Two sets of constraints, both binding
The legal constraints are specific to the industry and the jurisdiction: what may be claimed about treatment outcomes, what constitutes legal advice, what returns may be described. The second set is behavioural: models answer these categories more conservatively, hedge more, and prefer institutional sources over commercial ones. A strategy that ignores either will not work.
What cannot be claimed
Efficacy or cure rates in medical content without the approvals that permit them. Outcome guarantees in legal content. Return figures in financial content presented without the conditions and risks. These are not stylistic preferences - publishing them creates liability, and a model that detects a claim of this shape in a commercial source tends to discount the whole page.
What still works
Qualifications, stated with the issuing body and number. Process descriptions: what happens at each stage, what it costs, how long it takes. Scope: exactly which matters you handle and which you refer out. Published limits. None of these require a claim about outcomes, and all of them are quotable because almost nobody publishes them plainly.
Who the author is matters here
In these categories an identified author with verifiable credentials changes how a page is weighted. An anonymous page making careful claims is still an anonymous page. Put a real name, a real qualification and a way to check it on anything substantive.
Expect to be cited less, and plan for it
Even done well, commercial sources in these categories are cited less often than institutional ones. The realistic goal is to be the commercial source that is cited when one is, not to outrank a health authority. Setting the expectation early prevents the recheck from being read as failure.