Noise vs. Signal – Issue 04

There was no August core update, and no new AI content rule. Plenty of people spent the month reacting to both.

If your traffic wobbled in August and somebody forwarded you an article about “the August 2026 core update,” this one’s for you. The month gave us one real spam update, two explanations Google never gave, and a reporting error in the middle that quietly broke a lot of people’s baselines.

The movement itself was measurable. Third-party rank trackers registered heavy movement around August 1 to 3, again on August 5 to 6, and again on August 12 to 13. Nobody disputes that part. What nobody can tell you is what caused it. Google’s Search Status Dashboard, which is the official log of confirmed ranking incidents, recorded nothing at all across that first window. No ranking incident, no indexing, no crawling, no serving. Google does ship thousands of unannounced adjustments a year, so silence never means nothing changed. It just means nobody outside Google can honestly say what.

That gap didn’t slow anyone down. By August 8, vendor blogs were publishing recovery guidance for a “broad ranking recalibration” that supposedly targeted pages with weak originality and thin evidence, remediation steps included, for an update Google has still never confirmed. This deserves its own category, separate from the usual hype and separate from the usual doom. Somebody invented the event. And events get invented because an unnamed problem is pretty much impossible to sell a solution for.

There was one confirmed ranking event in August, and it landed a full fortnight after the panic. Google released the August 2026 spam update on Tuesday August 18 at around 12:30pm Eastern, and confirmed it had finished rolling out on August 21. Google’s entire public description: a normal spam update, all languages, all locations. No blog post, no new policy, no impact figure. Google also told Search Engine Roundtable that this one doesn’t target link spam or the site reputation policy. So if you weren’t breaking the spam rules in July, nothing new applies to you now.

Then there’s the detail that caught even careful people. Google logged two entries on its Search Console data anomalies page: a Generative AI (Search) logging error running August 13 to 17, and a separate one-day Discover error on August 13. So some of the mid-August “drop” people were busy diagnosing was a hole in Google’s own reporting rather than a change in their rankings. Google restored the missing data on August 21 and the reports now show complete metrics, which means anyone who drew a conclusion in that window drew it from numbers that no longer exist.

The second invented story is the one you’ve probably heard more of, because it sounds more serious. A wave of August coverage announced that Google was now treating mass AI-generated content as spam. Google announced nothing of the kind. The rule those articles describe is scaled content abuse, and it has been sitting in Google’s spam policies for years. It says that publishing many pages primarily to manipulate rankings is a violation, and that the method doesn’t matter, which means AI writing tools were already covered before anyone wrote a headline about it. The one genuinely recent change came in May, when Google added AI answers to the list of things you’re not allowed to manipulate. We covered that in the June issue. Nothing was added in August.

Which is not the same as saying nothing happened to those sites. On August 31, consultant Glenn Gabe published four case studies on what the update actually hit, and AI-written content runs through nearly all of them. A site in a high-stakes health and money niche that combined programmatically generated pages with AI-written filler lost rankings for more than 200,000 queries outright. Not dropped a few positions. Gone. A turn-key Amazon affiliate site, with blocks of text under its category listings testing at 97% probability of AI generation, lost over 14,000. A third site with 1.5 million indexed pages took damage across the whole domain rather than just the offending section. Gabe also notes the drops carry into AI Overviews and AI Mode, and into ChatGPT when it grounds its answers through Google.

Sources: Google’s Search Status Dashboard ranking history, where the absence of a logged incident is the evidence, and Google’s Search Console data anomalies page, which logged the August 13 to 17 Generative AI error and its restoration on August 21 (the sources that count) · Search Engine Land and Search Engine Journal on the spam update timing · Barry Schwartz at Search Engine Roundtable tracked both the unconfirmed volatility spikes and the confirmed rollout · Glenn Gabe’s four case studies on what the update hit, published August 31. Gabe sells algorithm recovery services, so he has a commercial interest in spam updates looking severe. He publishes the visibility charts behind each case and blinds the sites, which is why we’re using him, but you should know the conflict is there. The vendor recovery posts described above are deliberately not linked: naming them is the point, sending them traffic is not. And a lot of the August coverage in circulation is low-quality content built to rank on the update name, with dates that contradict each other, so everything load-bearing above comes from Google’s own logs or from Schwartz.

On some searches the AI answer now opens all the way by default. Everything else moves down the page.

On August 27, Chris Long, co-founder of the SEO agency Nectiv, posted that Google was showing the AI Mode prompt box by default on some searches. Barry Schwartz replicated it. The next day Google confirmed it in a statement: for some queries, AI Overviews may dynamically expand for topics where its systems decide that’s most useful, and Google says its research shows people find Search more helpful and explore follow-up questions more deeply that way.

Mechanically, it works like this. An AI Overview normally shows a compact summary at the top of the results, with a “Show more” button if you want the rest. In the expanded version that button is gone. The full answer loads immediately, and an Ask anything box sits underneath it, ready to start an AI Mode conversation. The ordinary results, ten blue links included, begin below all of that. On a laptop screen, a person may never see them without scrolling.

Definitions are the most consistent trigger. Long’s example was the single word “algorithmic” on a desktop search, and the expanded answer filled the screen. Google did not say what share of queries this affects, or whether it plans to widen it. Robby Stein from Google added one more detail: if you have already started scrolling, the expansion stops, so you don’t lose your place on the page.

This is where the numbers further down this issue come from. Brainlabs found the losses concentrated in educational and discovery searches, the ones where somebody is trying to understand a thing rather than buy it. This is the machinery behind that number. If a meaningful slice of your traffic arrives on explainer posts and definition pages, this is the shape of the pressure on them, and it is not going to reverse. If most of your traffic lands on pages people reach after they have already decided they need what you sell, you are further from the blast radius.

There is no clever response available, and I want to say that plainly, because somebody will try to sell you one. No markup, no schema, no file you can add gets your page a larger share of an expanded overview. You could switch off the generative AI control in Search Console, which we covered in July, but that removes you from AI answers without giving you anything back in ordinary search.

Sources: Google’s spokesperson statement to Search Engine Roundtable on August 28, plus a follow-up from Robby Stein of Google on the scroll behaviour (the sources that count) · Search Engine Land and 9to5Google on the behaviour and where it shows up · spotted by Chris Long of Nectiv and replicated by Barry Schwartz, who got Google on the record. Bear in mind this was never announced. It is an observed behaviour Google confirmed when asked, with no percentage of queries disclosed and no published measurement of what it does to clicks, so treat the scope as provisional and expect it to change.

Your AI search data arrived this month. The useful half is hiding in the old report.

Back in June we covered Google launching a dedicated generative AI report inside Search Console. It was a partial rollout at the time, so most of you couldn’t actually see it. That changed at the end of August. Google added a note to its original announcement confirming that as of August 31, 2026, the reports are live for all websites worldwide. So if you have a Search Console account, go open it. It’s there now.

The report gives you one thing and holds back three. You get impressions inside Google’s AI features, meaning AI Overviews, AI Mode and the generative features in Discover, broken out by page, country, device and date. You don’t get clicks, click-through rate, or a single query. And it only lives in the Search Console interface, never in the data export, so anything you want out of it comes out by hand.

The more useful discovery of the month was a bit of a happy accident. In early August an SEO practitioner named Anastasia Kourou posted a screenshot of her performance report on LinkedIn and asked Google’s John Mueller a fair question: why is Search Console listing things like “yes”, “yes go on” and “yes, pricing” as search queries? Nobody types that into a search box. Mueller’s answer was short, and it pointed at documentation that had been sitting there the whole time. Search Console includes AI Overviews and AI Mode data in the general performance report. Barry Schwartz wrote it up on August 6.

The reason is simple once you see it. Inside AI Mode, every follow-up someone types counts as a brand new query, and everything in the answer that comes back gets attributed to it. So when a person asks Google’s AI about your category, reads the answer, types “yes, pricing” to keep going, and your page turns up in what comes back, Search Console records an impression for your page against the query “yes, pricing.” Those conversations have been quietly landing in your ordinary performance report for months, with clicks and positions attached. Which is more than the dedicated AI report gives you.

They’re easy to miss for a pretty mundane reason. The usual trick for spotting AI-style prompts is to filter for unusually long queries, ten or more words. These fragments are one to three words, so they fail that filter from the opposite direction and read like ordinary short-tail data. Jean-Christophe Chouinard, an SEO strategist at Tripadvisor, published a pattern on August 14 that catches the conversational fragments directly in the report’s query filter. And the missing click data is on a clock. The UK’s competition regulator has already ruled that Google must give publishers impressions, click-throughs and click-through rate, with the page-level controls due by March 2027.

Sources: Google’s Search Central announcement with its August 31 global rollout note, and the Search Console help page for the report (the sources that count) · Search Engine Land and Search Engine Journal on the global rollout and the missing click data · surfaced by Anastasia Kourou, confirmed by John Mueller, documented by Barry Schwartz, with filtering methods from Jean-Christophe Chouinard and Glenn Gabe. The impressions, clicks and click-through-rate obligation and the March 2027 date come from the CMA’s June 3 conduct requirement.

Sources: Brainlabs analysis of Google Analytics data from 54 advertiser clients, January 2025 to April 2026, reported by Digiday in August 2026, with commentary from Sam Paez, senior manager of SEO at Brainlabs, on which query types are being absorbed. This is agency-held first-party data rather than an independent study, and the underlying dataset has not been published for review. We’re using it because it’s measured client data rather than a vendor survey, and because the caveats are disclosed rather than buried.