We isolated the prompts where an engine had both options genuinely in reach: the vendor's own site was indexed and current, and an independent profile for the same product existed on a major review platform. Then we looked at which one the answer actually cited.
This is not a mystery. A model asked to recommend something is looking for corroboration, and a page written by the company being recommended is the weakest possible form of it. The review profile carries volume, dates, named accounts and disagreement. The homepage carries a claim 1.
What actually moves on those profiles
- Recency. A profile whose newest review is eighteen months old reads as a product nobody is currently buying.
- Specificity. Reviews that name the use case get quoted; reviews that say "great tool" do not.
- Completeness. Empty feature fields and a stale screenshot are the difference between being described and being skipped.
- Disagreement. A profile with only five-star reviews is less useful to a model than one with a visible pattern of trade-offs.
The cheapest week of work available to most vendors we audit is claiming and completing the profiles they already have. It is unglamorous, it is not a content strategy, and it moves panels.
Receipts
01 sources- 01MoatWorks source-preference sample, Q2 2026Prompts where both vendor-owned and independent profile sources were available and indexed. Placeholder pending publication of the real run.Last checked 14 May 2026

