The entity consistency problem you're describing is brutal at ecommerce scale, where the same SKU often carries three or four different names and attribute sets across the website, the Merchant Center feed, and marketplace listings on Amazon or eBay. Retailers rarely fail at this because they don't understand the concept, they fail because no single team owns the product data across those surfaces, so the drift is structural rather than accidental. In practice, fixing it means treating the feed, not the website, as the actual source of truth for the product entity.
The entity map section is the most actionable part of this, and I'd split it in two. What you describe, mapping the primary entity, its relations and the evidence behind them, is an internal diagnostic you can build from GSC raw data today. Dixon Jones' EntityMap is a different animal.
Both are worth doing, but only the first pays off "now". As far as anyone has documented, there is no empirical evidence that major search engines (Google, Bing) or commercial LLM crawlers (OpenAI, Anthropic, Perplexity) natively ingest or give structural weight to entitymap.json or entitymap.html. While I can see initial access requests, including from Apple, in the server logs of some projects, I wouldn't call that evidence just yet.
The entity consistency problem you're describing is brutal at ecommerce scale, where the same SKU often carries three or four different names and attribute sets across the website, the Merchant Center feed, and marketplace listings on Amazon or eBay. Retailers rarely fail at this because they don't understand the concept, they fail because no single team owns the product data across those surfaces, so the drift is structural rather than accidental. In practice, fixing it means treating the feed, not the website, as the actual source of truth for the product entity.
That's a great point. I've not worked e-commerce sites properly for years but I can imagine the drift is a fucking nightmare.
And who owns it… always the same. This stuff just exists in the ether until someone's takes it on as an extra project
The entity map section is the most actionable part of this, and I'd split it in two. What you describe, mapping the primary entity, its relations and the evidence behind them, is an internal diagnostic you can build from GSC raw data today. Dixon Jones' EntityMap is a different animal.
Both are worth doing, but only the first pays off "now". As far as anyone has documented, there is no empirical evidence that major search engines (Google, Bing) or commercial LLM crawlers (OpenAI, Anthropic, Perplexity) natively ingest or give structural weight to entitymap.json or entitymap.html. While I can see initial access requests, including from Apple, in the server logs of some projects, I wouldn't call that evidence just yet.