Ranking Volatility August 2026: What My Own Numbers Say

On Saturday morning I pulled the Search Console Search Analytics API for seo-kreativ.de, because volatility reports had been sitting in my news radar for two days. I had expected a drop. What came out was the opposite.

The days from 10 to 13 August were the strongest of the entire period I pulled, measured in impressions. Average position was better than the week before. Clicks stayed where they were.

That leaves four findings side by side that do not resolve into one another. The community reports chaos. An independent volatility index shows a mid-range value for the same days. Google confirms nothing. My property sits better than in the previous week. This constellation is not unusual during an unconfirmed wave, and it is the reason the question “Did this hit me?” does not get answered in a forum.

So this article is not a chronicle of the wave. How a wave like this unfolds is something I wrote up for the February 2026 volatility with nine waves and three fronts. This one is about the diagnosis. And about the uncomfortable question of how good the data actually is that you use to run it.

Search Engine Roundtable reported notable spikes across several ranking trackers on 13 August. The trackers named include AWR, AccuRanker Grump, Mozcast, Semrush and Sistrix. The report does not cite any confirmation from Google.

What such reports also contain are posts from SEO forums. Drops of more than 70 percent, traffic jumps of 200 percent, a minus of 25 percent against January. I deliberately left those numbers out of this article. None of them is tied to a traceable measurement series, and none of them tells you anything about your site.

A tracker measures a sample of keywords in one country on one device type. That is a legitimate quantity. It is simply not yours. How to read a sensor spike without over-reading it is covered in the piece on interpreting SERP volatility with the Semrush Sensor.

Rather than just repeating the reports, I pulled an independent measurement series. The DataForSEO SERP Volatility Index is freely accessible and can be filtered by country, device and search engine. On the method, DataForSEO writes: “DataForSEO checks Google and Bing SERPs for a fixed list of keywords and measures the volatility of results’ changes daily.” The scale runs from 1 to 10, where “one is characterized by minimal changes in search engine rankings and ten corresponds to dramatic changes in Google search algorithms and major updates to search engine algorithms”.

The calculation formula and the keyword list are proprietary and are not disclosed. That makes the index useful as an orientation about the market. It is not a yardstick for any single site. I pulled four cuts. Germany and the United States, each for desktop and mobile, category “All Categories”, Google. These are the values around the reported days:

The rise against 11 August is there, in all four cuts. It is just small. Measured against the median of the last 30 days, 12 August sits exactly on the median for DE desktop. For US desktop it sits below, for US mobile clearly below. For US mobile it was the 29th percentile out of 100, a quieter day than a good two thirds of the period.

The chart shows what a single-day report cannot. From 19 to 29 July all four cuts sat between 8 and 10 for ten days. That is the range DataForSEO describes as “Significant changes in SERP rankings, might be caused by a major update to the Google or Bing search algorithm”. Against those weeks, 12 and 13 August are unremarkable. The actual event was three weeks earlier.

On the question of an update, the page itself notes: “No major Google updates have been reported in the last 30 days.” That covers both periods. So the July spike was unconfirmed as well.

The Google Search Status Dashboard is Google’s own record of ranking updates. Four entries are listed there for 2026, and I have linked my write-up for each of them:

The most recent entry is therefore the June 2026 Spam Update, and between it and 15 August 2026 as the cut-off date of this analysis there are roughly seven weeks without an official announcement.

That does not mean nothing was changed during those seven weeks. According to Google’s ranking systems guide, Google improves its ranking systems regularly and gives notice of updates “when those might be useful to content creators and others”. The announcement is therefore tied to a usefulness condition, and the inverse conclusion from an empty dashboard to an unchanged algorithm does not hold.

In practice this means: the dashboard answers the question about an official communication and not the question about an actual change.

Does this point to a Google update in August 2026?

That is the obvious question. It is not answered by the trackers but by the calendar. Each of the three preceding years has an August update in the same dashboard:

A fourth August update is therefore quite possible. None of the three fits 12 August, though. The earliest start was the 15th, and two of the three came in the final third of the month. The August 2025 spam update began on 26 August, fourteen days after the day at issue here. Reading the spikes of 12 and 13 August as precursors shifts the pattern by two weeks.

Whether spam or core cannot be derived either: twice core, once spam. Google notes in its documentation on spam updates that the systems run constantly and that “we occasionally make notable improvements to how they work”. Occasionally is not a calendar.

So the answer has two parts. An update in August 2026 is plausible on the record. The window for it is only just opening with the 15th. The spikes of 12 and 13 August are still not a signal for it. If an entry appears in the dashboard over the next two weeks, nothing about the diagnosis below changes. You would only have a confirmed cause to hold your own curve against.

All values below come from the Search Analytics API of Google Search Console for the property https://www.seo-kreativ.de/, pulled on 15 August 2026 with dataState=all and type=web. The comparison covers two Monday to Thursday windows so that the weekdays line up.

The four daily values in the event window are 16,250, 17,183, 16,743 and 17,162 impressions. Among the 34 complete days of the period they take ranks 1, 2, 4 and 5. Sitting between them is 13 July with 16,834 impressions, a day without any volatility report at all. The 14th and 15th of August stay out of the comparison because the data there was still incomplete.

The query level looks similar. Comparing the two windows gives 436 improvements of at least three positions against 326 deteriorations, and with a noise filter of 20 impressions, 22 improvements against 9 deteriorations remain. Of those 31 entries, 27 sit on English-language pages, and 19 of the 31 move in the range between position 20 and 45.

And that is exactly where the first limitation sits. Position 20 to 45 is the range with the largest fluctuation in my data anyway, which is why I would not derive any effect from these movements, in either direction. What I do derive is weaker and therefore sound: nothing collapsed on this property in this window.

Plus 24 percent sounds like a lot. What I still need to know is how much this series swings between any arbitrary pair of weeks. Until I know that, I cannot place the value. The number in my own headline is as affected by this as anyone else’s.

The calculation is simple. I build every pair of two non-overlapping four-day windows possible within my period and look at how their changes are distributed. Then I locate my observed value inside that distribution. If it sits in the middle, I have measured nothing.

Google first, so my own number. Thirty-four complete days produce 378 such pairs. Their median sits at plus 4.5 percent, and the band between the 10th and the 90th percentile runs from minus 22.8 to plus 29.5 percent. My rise of 23.9 percent lands in the 86th percentile. That is high, but not an outlier, and roughly every seventh arbitrary weekly comparison in my own data comes out larger.

Then Bing, which was meant to be the control group. A single property does not answer whether the rise came from Google or from demand. That needs a second search engine. This comparison is not a clean one to begin with: Bing has different volumes, a different index depth and a different user base, so it never measures exactly the same thing as Google. I therefore exported the search performance report from Bing Webmaster Tools for the same days.

These are the daily values of both sources side by side, for the same eight days:

Bing fell by 17.4 percent while Google gained. Two curves moving apart, and therefore an indication of something Google-specific. It does not hold up once you check it. On this property Bing delivers 258.6 impressions a day on average, with a standard deviation of 99.9. The median of all 253 window pairs sits at minus 15.0 percent, and the observed minus 17.4 percent lands in the 47th percentile. Any arbitrary four-day comparison in this data returns roughly the same result.

As a control group, then, Bing delivered nothing. The question of whether the rise was Google-specific stays open. The test delivered something else, namely the calibration of my own number. That is the part that stays.

And that is where all this arithmetic becomes practical. A tracker spike does not have to mean anything for your domain. For the same days, my own numbers tell a different story than the report. Nothing collapsed. There was nothing to repair. Anyone who looks into their data after a volatility report and finds nothing there has not done the measurement wrong. They simply had a quiet week.

While analysing this, I noticed that two numbers from the same API do not match. At day level my event window totals 67,338 impressions and 165 clicks. Querying the same four days by search query and page gives me 34,170 impressions and 28 clicks.

This is not an error in the query. The Search Console help on dimensions states: “Some queries are omitted from the report to protect user privacy.” And further: “They’re included in chart totals, unless a query filter is applied.” So the grand total contains everything. The table by search query does not.

For update diagnosis this has an uncomfortable consequence. If you open the performance report after a volatility report, group by search query and see nothing unusual there, then with coverage like mine you have not looked at half of your impressions. In my case it was considerably more for clicks.

How large the gap is on your site is something you have to work out yourself. It depends on how many of your search queries are rare enough to be anonymised. My rule of thumb for this: I recalculate the coverage once a quarter and start every diagnosis at day and page level, not at query level. For the analysis below that, the RegEx copy templates for Search Console help.

Step 1: day level, matching weekdays. Compare the event window with the same weekday block of the previous week. Not with the previous month and not with the previous year. Weekdays regularly have their own pattern in the projects I handle at my SEO consultancy. A Monday against a Saturday then produces a drop with no cause. Leave out the last two to three days. Google states in the documentation on Search Console data: “Normally, however, collected data should be available in 2-3 days.”

Step 2: read clicks and impressions separately. If both fall, a visibility issue is plausible. If only impressions fall while clicks hold, the cause is more likely to sit in serving or in counting than in a ranking loss. I took that exact case apart for seo-kreativ.de in July, written up under Impressions collapsed? How to narrow down the cause.

Step 3: page level before query level. The page dimension is not affected by query anonymisation. Sort by impression difference and look at the ten biggest losers. Only once you have an affected page is it worth looking at its search queries.

Step 4: only now look outward, and do it with numbers. If your own data shows something unusual, the question about an update makes sense. Before that it does not. Check the Status Dashboard for an official entry and pull a volatility index for your market and your device instead of reading a headline. The decisive question is not whether the index rose, but where the value sits compared to the last 30 days.

On 12 and 13 August 2026 several trackers spiked, the DataForSEO index sat mid-range within its own last 30 days, the Status Dashboard recorded no update, and my property sat better than in the week before, though only in the 86th percentile of its own fluctuation. All four sentences are true at the same time. They only contradict each other if you assume that a volatility report contains a statement about your own site.

What surprised me most in the index series is not 12 August. It is the ten days from 19 to 29 July, when all four cuts I checked sat between 8 and 10. I did not register those weeks as an event at the time, and they were not in my news feed either.

What occupies me more about this analysis than the wave itself is the coverage gap. I have worked with the query level of Search Console for years and had never quantified the share. In this window it was 50.7 percent of impressions. That changes nothing about the data. It changes the order in which I look at it.