Essay · September 2026
Most of your AI-search visibility is off-domain.
Teams respond to AI Overviews, ChatGPT search and Perplexity by writing more pages. Most of what decides whether an assistant names your brand is not on your website at all.
The entity, not the page
An assistant answering "which cloud testing platform should I use?" is not ranking pages; it is resolving entities and deciding which it trusts enough to name. That trust is assembled from everywhere the entity appears: review sites, package registries, developer communities, marketplaces, news, the knowledge graph. Most of that footprint sits off your domain, which makes the website the smaller lever.
What that changes
- Entity consistency beats one more blog post. One name, one description, one set of links, everywhere. A rebrand is the stress test: the old and new names must resolve to one entity, which is why we kept the old name in titles and in Organization schema as
alternateName. - The 40–60-word answer capsule is the new meta description. The passage that gets cited is the one that can stand alone: a direct answer, the specific term, one number, no throat-clearing.
- llms.txt is a courtesy, not a lever. Adoption by assistants is small and unverified. Ship it because it costs nothing; do not build a strategy on it.
- Measure mentions, not rankings. Rank-tracking tells you about ten blue links. For assistants, track whether you are named, alongside whom, and from which source the assistant took the claim.
A practical order of work
- Audit every profile the entity has that you did not create, and claim it.
- Fix the schema: Organization with
sameAsto every profile, product schema on product pages. - Rewrite the top hundred pages' opening passage as an answer capsule.
- Only then, new content — and only where the assistant currently cites a competitor.
Drawn from the off-domain work behind the rebrand. No company data appears here.