60% of my keyword list wasn't real. Here's the filter that fixed it.
I exported 4,122 queries from Search Console across five sites and nearly built a content plan around all of them.
Then I looked at the impression counts.
The noise
On two of the sites, 57–60% of queries had two impressions or fewer over three months. Hundreds of them showed exactly one.
Those aren't keywords. Nobody is searching for them in any meaningful volume — they're artifacts of Google showing a page once or twice, or long-tail phrasings that appeared a single time.
Treating them as keywords is worse than useless: it makes you write content for demand that doesn't exist.
The filter I use now
Three or more impressions before a query counts. Two or fewer, it goes in a separate bucket labelled "noise" and gets ignored for planning purposes.
That one number cut my working list by more than half, and what remained was actually worth writing for.
The second filter
Of what's left, I only act on queries ranking 21 to 60.
- Below 21: already competitive, needs links and authority, not another article
- Above 60: too far to close with on-page work alone
- 21–60: close enough that internal links and page improvements can actually move it
Combined, those two filters turned 4,122 queries into a few hundred actionable ones.
Why the raw export misleads you
Search Console is a log of what happened, not a list of what people want. It will faithfully record every single-impression phrase your pages matched, and present it to you in the same table as a term with 2,925 impressions. The spreadsheet doesn't rank them by importance — you have to.
Free diagnosis if you want help reading your own export: /