Written by Ram Kr Shukla, SEO and growth consultant

You open Search Console and the average position line has gone the wrong way. It was 8. Now it’s 19. You’ve paid for SEO for six months, and the one number that looks like a ranking just fell off a cliff.

Then you look at clicks. Flat. Maybe up a little.

That combination is one of the most common panic emails a local business owner sends, and most of the time nothing broke. The number moved because of what got added to it. This guide covers how Google computes the figure, why it falls while a site is growing, how to tell a fake drop from a real one in about 20 minutes, and the reporting mistakes that turn a small change into a scary one. It’s the same check I run in my local SEO work for service businesses, and two of them supply the real numbers below: a US roofing and exteriors contractor and a US lawn care and mosquito control company, both anonymised.

How Search Console Calculates Average Position

Google’s Performance report help defines it as “the average position of the topmost result from your site.” That sentence hides three rules, and every confusing drop comes from one of them.

Two quieter details matter later. Every item in a carousel gets the carousel’s position. And the chart is aggregated by property while the Pages table is aggregated by page, which is why the page rows never add up to the chart total. Google says as much in the same help page. Hold on to that, because it breaks a lot of reports.

Why Average Position Falls When Nothing Got Worse

The name for this is impression dilution. Average position is a ratio. When the bottom of the ratio grows with deep impressions, the result gets worse even if every ranking you had stays exactly where it was. Five things add deep impressions.

Google is testing your page in broader searches

When a page starts doing well for a specific term, Google often tries it in wider searches: a shorter head term, a national version of a local query, an adjacent service. Those tests usually land somewhere between position 15 and 30. Lots of impressions. Almost no clicks. Every one of them goes into the average. For local service businesses many of these test impressions come from outside the service area, which has its own diagnosis in why local service pages get impressions from outside their area.

New pages are ranking for long-tail terms

Publish ten new service or blog pages and they won’t enter at position 4. They enter at 40, 60 or 80 for hundreds of long, specific queries, each adding a few impressions at a deep position. Individually tiny. Together they hold the site average down for months, and that’s the normal shape of a growing site, as the honest SEO timeline lays out month by month. It’s also where the position 11 to 100 trap lives: a site ranking for thousands of terms on pages two to ten and getting almost nothing from them.

Demand shifted between your queries

If the query you rank best for gets fewer searches this month, it contributes fewer impressions, and your weaker queries take a bigger share of the weighting. Your rankings stayed put while the mix underneath them changed. Roofing and lawn care both swing hard with the weather and the season, so expect some of this every spring and fall.

Your device or country mix changed

Mobile and desktop rank differently, and so do countries. A jump in mobile impressions, or in impressions from searchers outside the US, changes the blend. Filter by device and by country before you trust the headline number.

Google changed what gets counted

This one runs in both directions. In September 2025 Google stopped supporting the num=100 results parameter that rank-tracking tools used to load 100 results at once. Lots of sites saw impressions fall and average position improve overnight while clicks stayed put (Search Engine Land covered the change). Nobody’s rankings got better that week. Fewer deep impressions were being logged. Same arithmetic, opposite direction.

A Worked Example: Same Rankings, Worse Average

Here’s the arithmetic with round numbers, for a made-up roofing site over two 28-day periods. Three queries bring in the work. In the second period, Google starts testing one page in broader searches.

Before: 400 times 3, plus 300 times 6, plus 300 times 8, comes to 5,400. Divide by 1,000 impressions and you get 5.4.

After: the same 5,400, plus 2,000 times 28, which is 56,000. That’s 61,400 over 3,000 impressions. Average position 20.5.

The chart says the site fell 15 positions. Not one of the three queries the business cares about moved. Clicks went from 65 to 67, because the test queries picked up two stray clicks.

The table shows two more things. A simple average of the four position values (3, 6, 8 and 28) gives 11.25, which matches nothing, so you can’t rebuild Google’s number by averaging rows. And if you compare only the three queries that existed in both periods, the set average is 5.4 both times. That fixed-set comparison is most of the diagnosis.

What It Looked Like for a Roofing Contractor

Here’s the real version, from a US roofing and exteriors contractor I work with. The owner asked why their soffit repair ranking kept going down. In Search Console, that query’s average position on its service page had slid to 19.7.

Split by day, the query had two kinds of day. On quiet days it got 3 to 10 impressions, at positions 5 to 7 early in the window and 7 to 9 three months later. Spike days came in three bursts, the longest running 16 days, and on those it got 50 to 150 impressions a day at positions 17 to 25. Zero clicks. That was Google testing the page in broader searches, mostly outside the area the company serves. The spike days carried so many more impressions that they set the average almost on their own.

Meanwhile the page was doing better than ever. Quarter on quarter, its impressions went from 8,170 to 32,053 and its clicks from 29 to 55. It was indexed, its canonical was right, and its review rich result was passing. An older page that used to compete for the same term had dropped to a single query, so this wasn’t cannibalisation either.

Not everything was dilution, though, and that’s the part worth copying. On the quiet days the page really had slipped 1 to 3 spots. And one local-modifier query (the service plus the town), which had sat at position 2 with 25 impressions, had stopped appearing at all. A real slip, hidden under a fake one. Stare only at the 19.7 and you’ll worry about the wrong thing and miss the right one.

The next 28-day comparison showed the reverse. As the broad fascia repair test impressions faded, the page’s impressions fell from about 12,600 to 7,200 while its average position barely moved, 11.6 to 11.9. Site clicks dipped from 211 to 188 the same month, which looked bad until brand and non-brand were split: about 17 of the 23 lost clicks were people searching the company’s name, down from 41 to 24. Non-brand clicks were roughly flat. That split is covered properly in branded vs non-branded traffic.

How to Diagnose an Average Position Drop

Run these in order. You need Search Console, about 20 minutes, and a spreadsheet for the last step.

Most drops are explained by step 3 or step 4. Steps 5 to 7 are for when the answer is mixed, like the roofing case, where dilution and a small real slip were sitting in the same number.

When the Drop Is Real

Dilution is common. It isn’t guaranteed. Treat it as a real ranking loss when you see any of these.

Google’s own guide to debugging Search traffic drops draws a line between a small drop in position and a large one, and it suggests looking at queries, pages, countries and devices separately. Do that for your fixed set and you’ll know which kind you’ve got. If it’s real, the usual suspects are in six real causes of a rankings drop, and the most common one on local sites is a change somebody made to the site.

Never Subtract One Date Range From Another

This is the mistake that turns a modest change into a headline. It happened while preparing a monthly report for a US lawn care and mosquito control company, and a recheck caught it.

The report needed first-half and second-half impressions for a rebuilt aeration page. The full month was 1,115 and the second half was 890, so subtraction said the first half was 225. A direct pull for the first half returned 609.

Search Console numbers aren’t additive. Data aggregated by page is counted differently from data aggregated by property, rare queries are anonymised and left out of query rows, and each date range is calculated on its own. On that site, for the same window, the page rows summed to 3,838 while the site total was 3,404.

The knock-on was big. The draft said the rebuilt pages’ impressions rose 325%. The true figure was 39%. It said the aeration page climbed 25 positions. It climbed 6. Still good news, just smaller than the draft said, and a client who later checks the numbers remembers the gap.

Freshness bites too. One day’s data went from 0 clicks and 30 impressions in a morning pull to 3 clicks and 215 impressions that evening. Google marks recent data as preliminary in the Performance report, and it does move.

What to Report Instead of Average Position

On its own, average position is a weak headline for a local business. It rewards sites that rank for fewer things. Put these in the monthly report instead, or at least next to it.

If you keep average position in the report, put impressions right beside it so anyone reading can see when the bottom of the ratio grew. And the new queries that caused the dilution are worth a second look: some of them are services you don’t have a page for yet. The service page gap analysis in Search Console turns that list into a build plan.