The problem this solves
When an assistant answers a question in your category, it cites somebody. If that somebody is repeatedly a competitor, the temptation is to guess at why - longer content, better brand, older domain - and start rewriting on instinct. Guessing is expensive: you can spend a quarter making pages longer when the winners are actually winning on structured data, or freshness stamps, or a statistics table the engines can lift cleanly.
The winners are public. Every cited page can be fetched and dissected, and across ten questions the pattern stops being anecdote: the same handful of traits shows up again and again. A teardown replaces the instinct with a ranked list of what actually correlates with getting cited in your category, on your questions, this month. It also settles internal arguments quickly, because the evidence is the competitor page itself rather than a consultant framework: when nine of ten winners lead with a dated statistic above the fold, the editorial debate about burying numbers is over.
How the mission runs
- Collect the winning citations. For each of the ten questions, the mission samples the engines and records exactly which pages get cited. Duplicates across questions are kept, because a page cited five times is itself a finding.
- Fetch and dissect each page. Every cited page is fetched and analyzed for the traits engines reward: how fast the page answers the question, stats density, freshness signals, question-shaped headings, schema markup, and how liftable the key passage is.
- Rank the traits by recurrence. Traits are scored across all winners: what shows up on nine of ten pages is signal, what shows up once is noise. Your own competing pages get the same dissection so the gap is explicit rather than implied.
- Deliver the teardown. The report lands with a page-by-page anatomy, the ranked trait list, and the three changes to your pages that close the biggest gaps first - each tied to the evidence that earned it a place.
The prompt
This is the exact objective the agent receives. Swap the obvious placeholders for your own domain, segment or channel and run it as-is from the console, Slack, or the API.
What comes back
A teardown report covering every page the engines cited across your ten questions: what each page does, the citable traits ranked by recurrence, your own competing pages measured against exactly the same rubric, and a shortlist of the highest-leverage changes with the evidence behind each one. It is the difference between rewriting on instinct and editing against evidence the engines themselves supplied.
Make it yours
- Run it against a single bitter rival only, dissecting every page of theirs the engines cite anywhere in your category.
- Focus on one money question and go deep: every cited source for it, across every engine, with a rewrite brief for your challenger page.
- Re-run after you ship changes to verify the engines started citing the improved pages, closing the loop.
Frequently asked questions
Do the engines all reward the same traits?
Substantially but incompletely. The teardown scores traits per engine as well as overall, and when one engine diverges - favoring community sources, say, while another leans on documentation - the report says so, because that difference changes where you should invest.
What if the winners are huge publishers we cannot outrank?
Citations follow citability more than domain size, which is precisely what the teardown demonstrates in your own category. Where a giant genuinely owns a question, the report says so and points your effort at the questions where the winning pages are beatable.
How current is the picture?
Everything is sampled live at run time and timestamped. Citation patterns drift as engines update, which is exactly why a teardown beats received wisdom about what worked last year.