Comparing your pages to a competitor's tells you almost nothing
Comparing your pages against a competitor's cannot tell you what to change, because the gap includes domain authority, age and backlinks you cannot copy. RankEcho compares a competitor's cited pages against that same competitor's uncited pages instead. Same site, same authority, same brand - so whatever separates the two sets is structure you can act on.
Why doesn't comparing your pages to a competitor's work?
The standard advice is to put the two side by side. Count the data points on their page and then on yours. Run a structured-data test on both sites. Compare the opening paragraphs. Count how many service pages they have and how many you have.
Every one of those comparisons is confounded. When a rival's page is cited and yours is not, the difference between them includes domain authority, how long the site has existed, how many other sites link to it, and how often the brand is named across the web. None of that is something you can copy by editing a page. So the comparison produces a list of differences without telling you which one mattered.
The confusion is easy to fall into because citation and ranking look like the same problem. They are not. Moz's analysis of forty thousand AI Mode queries found that 88 percent of citations came from pages outside the organic top ten - so the page winning the citation is frequently not the page winning the ranking, and authority alone does not explain it either.
Aggregate studies have the same weakness at a larger scale. AirOps reported that 61 percent of ChatGPT-cited pages carry three or more schema types, and that pages doing so were roughly 13 percent more likely to be cited. That tells you what cited pages look like across the whole web. It cannot tell you whether schema is what separates this particular rival's cited pages from their ignored ones - and that is the question you actually need answered before you change anything.
What does comparing their cited pages against their uncited pages show?
One site, split in two. Take the URLs on a rival's domain that AI engines cited when answering your prompts. Then take a sample of the rest of their site - pages the same engines had access to and did not cite. Measure both sets the same way.
Everything that made the first comparison useless is now held constant. Both sets sit on the same domain, carry the same authority, belong to the same brand, and address the same audience. Whatever still differs between them is treatment: how the page opens, how long its paragraphs run, whether figures sit in a table, whether claims link out, whether headings ask the questions buyers ask.
That difference is the finding. It reads as a sentence a writer can act on - their cited pages open with a self-contained answer in the first two paragraphs, and their ignored pages do not - rather than a list of attributes you cannot change.
RankEcho runs this comparison for up to eight rivals you name, and reports a trait only when the gap between the two sets is at least twenty-five points. Below three pages on either side it says the sample is thin rather than printing a percentage that reads like proof.
How do you know which of a competitor's pages AI engines actually cite?
You ask the engines. There is no way to derive it from their site, from a backlink index, or from search rankings - the only source of truth is the answer an engine gives to a specific question, and the URLs it attached to that answer.
This is the half of the method a crawler cannot supply. RankEcho records, for every prompt in an audit and every engine that ran it, the exact URLs each engine cited. When one of those URLs sits on a rival's domain, it becomes a known cited page for that rival.
It also means the comparison improves as your prompt coverage grows. A rival cited on two of your prompts gives a thin cited set; a rival cited on fifteen gives a comparison worth acting on.
Which traits separate a cited page from an ignored one?
Six are measured, chosen because each is visible in the page itself and each is something a writer can change in an afternoon.
An answer in the opening: a short, self-contained statement in the first paragraphs that could be quoted with nothing around it. Structured data: whether the page carries JSON-LD at all. A table: whether figures are presented as rows rather than buried in a sentence. Question headings: whether sections are phrased the way a buyer would ask. Outbound sources: whether claims link to where they came from. Quotable chunks: whether any paragraph runs long enough that an engine would have to cut it mid-thought.
Which of the six matters is not decided in advance, and that is the point. On one rival's site the separator might be tables; on another it might be nothing but paragraph length. A trait that appears on two or more rivals' sites is reported separately from one that appears on a single site, because one site's habit is a habit and three sites' habit is how the category gets cited.
What if a competitor has no cited pages yet?
Then this comparison is impossible, and RankEcho says so rather than producing something that looks like it. A rival you want to beat is not the same set as the sites engines currently cite for your prompts - the overlap is often partial, and for a rival you have just added it may be empty.
What runs instead is a site-to-site comparison measured the same way on both sides: their pages against yours, on the same six traits. It answers a different question - not what makes their cited pages citable, but where their pages are built differently from yours. That is weaker evidence and it is labelled as such.
Your own site is profiled by the identical crawler for that comparison, so neither side gets a more generous reading than the other.
How is this different from an AI visibility tracker?
A tracker answers who is cited. Several do it well, including at URL level, and some now distinguish sources a model consumed from sources it linked. That is the right first question and it is largely a solved one.
This answers a different question: what to change. Knowing a rival holds a citation you want tells you the score. Knowing that their cited pages open with an answer while their ignored ones bury it tells you what to write on Monday.
The two are complements rather than alternatives. RankEcho does the tracking as well - that is where the cited URLs come from - but the comparison is what turns an observation into an instruction, and it is only possible because both halves live in the same place.
What this method cannot tell you
It cannot tell you why a page was cited. It reports what differs between two sets of pages on one site; the causal story is yours to judge. A trait can separate the two sets for reasons that have nothing to do with citation.
It cannot read a site that refuses to be read. Robots directives are obeyed, and some publishers and preprint servers decline automated requests entirely - when that happens the report names the site and says it was blocked rather than implying nothing was there.
And it cannot promise movement. Matching what a rival's cited pages do is a better-founded bet than guessing, and it remains a bet. The only honest test is re-asking the prompt after you publish and recording what changed.
Frequently asked questions
Up to eight in RankEcho, and that ceiling is deliberate. Past eight you are describing an industry rather than the rivals you actually lose to, and each site added costs a crawl that produces a thinner comparison than a shorter, sharper list.
No. Every input is public: pages their own robots directives permit reading, and the URLs AI engines exposed when citing them in an answer. Nothing is inferred from traffic data, and nothing is accessed that a site has asked crawlers to leave alone.
A benchmark tells you what cited pages look like on average across the web, which is useful context and useless as an instruction. Within one site, authority and brand strength are constant, so a difference between its cited and uncited pages points at treatment rather than at reputation.
When your prompt coverage grows, when a rival visibly reworks their site, or when a citation you held moves. The cited set changes as engines re-answer prompts, so a comparison run months ago describes a page inventory that may no longer exist.
For the cited-versus-uncited comparison, yes - it reads the rival's site, not yours. The fallback site-to-site comparison needs enough of your own pages to be meaningful, and reports a thin sample rather than a percentage when there are fewer than three on either side.