The evidence standard behind every RankEcho judgment
Every competitor row in RankEcho carries a verdict - alternative, not an alternative, or unclear - with the reason and the two texts it was judged from: your site's own description and the candidate's. Brands are never silently dropped; when the evidence is thin the row says unclear and stays on the board. Your override outranks the judgment and persists for future audits. Fix copy is checked against the same audit evidence before it ships: a draft that introduces a number, authority claim, or comparison the audit did not observe is replaced by a template that makes no factual claims, and the reason is recorded.
How does RankEcho decide a brand is a competitor?
Two texts, one question. RankEcho reads your site and stores what it says about itself - the homepage description, the capabilities and topic clusters it actually publishes. For each brand an AI engine named in an answer, it takes that brand's own self-description, or the sentence the engine used when naming it. Then it asks a single question: would a buyer considering your business also consider this one as an alternative for the same need?
The verdict is stored with both texts and a short reason, so the row can be checked rather than trusted. That is the difference between a competitor list and a competitor judgment.
What happens when the evidence is thin?
The row says unclear, with the reason - no description available, a snippet that describes a page rather than a business, a category too broad to call. It stays visible on the board.
This is deliberate. An earlier version of this product used a deny-list to hide brands that looked like publishers. It hid real competitors too, and it never explained itself. Abstention is the honest answer when the evidence does not support a call, and a visible abstention is a working queue rather than a silent error.
Can I overrule a judgment?
Yes, and your call wins. Click a bar on the rival panel to mark a brand as a rival or as a source. The override is stored against your account and domain, applies to future audits, and shows as your call on the panel.
You know your market better than a two-sentence snippet does. The judge is a first pass, not an authority.
What stops the fix writer from inventing claims?
Generated fix copy is linted against the evidence it was given before it reaches you. Numbers that appear in no input, authority phrasing the audit never observed, and comparisons naming brands outside the audit are flagged.
A draft that fails the check is not published. RankEcho falls back to a template that makes no factual claims, and records the reason the generated version was rejected. A fix you cannot defend in front of a client is worse than a plain one.
Why publishers are sources, not rivals
In many markets the brands AI names most often are magazines, course platforms, and review sites. They matter - they are where citations come from - but they are not who you lose a customer to.
RankEcho separates them: judged rivals sit in the competitor panel, everything else in sources and voices, each with the reason it landed there. Nothing is deleted, because a source you can see is a place you can go and get cited.
What these labels do not do
Verdicts never change your numbers. Coverage, citation rate, and share of voice are counted from what the engines actually said, with or without a judgment attached.
Labels organise the board. Measurements stay measurements.
Frequently asked questions
No. Deny-lists were tried and removed: they hid genuine competitors and never explained a decision. Every brand is judged from its own description or the sentence an engine used to name it, and the verdict carries its reason.
The evidence was too thin to call - no self-description, or a snippet describing a single page rather than a business. The brand stays visible on the board so you can judge it yourself with one click.
Yes. An override is stored against your account and domain and applies to future audits, marked as your call on the panel.
It is not published. RankEcho ships the template version, which makes no factual claims, and records why the generated draft was rejected.
No. Scores and coverage are counted from observed engine answers. Verdicts organise the competitor board; they never alter a measurement.