Search Console's average position is not your rank

6 min read

Search Console is free, it's first-party data straight from Google, and it has a column called Average position. So it looks like the obvious place to find out where you rank.

It isn't, and the reason isn't that the number is wrong. The number is accurate. It's just not answering the question you asked.

What that column actually averages

Search Console records an impression every time a page of yours appears in someone's results. Each impression carries the position your page held in that particular search — for that query, in that country, on that device.

“Average position” is the mean of all of those, weighted by how many impressions each one produced. Which means it isn't your position for a keyword. It's your position across every keyword Google decided your page was loosely relevant to, blended together.

Those two things can be extremely far apart.

A worked example

Say you have a page targeting best running shoes, and it genuinely sits at #4. Over a month, Search Console records a thousand impressions for that page, spread like this:

The arithmetic Search Console does:

(600 × 4) + (150 × 22) + (120 × 46) + (130 × 88)
--------------------------------------------------  =  22.7
                     1000

You rank #4. Search Console reports #22.7. Both are correct. Only one of them is the thing you wanted to know.

And notice that the error only runs one way. There is no position better than #1, so every extra query you appear for can only pull the reported number higher — which is to say worse. The more things Google decides you're marginally relevant to, the worse your headline position looks, even when your actual ranking hasn't moved at all.

The artifact that catches people out

Here's the one that genuinely surprised me. Average position is computed only over impressions. If a query stops producing impressions, it stops contributing.

So imagine one of those long-tail queries drops from #46 to somewhere past #90. Almost nobody scrolls that far, so the impressions for it collapse — and its heavy, average-dragging weight collapses with them.

Your average position improves. Because you got worse.

It works the other way too: publish something that starts ranking at #70 for a batch of new queries, and your average position gets worse while your traffic goes up. The metric moves for reasons that have nothing to do with the ranking you care about, which makes week-on-week comparisons close to meaningless.

You can get a real number out of it

To be fair to the tool: the data is in there. You have to ask more narrowly.

• Filter to one query. Not the page — the query. Then the average is over impressions for that keyword only, which is a number that means something.
• Filter to one country. Positions differ per market, and the unfiltered view blends them.
• Filter to one device if mobile and desktop matter differently to you. They often do.
• Remember the delay. Search Console data lands a couple of days late, so it can't tell you about today.

Do all of that and you have a usable measurement. What you don't have is something you'll keep doing every week for twenty keywords, which is the actual problem — the friction is why people end up quoting the blended number instead.

Your own Google isn't a control either

The usual next move is to just search for the keyword and count. That's a reasonable instinct and it's also not neutral: results are shaped by your location, your account history, and your device. You will tend to see yourself higher than a stranger does, partly because you visit your own site constantly.

I've written up the practical side of this — incognito checks, what they do and don't fix — in how to check your Google ranking for a keyword.

What a measurement has to do to be trustworthy

Strip it back and there are only four requirements, and Search Console's headline number satisfies none of them:

• One keyword. Not a blend of everything you were marginally relevant to.
• One country. Chosen deliberately, because it changes the answer.
• De-personalised. The result a stranger gets, not the one your own history produces.
• Repeated identically. This is the one people skip. A single check is a snapshot; the only reason to care about position is to see it move, and movement only means something if the measurement didn't change underneath it.

That last point is why the fix isn't a better dashboard reading. Two numbers taken the same way a week apart tell you more than one number taken carefully today. A blended average taken twice tells you almost nothing, because you can't tell which part of it moved.

If you want a one-off de-personalised reading for a single keyword right now, there's a free checker here that needs no account. But the honest advice is: whatever you use, pick one method and repeat it. Consistency beats precision, and Search Console's average position gives you neither for the question “where do I rank?”