A client page showed 330 impressions over ninety days at an average position of 24. Read that as demand and you would write the page off. Strip out the search operators and 122 impressions remained, at position 34 — and inside those 122 were two buyer queries already sitting inside the top ten. Three different numbers for one URL, and only one of them describes a customer.
208 of those 330 impressions, 63%, were site: and allintext: queries. That is me and the site owner inspecting the page, recorded by Search Console as though it were demand.
Your own inspections are in your data
When you run site:example.com/page to check indexation, Google logs an impression. So does allintext:, inurl: and intitle:. Search Console does not distinguish between a buyer and an SEO holding a magnifying glass.
On a page with real traffic this is rounding error. On a page that is nearly invisible — which is exactly the kind of page you inspect repeatedly — it can be most of the record. The pages you check most are the pages this distorts most.
Why it gets the decision backwards
Two failures, pulling in opposite directions.
Leave the operators in and average position looks better than it is. Operator queries return your page at position 1, because you asked for that page by name. Those ones drag the average toward the top and hide how far the real queries sit from page one.
Then the volume looks like demand that is not there. In the case above the page appeared to have 330 impressions of interest. The honest figure was 122, and the recommendation that came out of the blended number was to stop linking to the page at all. That recommendation was wrong, and the operator filter is what showed it.
The filter, and where to apply it
In the Search Console interface, add a query filter set to Custom (regex), choose Doesn’t match, and use:
(site:|inurl:|intitle:|allintext:|allinurl:|cache:|related:)
Through the API, pull the query dimension and drop matching rows before you aggregate. Either way the rule that matters is this: apply it to both sides of any before-and-after. Filter the after and not the before and your improvement gets divided into an inflated baseline, so real lift reads smaller than it was.
Never read a page average as a position
The wider lesson is the one the three numbers demonstrate. An average position across a page is a weighted blur of every query that page appears for. It is not a location. The page above averaged 34 while ranking 6.8 for “who is the best seo consultant to help me grow my $10m arr b2b saas” — a specific, high-intent, conversational query that a person actually typed.
Aggregate to the page and that query disappears into an average that describes nothing. Look at the query rows and the page’s real situation is obvious in about a minute.
Check this before your next audit
- Any page you have inspected a lot. Staging URLs, recent migrations, pages you were asked to diagnose.
- Any page with impressions and no clicks. That pattern is what operator traffic looks like.
- Any baseline you are about to measure against. Filter it now, before it becomes the number you are judged on.
Related: how I run a technical audit, and fractional SEO services if you would rather not do this yourself.
The limits of this
One property, ninety days, measured on 11 September 2026. The client is not named because the numbers are theirs. I have not sampled enough sites to tell you how common a 63% operator share is, and I would expect it to be far lower on anything with real traffic. What I can tell you is that it was high enough here to reverse a recommendation, and that checking costs one regex.