Find page ideas in your own Search Console

Volume in a keyword tool is somebody's model of a market. Position in Search Console is a record of what happened to your pages. The first kind of number tells you where demand might be. The second tells you where you already are, which is normally the better place to start planning from.

Most sites barely read that record. The report is built for someone checking one page at a time, so the patterns across a whole property never surface, and the easiest wins in the data are the ones nobody assembles. Those patterns are what this guide is about.

Measured against modeled

The distinction matters more than it sounds. A tool estimates how often a term is searched, then estimates how hard it would be to rank, then estimates what a ranking would send you. Each step carries the previous one's error.

Search Console skips all of it. It counts impressions your pages actually received, positions they actually held, and clicks that actually arrived, for queries that were really typed. Nothing is inferred about a market. Nothing needs to be defended in a planning meeting as "an estimate".

What it cannot do is show you searches you have never appeared for. That limitation is real, and it is why keyword research still has a job. The two are complementary rather than competing: research finds territory, Search Console shows where your existing ground is being wasted.

Near-miss rankings, grouped by page

Some of your queries sit just outside the visible results. They get shown and rarely clicked, and they are the closest thing to a shortcut in this data, because the page has already been judged a plausible answer and all that remains is climbing.

To find them, pull all of the property's queries over the widest range the report offers, keep the ones whose average position falls between the bottom of page one and the middle of page three, and drop anything with too few impressions to interpret. What survives is a list of searches you nearly win.

Then do the step that makes it actionable: group them by the URL that ranks. One query is a footnote. Twelve queries attached to one URL is a single editing job with a dozen reasons to do it. This is clustering applied to first-party data, and it is the difference between a spreadsheet and a plan.

Inside each URL's group, put the highest-impression query first. Position is the thing you can move. Impressions are the thing that moves once you do.

Shown often, chosen rarely

Read the same data from the other end and look for pages that appear constantly and get picked rarely.

High impressions are proof that demand exists, because Google would not keep surfacing the page otherwise. Weak clicks next to that means something in the chain is broken, and there are two usual breaks. Either the title and description lose the comparison against everything else on the results page, or the page is being surfaced for searches it does not actually answer.

The attached queries tell you which. When they describe what the page covers, rewrite the title and the description: a short job with an outcome you can measure within weeks. When they describe something else, the page has drifted, and editing its headings will not fix a mismatch. That one needs splitting, retargeting, or a redirect to whatever genuinely covers the subject.

Both responses are cheaper than commissioning a new page, and neither needs a tool beyond this one.

Where the unnamed clicks went

Add up the clicks on a page's named queries and compare the total with the page's actual clicks. The named sum falls short, sometimes badly.

Google omits a query from the report when too few people searched it. The click still lands; the label never appears. There is no setting to change and no way to recover the individual terms.

What you can still do is useful. Page-level totals are complete, so reading the report by page rather than by query puts most of the missing clicks back into view. And the size of the gap is itself a signal worth noting: a page where the named queries explain only a small share of its traffic is a page you understand less well than you assume, and it deserves a read before anyone rewrites it.

Use the whole window

Google keeps roughly sixteen months of performance history. Most analysis looks at a quarter of that, and the habit costs more than it appears to.

Seasonal searches vanish from a short view and read as a decline. Rankings a page held before a redesign disappear from the record entirely. A page sliding off the subject it was originally aimed at looks stable over ninety days and obvious over a year. Long windows do not just add detail; they change which conclusions are available.

Any repeatable run should take the widest range the report allows, and the near-miss work above gains the most, because positions move slowly enough that a short slice mistakes a trend for a steady state.

Do it with an agent

The Doverity MCP puts Search Console next to your ranking data, so one prompt covers both. How that connection works is documented on the Search Console MCP page. Otherwise, run this.

Use the Doverity MCP. Call get_search_console_performance for
[mydomain.com] with dimensions ["query","page"], a high rowLimit, and
the full 16-month range.

A. Near misses
Keep rows averaging position 6 to 30 with impressions over [50].
Group them by URL and order each group by impressions. Give me the
ten URLs holding the largest impression pool just outside page one.

B. Shown, not chosen
Find the URLs with the most impressions and the weakest CTR. Print
the queries behind each and say whether those queries describe what
the page covers.

C. Unnamed clicks
For the ten URLs with the most clicks, subtract named-query clicks
from total clicks and report what is left.

D. Hydrate
Take the top URL from part A and run its queries through
get_keyword_metrics so each one carries volume and intent.

What comes back is a to-do list ordered by how little work separates each row from the clicks it is missing, which beats a volume column as a way of deciding what to do first.

Mining Search Console: FAQ

Can Search Console replace a keyword tool?

It cannot, because it only reports searches your site already appears for. The searches you have never reached are invisible to it, and so is anything below Google's reporting floor. Treat it as the first place to look for work worth doing and a research tool as the way to find territory you have not entered yet. Running them in that order keeps the expensive data for the decisions that need it.

What is a striking-distance keyword?

One sitting just below the fold in the results, usually somewhere in the eleven-to-thirty band of average position. The label is approximate and the reason for it is not: the page has already convinced Google it belongs in the conversation, so improving it costs less than starting a ranking from nothing. Let impressions rather than the exact band decide which rows you work on.

How far back does Search Console data go?

About sixteen months, which is the report's ceiling. Anything relying on a ninety-day view will miss seasonality, pre-redesign rankings, and gradual drift, so take the full range whenever the question concerns a page's history rather than last month.

Why do named queries not add up to my click total?

Google withholds any query searched by too few people. The clicks stay in the totals; the rows stay anonymous. Page-level figures are complete, which is why reading the report by page recovers most of what the query view hides, and why a large unexplained share on a page is worth investigating in its own right.