How to improve average position in Search Console (and what the number hides)

What average position actually measures

Most people read this number as a rank. It is not one, and that misreading causes most of the confusion around it.

Google's own definition: position is the topmost position occupied by a link to your property in search results, averaged across all queries. Three things follow from that, and all three matter.

• Only your best result counts per query. Rank 2, 4 and 6 on one query and Google records 2. Your other listings are invisible to this metric.
• Every query with an impression is in the average. Including the hundreds of junk long-tail queries you never targeted.
• An AI Overview holds a single position, and every link inside it is assigned that same position. So a citation inside an AI Overview can register as position 1 — and it is not a blue link at all.

A real turnaround, shown with its flaws

We ran the organic desk for a US marketing agency. Here is the Search Console compare, exactly as it was reported.

The raw totals on the screenshot read better than that: impressions 39.3k → 537k, clicks 1.71k → 5k. We are not using those, and here is why.

Three reasons to discount part of our own result

Every agency case study you have read leaves this section out. It is the most useful part of the page.

1. 01The windows are not the same length. Search Console compared 273 days against 121. The current period is 2.26x longer, so the raw totals are inflated. Per day, impressions rose about 6x and clicks about 1.3x — real, but a long way from 13.7x and 2.9x.
2. 02Google says the impressions data was faulty. Google's own Data anomalies page states a logging error prevented Search Console from accurately reporting impressions from May 13, 2025 until April 27, 2026. That covers nearly all of both windows. Treat impression totals as directional.
3. 03A Google change helped us. Google disabled the `&num=100` parameter around September 12–14, 2025 — right between our two windows. That removed scraper-driven impressions at positions 11–100 and improved reported average position across most of the web. Some portion of 51.1 → 13.2 is that, not us.

So what is left that we can actually claim? One thing, and it is the interesting one. After `&num=100` was removed, 87.7% of studied properties lost impressions. This property's impressions per day went up about 6x through the same change. Losing what everyone else lost and still growing six-fold is the part the work earned.

And the metric that moved against us

CTR fell from 4.3% to 0.9%. It is on the same screenshot. Here is the honest read.

• A filling long tail buys cheap impressions. You start appearing for hundreds of queries where you rank at depth and nobody clicks. The denominator explodes.
• AI Overviews take clicks off the table. Pew Research found people clicked a traditional search result on 8% of visits when an AI summary appeared, versus 15% without one — and just 1% clicked a link inside the summary.
• Both can be true at once. In this case the click count still rose per day. The rate fell because we were being shown far more often, in far more places.

If your CTR falls while clicks rise, that is usually growth. If CTR falls while clicks fall, that is a problem. Never read the rate on its own.

Read the band, not the number

We triage by band. These are our operating rules from running these programs — not Google benchmarks.

Find your symptom

Most people land on this topic because something looks wrong. Match the symptom before you touch anything.

Step 1: Structure before content

The first fix on this account was not writing. It was arranging. Pages were competing with each other, and nothing linked in a way that told Google what mattered.

• One page, one job. Two pages chasing the same query means neither wins. We merged and redirected.
• A clear hub for each theme, with supporting pages linking up to it using descriptive link text. Google's starter guide says link text tells users and Google what the linked page is about.
• Money pages close to the homepage — we aim for two clicks. That is our operating rule, not a Google threshold; what Google has said is that pages far from your main navigation get crawled less often.
• Breadcrumbs everywhere, marked up so the relationship is machine-readable.

Step 2: On-page work, but only where it can pay

We only touched pages already ranking between 6 and 30. A page at position 60 needs structure, not a new title tag.

1. 01Title tag matching how people actually search, with the primary term near the front.
2. 02A heading that restates it in human language. Google does not care how many H1s you use — this is for readers and for anything parsing your page.
3. 03A direct answer up top. That is our house rule for making a page easy to quote, not a documented Google threshold.
4. 04H2s written as the questions the page actually answers.
5. 05Named entities — products, cities, tools, brands — instead of vague phrases.
6. 06One clear next step per page. A page that ranks and converts nothing is a hobby.

Step 3: Publish on a rhythm, and expect the dip

Ten posts in one week then nothing for two months teaches Google nothing. A steady cadence does. But be ready for what it does to your headline metric.

Publishing more pages surfaces you for more long-tail queries, and new pages enter at weak positions. That mathematically drags the site average worse even while the program is working. This trips up more teams than any other single thing on this page.

• A brief per piece: target query, the intent behind it, the competing pages, and which internal links must appear.
• Every new post links to a money page. Every money page gets a fresh supporting link each month.
• Refresh before you write. Updating a page at position 12 usually beats a new page starting at 60.
• Kill what does not earn. Thin pages with no impressions after six months get merged or removed.
• Track a fixed set of named target queries alongside the aggregate, so dilution cannot hide progress.

What we did not do

• No bought links.
• No mass AI publishing. Google's spam policies treat generating many pages without adding value as scaled content abuse — and that applies to human, AI, and hybrid production equally.
• No keyword stuffing or doorway pages by city.
• No changing dates to look fresh. Google's helpful-content self-assessment asks about that by name.
• No monthly ranking PDF nobody reads. Search Console was the scoreboard.

A timeline, from our own engagements

This is what we have seen across the programs we run. It is not a forecast — competition, site age, and existing authority move all of it.

Do this in the next 30 days

1. 01Open Search Console → Performance → Queries and sort by Position. There is no position filter — position is a metric, not a dimension — so export to Sheets and filter to rows between 6 and 15 there.
2. 02List the pages that own those queries. That is your work queue, in order.
3. 03For each: sharper title, a direct answer up top, three internal links in, one clear next step.
4. 04Find money pages with fewer than three internal links. Link to them from your best-performing posts.
5. 05Write down 20 named target queries. Track those separately from the site average, forever.
6. 06Set a publish cadence you can hold for six months, even if it is two posts a month.
7. 07Re-check in 30 days. Watch clicks and named-query position. Ignore the aggregate for now.