When an AI answer names two competitors and skips your brand, the cause is rarely a single broken tag. Retrieval-augmented models build an answer from pages they can fetch, passages they can quote, and entities they can verify. If any one of those three links is missing, the model defaults to a brand it has already seen cited elsewhere.
The first link is access. Retrieval crawlers and training crawlers are separate agents with separate rules. A robots.txt that blocks every bot with "User-agent: *" removes your pages from the retrieval pool while leaving competitors untouched. Check the declared rules for each crawler, and confirm that the blocked path is not the one holding your product or service pages.
The second link is quotability. Models prefer pages that state a conclusion in the first two sentences and then support it. Pages that open with a brand slogan, a carousel, or a cookie banner give the extractor nothing to lift. Rewrite the opening of each key page so the answer exists as plain text before any interaction is required.
The third link is entity evidence. A model recommends a brand it can describe: what it sells, where it operates, who stands behind it, and which third parties confirm it. Organization and Product structured data help, but they only restate what the visible page already says. A policy page, a verifiable address, and one or two independent mentions do more than any schema field.
The practical order is access first, structure second, entity third. Fixing structured data while retrieval crawlers are blocked changes nothing. Start by fetching robots.txt and the key pages as a crawler would, then compare the text a model can extract with the text a buyer sees.