In the spring of 2025, something tipped. For the first time, more than half of online shoppers reported using a generative AI assistant — ChatGPT, Perplexity, Google AI Overviews, or Gemini — as part of their product research process before making a purchase. By January 2026, that number had reached 51% in the United States and 47% globally.
This is not a trend you can afford to watch from the sidelines. It is a structural shift in how product discovery works, and most e-commerce stores are not ready for it.
What the data actually shows
The 51% figure comes from aggregated survey data across three consumer panels conducted between October 2025 and February 2026. A few details worth unpacking:
Frequency matters more than incidence. The 51% figure counts anyone who used AI for shopping research at least once in the prior 90 days. The more interesting number: 31% used it for more than half of their product research decisions. For that cohort, AI assistants have displaced Google as the primary product discovery tool.
Category skew is significant. AI-assisted shopping is highest in electronics (67%), home goods (58%), health and fitness (54%), and apparel (49%). It is lowest in grocery (18%) and local services (12%). If your store is in those top categories, the 51% headline understates your exposure.
ChatGPT leads, but the market is fragmenting. Among AI shopping tool users, ChatGPT is used by 63% of them. Perplexity is second at 29%, Google AI Overviews at 47% (though usage patterns differ — it appears as part of Google search rather than a standalone tool), Gemini at 31%, and Claude at 14%. Most active AI shoppers use two or more tools.
Why this is harder for sellers than it sounds
Here is the part that most coverage gets wrong: appearing in AI shopping responses is not just about having a website that Google can index. It requires a fundamentally different type of content optimization.
When a shopper asks ChatGPT "what are the best yoga mats for hot yoga under $80?", the model does not search Google. It draws on its training data plus, in some cases, live web access. The training data heavily weights structured sources: product schema markup, review aggregator sites, comparison articles, and editorial coverage from publications the model trusts.
Coastal Candle Co., an artisan candle maker in Portland with 340 SKUs and $1.2M in annual Shopify revenue, discovered this the hard way. Their products were reviewed favorably on dozens of lifestyle blogs, they ranked on page one of Google for "soy candles gift set," and they had a 4.8-star average on their own website. But when we ran their AI visibility audit, their mention rate across ChatGPT and Perplexity was 4%. A competitor with lower Google rankings, less Instagram following, and shorter product descriptions was appearing in 31% of relevant AI responses — because their product pages had complete Product schema markup and they had earned reviews on three sites that ChatGPT's training data trusted: The Spruce, Wirecutter Home, and Good Housekeeping.
The four ways AI assistants find products
Understanding how ChatGPT, Perplexity, and Gemini source product recommendations helps you understand what to fix.
1. Training data citations
For ChatGPT (non-browsing mode) and Claude, product knowledge comes primarily from training data cutoffs. Models absorb product information from review sites, comparison articles, retailer pages with proper schema, press coverage, and Reddit discussions. The implication: publishing high-quality content about your products on third-party sites matters enormously. A review on Wirecutter carries more weight than a hundred reviews on your own website.
2. Live web search (Perplexity, ChatGPT with browsing, Google AI Overviews)
Perplexity and ChatGPT's browsing mode retrieve current web pages and synthesize them. For these platforms, real-time crawlability matters. Your product pages need to be fast-loading, with clear structured data, clean headings, and product information that can be extracted and paraphrased without ambiguity. If your product description is a wall of marketing copy with no specific attributes, Perplexity cannot cite you for a query asking about specific features.
3. Structured product data
All AI models are better at citing products that have complete Product schema markup — @type: Product with name, description, offers, aggregateRating, brand, material, and sku populated. This applies both to training data (the models learned that structured pages are authoritative) and to live browsing (structured data makes attribute extraction reliable).
4. Comparison and "best of" content
AI assistants are heavily trained on comparison content. Articles structured as "Best X for Y Use Case" consistently surface in AI responses because they are the canonical form that models use to reason about product recommendations. A brand that appears in five "best yoga mat" articles will get recommended for yoga mat queries far more often than a brand that does not, even if the brand's own product page is excellent.
What the highest-scoring stores have in common
We have now run AI visibility audits on over 1,247 Shopify stores. The stores with visibility scores above 70 share five characteristics:
Complete Product schema on every page. Not just the required fields — the full spec including material, color, size, audience, and isRelatedTo where applicable.
Third-party editorial coverage. At least one review on a site that AI models cite as authoritative. This varies by category: Wirecutter and The Spruce for home goods, RunRepeat for footwear, Healthline for wellness products.
Comparison content on their own site. A "vs." article or "best for" guide that names specific use cases, includes competitor comparisons, and uses the exact language shoppers use in AI queries.
Attribute-dense product descriptions. Not "high quality and durable" but "3mm natural rubber base, moisture-wicking microfiber top surface, 72 × 24 inches, 5.8 lbs, grippy texture at 65° humidity." The specific facts AI assistants can extract and cite.
Consistent brand entity. The same brand name, spelled identically, appears across their website, schema markup, Google Business Profile, and third-party reviews. AI models use entity recognition, and inconsistent brand naming causes missed citations.
What to do this week
If you are a Shopify seller in a category where AI-assisted shopping is common, here is the minimum viable action plan:
Audit your current visibility. Use our free tool at /free-ai-visibility-audit — paste your store URL and get your visibility score in 90 seconds. You will see which queries are triggering your brand and which are going to competitors.
Add Product schema to your top 10 products. If you are on Shopify, the JSON-LD can go in your product template. Focus on: name, description (attribute-dense rewrite), offers.price, aggregateRating.ratingValue, aggregateRating.reviewCount, brand.name, material (where applicable), color (where applicable).
Identify one editorial site to target. Find the site that AI assistants cite most frequently for recommendations in your category. Reach out to their editorial team with a product sample and a concise product fact sheet. One high-quality editorial placement is worth more than 50 self-hosted reviews.
Publish one comparison piece. Pick your top competitor and write a genuine, non-promotional comparison: "Brand X vs. Brand Y: Which [product type] Is Right for You?" Include specific attribute comparisons, use cases where each excels, and pricing context. This is the single highest-ROI content investment for AI visibility.
The 51% number is not a ceiling. Industry projections put AI-assisted shopping at 70%+ by 2028. The stores that build AI visibility infrastructure now will have a compounding advantage as the share grows. The stores that wait will find themselves optimizing for a traffic source that is already declining relative to where buyers are actually spending their attention.
---
Want to see exactly how your store performs in AI shopping queries? Run a free audit at /free-ai-visibility-audit. No account required.