On 4 August 2026 our site was opened for indexing for the first time. Until that day it sat under a site-wide noindex, and over the previous 28 days Google had shown it exactly three times. What followed was a month of data from a blank sheet: no history, no seasonality, no causes stacked on top of each other.
That makes it useful for one thing. It shows what «search volatility» looks like from the inside when we know for certain that nothing happened to the site — except for one renaming, which we will come back to.
Below are the daily figures for our own site in August and the method we use to separate movement in search from a fault of our own. The data was pulled from Search Console on 2 September 2026 for the period 3–31 August.
A day lies, a week tells the truth
Four consecutive days from the middle of the month:
| Date | Impressions |
|---|---|
| 15 August | 55 |
| 16 August | 148 |
| 17 August | 381 |
| 18 August | 137 |
In two days impressions grew sevenfold, and the next day they fell almost threefold. A week later the same thing repeated elsewhere on the graph: 252 impressions on 25 August, 579 on 26 August, 321 on 27 August.
Now the same days grouped into weeks:
| Week | Impressions |
|---|---|
| from 3 August | 421 |
| from 10 August | 858 |
| from 17 August | 2,127 |
| from 24 August | 2,250 |
No break anywhere. Steady growth that the daily graph does not show at all.
It is the same sample, only the step is different. A several-fold swing over a single day is normal behaviour for search, and the smaller the numbers the stronger it gets: one page picks up a broad query and adds three hundred impressions, the next day it is shown less often and loses two hundred. The cause is not the site — it is how many times somebody searched that day.
The practical conclusion is simple and inconvenient: a daily graph is not grounds for any conclusion. Compare stretches of equal length — a week against a week, four weeks against four. If a drop is visible only in daily data and disappears in weekly data, there is nothing to fix.
The other half of the same rule: there is nothing to celebrate either. On 26 August our site collected 579 impressions — the best day of the month — and exactly zero clicks. The day before, on 252 impressions, there were three clicks. A one-day record means precisely as much as a one-day collapse: nothing.
The average position trap
Now the most treacherous number in the console. Our average position by day:
| Date | Average position |
|---|---|
| 4 August | 5.3 |
| 16 August | 58.7 |
| 28 August | 29.6 |
Read literally, this is a story of catastrophe and heroic recovery. Neither happened.
On 4 August the site had only just opened, a handful of pages were in the results, and nearly all of them for our own name: 54 impressions, 7 clicks, position 5.3. Then Google started finding the rest — by the end of the month 562 addresses had received impressions. New pages enter the results from the bottom, from the fiftieth and hundredth positions, and every one of them drags the average down. Once they outnumbered the brand pages, the average slid to 58.7 — while no individual page had lost anything.
By the end of the month the average came back to the 30–35 range for the mirror-image reason: some of those new pages climbed out of the hundreds into the second and third dozen.
The weekly averages show the same shape: 34.9 → 54.4 → 40.0 → 33.6. The curve looks like a collapse followed by a recovery, although what it actually describes is not where the site ranks but how fast new pages were entering the index. Impressions over those same weeks did nothing but grow.
One example of how easily a single page moves that figure. Our SEO audit section collected 1,193 impressions over the month at an average position of 57.8 and zero clicks. It is the most-shown page on the site — and on its own it noticeably drags the site-wide average down, although nothing is wrong with it: it simply sits deep for broad queries.
The rule that follows: site-wide average position is not a health indicator. Look at the position of a specific page for a specific query and compare it with the same set of queries in the previous period. We broke down how impressions are spread across positions on our site in the one-day audit — seventy percent of impressions sit below position twenty, and that, not the moving average, is the real picture.
Our own breakage that looks like volatility
One thing did happen to the site in August. At the end of the month we renamed the sections of the service network to localised addresses and put permanent redirects on the old ones.
Here is what that produced in the data: of the 562 addresses Google showed the site for, 52 are old ones that no longer exist on the site. They answer with 308 and collected 1,065 impressions — almost a fifth of the total. Those figures and the check behind them are covered separately; what matters here is different: on a graph this is indistinguishable from «the algorithm recalculated something». Impressions crawl from the old address to the new one unevenly, over several weeks, and during that time both pages keep appearing and disappearing from the results.
There is one difference — the outline. A fault of your own almost always has a boundary: one section, one language version, one page template, one type of address. A search update does not behave that way. So the first question after a drop is not «when was the update» but «what exactly dropped». How to check addresses after a move is in the breakdown of redirects after a migration.
Five steps we use to separate the two
This is what a traffic drop diagnosis looks like in practice.
1. Break the drop down instead of staring at the total. In Search Console, one dimension at a time: pages, queries, countries, devices, language versions. The goal is to find the outline. While all you can see is «minus forty percent site-wide», there is nothing to check, because everything has to be checked.
2. Match the date against two calendars. The first is Google's public list of updates: the Search Status Dashboard shows what rolled out and when, and dates should be taken from there rather than from memory or somebody's post. The second calendar is your own: releases, address changes, a new template, a hosting move, a bulk text edit. A match with the second calendar happens more often than anyone would like.
3. Separate a drop in impressions from a drop in clicks. These are different diagnoses. Our own example: in the week from 3 August, 421 impressions produced 17 clicks; in the week from 10 August, 858 impressions produced 2 clicks. Impressions doubled, clicks fell several times over — and nothing had broken: the mix of queries the site appeared for had changed. If clicks fall while impressions hold, the question is about the snippet and the position. If impressions fall, the question is about indexing and demand.
4. Run the technical check in a fixed order. Response codes for every address that received impressions; redirects — where they lead and whether they chain; sitemaps with unique addresses counted rather than lines trusted. The most frequent finding here is a second, outdated sitemap living alongside the canonical one and handing the crawler a different, incomplete list of pages. What to do about it is in the breakdown of sitemaps and robots.txt.
5. Compare the two periods mechanically. Doing it by eye works badly: the console is convenient for totals and inconvenient for a difference spread across hundreds of rows. That is why we built a Search Console period comparison: it shows which pages and queries actually lost impressions, not by how much the total sagged.
When to wait instead of act
After addresses are renamed, search does not rebuild instantly. Technical changes are re-crawled in roughly 2–6 weeks — hence our checkpoints at 4 and 8 weeks after the fixes. In that window the graph behaves exactly as it does at the top of this article: in jumps.
The most expensive mistake here is to edit again before waiting it out. The second change lands on top of the first, and afterwards there is no way to say what came from what. If there is no movement after eight weeks, the hypothesis was wrong, and we write that down as a result too.
In practice this implies a different order of work: a fix plan is better split into steps with pauses than shipped in a single day. If addresses, template and texts all change within one week, the next measurement shows only the combined effect — and if that turns out negative, everything has to be rolled back, because there is no telling which part did the damage.
How much crawling pages actually get and where it goes is a separate breakdown: where a site's crawl budget goes.
Honest limits
Three things this data does not prove.
A month of data from a young site is not statistics. 5,656 impressions and 41 clicks in August are enough to show the mechanics of the swings, but not enough to claim any regularity. On a site with thousands of clicks a day the daily jumps will be smaller in percentage terms, and the threshold for «this is a drop now» is different there.
A diagnosis does not promise traffic will come back. Technical causes — a language version dropped out, a redirect broke, a sitemap returns an empty body — are usually reversible. Losing positions because competitors built better pages is reversible only through work, and certainly not in a week. We split the findings into «we will recover this», «we will partly recover this» and «we will not recover this», and we name no figures we will return to.
Sometimes the cause is not in the site at all. A season, a shift in demand, a change in how results are laid out for your query — any of these produces a drop without a single mistake on your side. That is why we check the trend against demand in your industry before looking for culprits inside.
If traffic has fallen and the cause is unclear: the first drop diagnosis is free and takes 2–4 working days — the breakdown, the outline, the date and a hypothesis with evidence behind it. If it turns out there are several causes, a full audit follows, 5–25 working days depending on how many. Work under a written contract, 30-day warranty from the date the acceptance act is signed. Send your site in for a diagnosis.
