Two sources of truth: your page and the live SERP

The agent fetches your URL directly and parses the real HTML: title tag, meta description, H1, up to 25 H2 headings, published and modified dates from article meta or JSON-LD, total word count, and a sample of the body text. It then reads the current top 10 for your target keyword, or one derived from the page title if you leave that field blank. Every recommendation is anchored to that evidence. Pages that time out, return an error status or serve non-HTML come back with a plain explanation rather than a fabricated plan.

The plan you get back

Updating decaying pages fails when the advice stays vague, so the output is deliberately concrete. Sections to add arrive as a heading plus the reason it is needed. Sections to update name the existing section and the specific change. Cuts are listed outright. Alongside those sit the decay signals found on the page, competitor gaps drawn from the pages currently ranking, a freshness checklist, keywords worth re-targeting, and an overall priority of high, medium or low.

• ›Sections to add, each with the reason it is needed
• ›Sections to update, each with the exact change
• ›Sections to cut
• ›A new title and meta description, both copyable
• ›High, medium or low refresh priority per page

Working through a backlog

Single mode suits one page at a time. Bulk mode accepts up to 25 URLs from a paste, a spreadsheet or a Google Sheet and returns priority, current title, and the number of additions and cuts for each, exported as one file. Sort that by priority and you have a refresh queue you can work down over a sprint instead of arguing about where to start. The plan is only as good as the page it reads, so run it against the live URL, not a staging copy.