Resolution comes before recommendation
A model answering a buying question works from entities, not strings. If your name resolves to three different organisations, or to an ordinary word, the evidence you have published gets split across those readings and none of them accumulates enough to support a recommendation.
Publish the identifiers that tie the name down
A single canonical "about" page carrying the legal entity name, the registered address, the founding year and the business registration number does more for disambiguation than any amount of marketing copy. Mark it up as an Organization with sameAs pointing at the profiles that already exist for you elsewhere, so the separate readings get stitched into one.
Use one spelling everywhere. A brand written three ways across your own site is three entities as far as resolution is concerned.
Short and common names are a real handicap
A one or two character Chinese name, or an English word in ordinary use, will collide with unrelated text constantly. There is no markup that fixes this - the fix is to always publish the name together with a qualifier, so that the pairing rather than the bare word becomes the recognisable unit.
This also affects measurement: a short name matches inside unrelated words, so mention counts based on substring matching will be wildly overstated. Measurement should ignore names below two characters rather than report a fabricated figure.
Check what resolution currently happens
Ask each platform plainly who your brand is, with no other context, several times. If the answers describe a different company, or hedge between two, disambiguation is the blocking problem and no amount of content placement will get past it.