Fabrication does not look like fabrication
The intuition most people carry into reviewing model output is that a made-up fact will feel made up. It does not. A fabricated statistic is generated by the same process as a real one and arrives in the same register, with the same confident phrasing and often more polish than a real citation would carry.
What does vary is the evidence around it. Real numbers tend to come attached to a source, because whoever wrote them had one in front of them. Invented numbers arrive bare, or wrapped in the phrases that stand in for a source: studies show, research indicates, industry data suggests. Those phrases are not evidence of dishonesty, they are evidence that the sentence was completed rather than reported.
Auditing for verifiability rather than for truth is the practical response. You cannot tell by reading which figure is real. You can tell, quickly and reliably, which ones you would be unable to check.
Why the record matters more than the flag
The reason this tool sits on a property research site rather than being a generic writing checker is that the useful half of the answer is jurisdictional, and most claims an advisor makes in client-facing copy resolve to a specific public record.
A claim about who owns a property resolves at the county recorder. A claim about what it is worth resolves, imperfectly and with caveats, at the assessor. A claim about who is behind an LLC resolves at the Secretary of State in the state of formation, unless it was formed in Wyoming, Delaware, New Mexico or Nevada, in which case it does not, and the route in is the recorder instead. A claim about a firm's own disclosures resolves in its Form ADV. A claim about a competitor's pricing resolves on that competitor's pricing page, read on a date you write down.
None of that requires a subscription to anything. It requires knowing which office holds the answer, which is the part a general purpose fact-checking prompt does not give you.
The claims that are worth the most scrutiny
Not every unsourced sentence deserves the same attention, and treating them equally is how a review turns into noise nobody acts on.
The ones that matter are the ones a reader could act on. A specific figure about time saved, money made, or households reached is load-bearing: somebody reads it, forms an expectation, and holds you to it. A superlative about your service is not load-bearing in the same way, because nobody was ever going to check it, and it fails for being empty rather than for being wrong.
The second category worth real scrutiny is anything with a date buried in it. Prices, thresholds, contribution limits, rankings, rate cards. A model will state last year's figure in the present tense without any signal that time has passed, and this is the failure that ages a piece of published writing into something quietly incorrect while nobody is watching.