FAQ and policy pages covering sizing, returns, sustainability, and warranty

The 70% (off-site) includes:

• Reddit threads and forums with real-world usage and troubleshooting
• Independent reviews from gear labs, YouTube testers, and category experts
• Media and analyst mentions in reputable news outlets and trade publications
• Standards bodies, NGOs, government sources, and academic certifications
• Marketplace Q&As, common questions, and comparative reviews

The takeaway: a strong generative engine optimization strategy has to go beyond your own website. You need an ecosystem approach.

Where does AI actually get its product recommendations from?

LLMs have a trust hierarchy baked into how they weight sources, and brand-owned content sits at the bottom. User-generated content (UGC) sits at the top.

That means reviews on your site, reviews on retailer and marketplace pages, Reddit discussions, YouTube reviews, and expert comparison articles all carry more weight with AI than your own product descriptions or brand blog.

This is a mindset shift for most marketing teams. The content you control least has the most influence over AI recommendations.

The implication isn’t to stop creating brand content. It’s to stop trying to control the entire narrative and start generating authentic content at scale. Make it easy for real customers to leave detailed, specific, verifiable reviews. That’s the content AI learns to trust.

What does AI look for in product reviews?

• Authenticity: verified buyers, no brand affiliation
• Recency: fresh reviews signal an active, in-stock product
• Specificity: detailed use cases, not just “great product!”
• Visual content: photos and videos
• Community sections: Q&As and FAQs
• Natural language: conversational phrasing that mirrors how people actually search

How does AI decide which products to recommend?

AI systems don’t flip a coin when they recommend products. They’re building something like a confidence score for each brand and product before deciding what to surface. Three signals drive that score:

1. User sentiment

Customer reviews, Reddit discussions, and social media mentions all feed the AI’s sense of how real buyers feel about your product. Positive, specific, and recent sentiment pushes your score up.

2. Citation-worthy content

When media outlets (think WSJ or Forbes), category experts (Runner’s World), or testing labs (Consumer Reports) mention your brand in a relevant context, that signals authority. Wiki pages and brand blogs count too, but third-party citations carry more weight. This is closely tied to the E-E-A-T principles that Google uses for search quality, and AI systems apply similar logic.

3. Quality structured data

Schema markup, JSON-LD structured data, your robots.txt, and increasingly, direct API connections via Model Context Protocols (MCPs) all help AI pull accurate information about your products. For the full tactical breakdown on implementing this, see our guide to actionable GEO tips for retailers.

Does generative engine optimization replace SEO?

No, and this is important to get right. GEO isn’t a replacement for SEO. Traditional SEO fundamentals still matter. Backlinks, domain authority, structured markup, quality content that addresses user intent, all of it still plays a role.

That’s because most AI models still lean heavily on the open web and search indexes. They add reasoning on top of search results, not instead of them. Strong SEO means you’re more likely to be in the pool of sources the AI is drawing from.

But here’s where it gets more nuanced. When AI platforms have access to direct APIs, they prioritize those over web crawling. Think of it as two tiers:

• Tier 1 (prioritized): Direct API access, where AI platforms partner directly with commerce systems to get real-time inventory, pricing, and product attributes without having to parse HTML
• Tier 2 (fallback): Web crawling, where AI indexes your site the same way Google does, only when no API is available

Brands connected via structured data feeds will have a growing edge over those that rely solely on crawlable web pages.

For a deeper look at how GEO and SEO work together: Read more

Why isn’t your existing content showing up in AI answers?

Most brand content today is aspirational. Beautiful lookbooks. Lifestyle photography. Inspirational room settings. Source books and catalogues.

That content is great for inspiring humans. It’s largely invisible to AI.

What generative engine optimization actually rewards is educational content. The kind that answers real, specific buyer questions. Consider a luxury furniture brand as an example. Instead of only publishing a lookbook, they’d also need articles like:

• “How to choose fabrics for a coastal home”
• “Best materials for high-traffic family homes”
• “Certified sustainable hardwoods: FSC vs PEFC ratings explained”
• “Sofa size calculator for your space”
• “Fabric performance data: Martindale rub test results by material”

This kind of content serves a dual purpose. It answers buyer questions AND it gets cited. It’s highly AI-citable precisely because it’s detailed, specific, and genuinely useful.

The aspirational content inspires and sells. The educational content gets recommended. You need both.

How do you make product pages readable by AI?

Your product detail pages need a rethink for the GEO era. The core principle: surface everything, and avoid hidden accordions.

AI can’t cite content it can’t easily parse. If your size chart is buried in a collapsed tab, or your dimensions are stored in an image instead of an HTML table, that data might as well not exist from an AI’s perspective.

This also means moving from human-optimized copy toward AI-optimized structured data. Compare these two approaches for a sofa listing:

Human-optimized:

“Expertly crafted with premium materials and timeless design, this sofa brings sophisticated comfort to your living space…”

AI-optimized:

• Category: Sectional Sofa
• Dimensions: 98″W x 40″D x 32″H
• Materials: Performance linen, hardwood
• Style: Modern traditional / coastal
• Pet-friendly: Yes (performance fabric)
• Child-friendly: Yes (durable construction)
• Lead time: 10 to 12 weeks
• Use cases: High-traffic home, beach house
• Price range: $8k to $12k (fabric dependent)

The prose is beautiful and on-brand. The structured version is machine-parsable, searchable, and recommendable. Ideally, you’d have both. A practical tip: avoid using images of tables or charts. Use actual HTML tables and semantic elements so AI can read and cite the data directly.

Step-by-step instructions for structuring product data for GEO: Read more

How do you audit your brand’s generative engine optimization across the funnel?

One of the most useful exercises you can do right now: step into your customer’s shoes and test what an AI actually says about you at each stage of the buying journey.

Top of funnel (discovery queries)

• “What are some of the best-reviewed [product] available online?”
• “What are the top [product] for [use case]?”
• “Top [category] brands with sustainable materials?”

Middle of funnel (comparison queries)

• “Which [category] brand has the best durability?”
• “[Your brand] vs [Competitor]: which is better for [use case]?”

Bottom of funnel (validation queries)

• “Is [your brand] worth it compared to others?”

Run these through ChatGPT, Perplexity, and Google AI Overviews. Is your brand mentioned? Are you described accurately? Is the information current? What are your competitors saying that you’re not?

The gaps you find are your GEO content roadmap

How do you audit your brand’s generative engine optimization across the funnel?

Here’s a simple three-step exercise:

• Feed your product data into an LLM like ChatGPT or Claude
• Ask: “How can I structure this for AI discovery?”
• Implement its suggestions

It’s a fast, low-cost way to identify gaps in how your product information is formatted. You’ll quickly see what’s missing, what’s buried, and what could be rewritten to be more citable.

For a more comprehensive audit framework, see Shopify’s GEO discoverability checklist, which breaks the whole process into actionable checkboxes.

Is it too late to start with generative engine optimization?

Not even close. GEO is still forming. Nobody has a perfect playbook yet, and that’s actually the point. The brands that experiment now, while the landscape is still taking shape, will have a real advantage when it solidifies.

The early days of SEO looked very similar. Google launched in 1998, but its guidance for webmasters didn’t follow for years. The brands that won were the ones that tested, iterated, and built authority early.

The core principles of generative engine optimization aren’t that different from what’s always mattered: be genuinely useful, be findable across many different question types, make your data clean and structured, and build a reputation that third parties want to cite.

What’s different is the channel. And the window to get ahead is right now.

Less traffic. Higher intent. Bigger baskets. That’s what’s waiting on the other side.