Which posts died, and which are worth updating

Last September, to compare this September against.

Search Console holds roughly sixteen months and then deletes a day at a time, which is enough for a year-on-year comparison right up until the month you need it. On connection we copy in every day Google still holds, about sixteen months of it, and then keep appending, once a day on either plan. A paid site keeps a rolling 24 months, about eight more than Google, and a free site a rolling 14 days, each trimmed back to that at the end of every sync, so holding a year and more of daily page and search detail, including the month Google has already dropped, is the thing paying for a site buys.

Google's own BigQuery export cannot do that half at any price: it records forward from the day you switch it on and never fills in the past, which is why it is worth switching on today and is not a substitute for having copied the past already.

An update moves the site; a page list says which posts it moved.

Pick the weeks either side of it and the Changes screen ranks what gained and what lost as pages, not only as queries, with impressions (the number of times a post appeared in someone's results), clicks, and average position, meaning where in the results you sat on average across all of them. Posts whose clicks held steady while their ranking moved are fetched as their own group so the big losers cannot crowd them out, and each one is tagged rank in the table, because those are the ones a rewrite still reaches.

The ranking column holds a line the others do not: under twenty impressions in either period it prints the rank but withholds the change, and the cell says on hover that a shift needs twenty in both periods before it counts as a shift. Otherwise a post with a handful of impressions shows a double-digit gain that is really just noise. Clicks and impressions are not filtered that way, so a small post that lost its few impressions is still in the list, with the numbers beside it.

The posts that stopped showing at all.

Page coverage reads your sitemap (robots.txt and the sitemap files it points at are the only things we ever ask your own server for) and sets what it declares against every page Google served in the range you picked, sorting them into four groups that add up to the whole: still showing, stopped showing, declared but never served in that range, and pages Google serves that your sitemap never mentions.

If the sitemap cannot be read, the page says so and says why, rather than filing your whole site under undeclared. You set what counts as stopped: three, seven, fourteen, twenty-eight, sixty or ninety days with no impressions, offered only where the threshold fits inside the range you are judging. A site crawled once a fortnight needs a longer silence before the silence means anything, and the floor is three days because Search Console itself publishes two to three days late.

Open a post to see the day it went quiet, drawn on a scale that starts at zero so a slide from healthy to nothing looks like a slide. Where the range reaches back past the per-page history we hold, those older days are marked unknown instead of drawn as zeros.

Every search term, not the first thousand.

The query list pages through the entire set rather than a top slice, and sorts and searches across all of it over whatever range you pick. The Pages tab download is the one that keeps a window of its own, a fixed last 90 days, while the query list downloads over the range you picked (CSV with no row cap, Excel to 200,000 rows with a line in the file where it stops) and it carries that window and its coverage in the file's own header.

Under the on-screen count sits a sentence measuring what share of your real clicks and impressions those queries account for, worked out against your property's own daily totals, because Google anonymizes rare searches away and a single day's query pull stops at about 5,000 rows, which is our own measurement against the API rather than a documented limit, and the days that hit that ceiling are marked as capped.

That sentence disappears the moment you filter, since the denominator is then no longer your site. A long tail that will not admit what it is missing is worse than a short one.