Which ecommerce brands survive the AI shopping journey?

The brands that survived the journey were not the high-GEO brands. Casper, Allbirds, Athletic Greens, and Glossier — all heavily-SEO-optimized DTC landing pages — scored zero across all stages. Meanwhile Chewy, Hoka, and Helix dominated their categories.

Our best read of why: the high-GEO pages in this set are textbook conversion-optimized landing pages — long hero, social proof bands, FAQ accordions, expert-reviewer bylines, the works. AI shopping flows appear to skip past those in favor of mentioning the brand by name and (sometimes) linking to simpler product or category pages. Heavily-answerable pages also tend to read like SEO bait, which the platforms may be down-ranking in shopping contexts.

The honest caveats: brand recognition is a huge confound we can't measure directly — Chewy survives partly because Chewy is the obvious answer for pet supplies, not because of anything on chewy.com. Some of our high-GEO brands have drifted from their original positioning (Allbirds no longer makes serious running shoes), so the AI is correctly not recommending them — that's category-fit, not page quality. And with 40 brands, individual results have meaningful uncertainty; read the overall pattern, not any single brand's number.

Despite those caveats, three independent rounds of our research (the original citation study, the E-E-A-T follow-up, and this ecommerce study) all point at the same thing: a high GEO Authority/E-E-A-T pillar score is not a reliable predictor of AI citation or recommendation. For ecommerce specifically, it appears to anti-correlate. That's a strong enough pattern to bake into the algorithm even with the caveats.

We picked 12 product verticals where AI assistants are plausibly part of the buying journey (running shoes, memory-foam mattresses, eyewear, minimalist skincare, sustainable apparel, coffee subscriptions, meal kits, pet supplies, DTC furniture, wellness supplements, razors, bedsheets). For each, we selected 3-5 well-known brands competing in that category — 40 brands in total.

For each category, we wrote four queries representing a typical shopping journey: discovery ("what are the best running shoes"), filter ("for flat feet and overpronation"), compare ("compare the top running shoes for stability and cushioning"), and purchase intent ("recommend the top 3 with links"). Every brand in a category was tested against the same four queries.

Each brand × stage was checked across ChatGPT, Perplexity, and Claude — 475 citation checks total. We compute per-stage citation rate and aggregate across stages to a per-brand survival rate. We then look at how each of the 12 GEO pillars relates to per-brand survival to find the strongest predictors.

The Recommendation Readiness Score weights are derived from those relationships. The score uses the same pillar scores the GEO score uses — only the weighting and direction are new.

Limitations. Concept queries don't capture conversational state. Each stage is an independent API call — we approximate a multi-turn conversation, not the real thing. Survival ≠ purchase. Being recommended at the buy-now stage is the closest proxy we can build, but actual purchase outcomes require attribution data we don't have. Brand recognition dominates. A small DTC brand with a great page may lose to a well-known competitor on every stage. Survival isn't purely a function of page quality.