Entity recognition is what turns a source into a name
There is a meaningful difference between being used and being named. Generative answers routinely draw facts from pages whose owners are never mentioned, and the deciding factor is usually whether the system can resolve the site to a known entity with confidence. Unresolved sources get paraphrased anonymously; resolved ones get named, and naming is what produces branded search later.
Building that resolution is unglamorous work: one canonical organization name used everywhere, organization schema with sameAs links to real profiles, a substantive About page that states what the company does in plain language, author pages with verifiable credentials, and third-party coverage that describes you in consistent terms.
The payoff shows up in an odd place - branded search volume rising while informational clicks fall. That combination is the fingerprint of successful entity work under heavy summarization, and it is worth reporting explicitly because a clicks-only dashboard will read it as failure.
Passage design for stitched answers
Because overviews stitch spans from several documents, the ideal contribution is a self-sufficient eighty-to-one-hundred-fifty word block that resolves one sub-question completely. Longer blocks get truncated mid-thought and are less likely to be selected; shorter ones lack the qualifying detail that makes a span safe to reuse.
Write each block as though it will appear next to spans from two competitors. What would make a reader trust yours? Almost always a concrete figure, a stated condition, or a named example. Add exactly one of those per block and the selection rate changes measurably.
Finally, order matters within the page. Put the most extractable, most specific blocks in the first half of the document. Sampling favours the top of a page, and a brilliant passage in the closing section competes with everything above it for attention.