What does an AI system see when it reads a real store? We ran ecommerce.com on a well-known public store to find out, and this post reports what one run found.
The facts of the test:
- Store: allbirds.com, a Shopify store.
- Date: 24 September 2026, one run.
- Data: public data only: the product feed, the product pages, the sitemap, Google results (United States, English, desktop) and the latest public Trustpilot reviews.
- Relationship: Allbirds is not a customer of ecommerce.com. We picked the store because its catalog is public and well kept, which makes the gaps more useful to study.
One run is one sample on one date. Google results change from day to day, so read these numbers as a snapshot, not as a ranking of the brand.
Product data: 145 of 293 products had a problem
The run read all 293 products of the store's feed and every product page. Each product got two scores: an SEO score (17 on-page checks plus its place on Google) and a GEO score (does the page state the facts that shoppers ask about in its category, and does Google's AI Overview use it).
145 of the 293 products had at least one data problem. 30 of 293 scored under 50 for AI answers.
Google: in the first ten for 3 of 24 buyer searches
For its products, the run wrote buyer searches the way a shopper types them, with no brand name (for example "women's running shoes"), and read the real Google result pages.
- The store was in Google's first ten results for 3 of 24 buyer searches.
- The store's own pages were on 4 of 24 buyer result pages at all.
- amazon.com was on 21 of 24 of the same result pages.
On the search "women's running shoes", Google showed an AI Overview. It named two products from other brands and did not name allbirds.com.
What this means for a store
A good brand is not enough for a generic search. The shopper who does not type the brand name meets marketplaces and competitors first, and the AI summary above the results names the products whose pages state the facts it compares.
Pages: 801 of 1,845 sitemap pages were dead
The run listed every URL in the store's sitemap: 1,845 in total. 801 of them were dead: pages that return no useful content, such as empty collection pages. Search engines keep crawling them, which spends crawl time on pages that cannot rank or answer anything.
| Finding | Result | Source in the run |
|---|---|---|
| Products with a data problem | 145 of 293 | Product feed and product pages |
| Buyer searches with the store in the first ten | 3 of 24 | Google, US, desktop |
| Buyer result pages that show the store | 4 of 24 | Google, US, desktop |
| Buyer result pages that show amazon.com | 21 of 24 | Google, US, desktop |
| Dead sitemap pages | 801 of 1,845 | Sitemap and page reads |
| Latest Trustpilot reviews with 1 or 2 stars | 102 of 200 | Trustpilot |
| Of those 200 reviews, replied to by the store | 0 of 200 | Trustpilot |
How the run reads a store
Every number in this post comes from the same method that runs on any store.
Two scores for each product
- SEO score: 17 weighted on-page checks (60% of the score) plus the product's place on real Google result pages (40%).
- GEO score: content readiness (70%), such as whether the page states the facts of its category in the HTML, plus how Google's AI Overview treats the page when the buyer search shows one (30%).
The facts of each category come from the open Shopify product taxonomy, which lists standard attributes and return reasons for each category. No study publishes the right weights, so the weights are our judgment, and the dashboard shows every check behind a score.
Real Google searches
The buyer searches are written by a model from the product data, with no brand name, and the dashboard labels them as model-written. Each search is one real Google result page, United States, English, desktop, read to a depth of 20 results.
What the numbers do not say
- They do not say that the store sells less than it could. The run reads public data only; sales, traffic and ads stay private.
- They do not rank the brand against its competitors. One run on one date is one sample.
- They do not say every flagged product needs work now. The dashboard orders the fixes by the points they add, so a team starts with the few that matter most.
Reviews: 102 of 200 gave 1 or 2 stars, 0 got a reply
The run read the latest 200 public Trustpilot reviews of the store, sentence by sentence. 102 of the 200 gave 1 or 2 stars, and the store replied to 0 of the 200. The complaints were mostly about service, returns and delivery.
AI answers read public reviews too. A review site with many unanswered complaints is part of what an agent learns about a store before it recommends it.
What to fix first
If your store looks like this one, the order of work is simple:
- Add the missing category facts to the product pages that sell most, in plain text.
- Remove or redirect dead sitemap pages, so crawlers spend their time on pages that sell.
- Find the buyer searches where marketplaces stand in front of you, and give those product pages the facts the AI summary compares.
- Reply to recent complaints on public review sites.
For how to write the product facts, see How to write a product page that AI answers can quote. For why this matters now, see Agentic commerce: what changes when AI shops for your customers.
Run it on your own store
Every number in this post came from one ecommerce.com run. Enter your store address to see the same dashboard for your store: each product with its SEO and GEO score, your Google results, your dead pages and your reviews, with the fixes that add the most on top.
Questions and answers
Was Allbirds a customer?
No. We ran ecommerce.com on the public store allbirds.com on 24 September 2026 with public data only. Allbirds is not a customer.
Are these numbers a ranking of the brand?
No. They come from one run on one date. Google results change daily, so each result page is one sample.
What counts as a dead sitemap page?
A URL in the sitemap that returns no useful content, for example an empty collection page. The run found 801 of 1,845.
