Strings vs things

Classic search matched strings; AI search reasons about things. When a buyer asks Perplexity about your company, the engine must resolve which entity the name refers to — then answer from facts attached to that entity. The resolution draws on: your own structured data (Organization schema with sameAs links), Wikidata and Wikipedia, business directories (Crunchbase, LinkedIn, local registries), press coverage, and the sheer consistency of how the web describes you.

The anatomy of a clean entity

  • One canonical name — used identically everywhere, including legal suffixes handled consistently. "Haipengyun Holding Group" and "Haipengyun Holdings Ltd." on different sources is a fragmentation bug.
  • Schema markup — Organization (or LocalBusiness) JSON-LD on your site with name, url, logo, foundingDate, address, and sameAs pointing to every authoritative profile.
  • Wikidata entry where eligible — the shared entity backbone engines consult. Wikipedia if notable; Wikidata has a lower notability bar and feeds the same graphs.
  • Directory and profile consistency — identical descriptions, founding year, address formats across every listing.

Entity collisions

The failure mode: your name collides with a better-documented entity, and engines answer about them when asked about you. Symptoms: wrong founding year, wrong industry, a competitor's product attributed to you. The cure is differentiation evidence — distinctive descriptors used consistently ("the Beijing-based GEO firm", not just the name) until the entity is established enough to stand alone.

Operational checklist

  1. Audit what engines currently believe: "what is [brand]" across five engines, log every claim.
  2. Deploy Organization schema with sameAs on the homepage and About page.
  3. Create/claim Wikidata, Crunchbase, LinkedIn, relevant directories — identical facts.
  4. Pick canonical descriptors and enforce them in every press release and profile.
  5. Re-audit quarterly; entities drift as sources change.

Frequently asked questions

Do I need a Wikipedia page for GEO?
Wikipedia helps but has a real notability bar — forced entries get deleted and the deletion itself becomes a negative source. Wikidata is the pragmatic move: lower bar, machine-readable, feeds the same knowledge graphs. Build press and directory evidence first; Wikipedia follows when notability is genuine.
How do I know if I have an entity collision?
Ask multiple engines 'what is [your brand]' and check the industry, location and founding facts they state. If the answers describe a different company — or hedge with 'there appear to be multiple companies' — you have a collision. The fix is consistent differentiation evidence across sources, not more volume.