How do I compare my declining page with the pages that now win the SERP?
Short answer
Compare on what the reader can use, element by element, not on length. Start with format, because if the top five are tools or comparisons and you are an article, nothing else on the list matters. Then check for the specific things they have and you do not: a number, a decision, a worked example, a table, current dates.
Check format before anything else
This is the first question and it disqualifies most of the other work.
Sort the current top ten into page types: article, comparison, tool, case study, product page, documentation. Then look at where you sit. If six of the top ten are comparison pages and you are an explainer, you have not been outwritten. You are the wrong shape for what the searcher now wants.
That is a rebuild rather than a refresh, and it is a different conversation about budget. Should this be an article, comparison page, case study, tool, or service page covers picking the format.
If the format matches, carry on down the list.
Diff the elements, not the word count
Open the top five and your page side by side, and fill in a grid.
| Element | Do they have it | Do you |
|---|---|---|
| A specific number or threshold | ||
| A decision or verdict, not a survey | ||
| A worked example with real figures | ||
| A table for the comparison or thresholds | ||
| Dates on claims and data | ||
| Original data or a named source | ||
| Screenshots of the actual thing | ||
| An answer in the first 100 words |
The pattern that usually emerges is not that they wrote more. It is that they resolved something and you described something. Three or four missing rows explains most declines, and each one is a specific edit rather than a vague instruction to improve.
Word count is the trap here. Matching their length by padding makes the page worse, and Google's own guidance warns against writing to a word count because you heard there was a right one.
Fetching your live page and commenting against passages is how the diff turns into named edits rather than a general impression.
Note what they did that you cannot copy
Two rows in that grid are worth separating out, because they are not edits.
Original data. If the page above you ran a study, you cannot match it by rewriting. You either run your own or you compete on something else, which is the argument in how do I prove an article adds net-new value instead of summarizing what already exists.
A stronger domain. If everyone above you is far larger, some of the gap is not about the page. Focus on the narrower version of the query where that matters less.
Being honest about these two saves you from a refresh that was never going to work.
Compare on the AI side too
The result page is only half of it now, and the winners can be different.
Analyze AI's Sources view shows which pages assistants built their answers from for the same question, and Citation Pages shows whether yours is among them. Those two lists frequently disagree with the top ten, because selection follows whether a page adds something to the answer rather than how strong the site is.
That disagreement is useful. A page that ranks fourth and is never cited is missing something extractable, usually a clear answer near the top or a specific figure. A page that ranks tenth and is cited constantly is doing something right that the ranking has not caught up with.
Competitor tracking shows which rivals are named on the same questions, which is the AI-side version of reading the top ten.
Run the comparison as an agent
Start (manual, your URL and the target query) → Google Search returning the live top ten → Web Page Scrape on each result and on your page → Prompt LLM sorting each into a page type and filling the element grid → Sources and Citation Pages for the AI-side winners on the same question → Conditional stopping early and reporting a format mismatch when six or more of the top ten are a different type → Export Markdown as a comparison note.
Stopping early on a format mismatch is the design choice worth copying. Without it the agent dutifully produces a forty-row element diff for a page that needed to be a calculator, and someone spends a week acting on it.
FAQ
Related answers
- How do I diagnose why a page declined before rewriting it?
- What should a content-refresh brief contain?
- Should this be an article, comparison page, case study, tool, or service page?
- Why did my page lose its AI citations, and how do I win them back?
Diff your page against the ones beating it
Analyze AI reads the live top ten and your page together, and tells you when the gap is format rather than content.
Start your free trial