Google’s August 2026 spam update, and why unedited AI content lost.

Between August 18 and August 21, 2026, Google rolled out its third spam update of the year. It shipped with no blog post and no new policy, and a lot of people are calling it a core update. It wasn't one. But the sites that lost looked very alike, and they matched what Google has been saying about scaled content for two years. Here is what happened, what Google says it targets, what we saw across the sites that dropped, and what to change so your content holds up through the next one.

What actually happened in August 2026

Google announced the August 2026 spam update on its Search Status Dashboard on August 18, a little after 12:30pm ET, and marked it complete on August 21 at 4:51am ET. That is a rollout of about two days and sixteen hours, which is fast even for a spam update. It applied globally and to all languages.

Google called it a “normal” spam update. There was no accompanying blog post and no new spam policy. It came after spam updates in March and June, which makes it the third of 2026.

Normal or not, it hit hard. SE Ranking tracked US results through the rollout and found that 16.71% of URLs that had been in the top 10 fell out of the top 100, against 9.2% over a comparable stretch in July. That is roughly 82% more pages disappearing entirely than in a typical week. These were not small shifts of a few positions. Pages were effectively removed from view.

It was not a core update

A lot of the coverage you will find calls this the “August 2026 core update.” Google has run two core updates this year, in March (March 27 to April 8) and May (May 21 to June 2). There was no core update in August. The difference matters for diagnosis:

• Core updates re-assess overall quality and relevance. Losses are often relative. Someone else got better, or Google re-weighted what a good answer looks like.
• Spam updates are aimed at pages that break a written policy. Losses tend to be steeper and more binary, and recovery depends on removing the thing that tripped the policy, not on general polishing.

If your traffic fell between August 18 and 21, you are dealing with a spam classification. That changes what you should fix and how long you should expect to wait.

What Google says it is targeting

Google's spam policies define scaled content abuse as “when many pages are generated for the primary purpose of manipulating search rankings and not helping users.” The policy lists examples, and several describe the sites that lost in August almost word for word:

• Using generative AI tools or similar tools to generate many pages without adding value for users.
• Scraping feeds, search results or other content to generate many pages.
• Stitching or combining content from different pages without adding value.
• Creating many pages that contain search keywords but make little or no sense to a reader.

The policy is explicit that it applies however the content is produced: by people, by automation, or by both. That is the most misunderstood point about AI content and search. Google does not penalize a page because AI helped write it. Its guidance on generative AI says appropriate use of AI is fine, and using it to produce content mainly to manipulate rankings is a violation.

So the question Google's systems ask is not “was this written by a model?” It is “why does this page exist, and would anyone miss it?” AI makes it cheap to publish thousands of pages that fail that test, so AI-heavy sites are where the damage concentrated.

The who, how and why test

Google's helpful content guidance gives a simple lens that works well as an audit. Who created this, and is that obvious to the reader? How was it created, and would a reader expect to be told if automation did most of the work? Why was it created: to help someone, or to catch search traffic? Pages that fail all three are the ones spam updates are built to find.

The sites that lost: what we saw

We looked at the sites that came to us after the update and at others we track. Glenn Gabe also published a set of case studies from the rollout. The same patterns came up again and again. To be clear about sourcing: the policy language above is Google's. The patterns below are our observations and those of other practitioners. Google has not confirmed them site by site.

1. AI content published at volume with no human edit

This was the largest group by a distance. The typical profile was a site that went from a few dozen posts to several hundred or several thousand within months. The pages were plainly generated from a keyword list, and nobody had read them before they went live. The tells were consistent:

• The same article skeleton on every page: definition, “why it matters,” five generic tips, a conclusion restating the intro.
• No specifics. No prices, no dates, no named tools, no numbers the writer could only know from doing the work.
• Stock phrasing (“in today's fast-paced world,” “it's important to note,” “navigating the landscape”) repeated across hundreds of URLs.
• Facts that were slightly wrong or out of date, because nobody checked them.

2. Programmatic pages with a keyword swapped in

“Best [service] in [city]” and “[tool] alternatives” pages, generated from a template with only the variable changed. Some sites had thousands. Where each page carried real, local, unique data, it generally held. Where the city name was the only thing that changed, it generally did not.

3. No author, or an invented one

Many of the affected sites had no bylines at all, or credited everything to “Admin” or “Staff Writer.” A worse version was a made-up expert persona with an AI-generated headshot, a bio full of claims nobody could verify and no presence anywhere else on the web. Google does not rank on bylines directly. But anonymity is a strong sign that nobody stands behind the content, and that correlates heavily with the pattern above.

4. Thin affiliate and roundup content

Review and “best of” pages that clearly had never touched the product: specs pulled from manufacturer pages, the same pros and cons as every competitor, and affiliate links doing the real work. Gabe's case studies flagged this category too.

5. Sudden publishing spikes

A common thread even among sites with better-looking content: output jumped from a few posts a month to dozens a day. That kind of spike is exactly the shape scaled content abuse describes, and it made otherwise borderline sites easy to classify.

What held or gained

The sites that came through well were not necessarily AI-free. Many use AI in drafting. What they had in common was that each page clearly had a reason to exist. It carried the writer's own experience, named a real and verifiable author, cited primary sources, included details only a practitioner would know, and was published at a pace a real team could plausibly edit.

Humanize the content, not the detector score

“Humanizing” has become shorthand for running AI text through a paraphraser until a detector stops flagging it. That misses the point entirely. Google is not running an AI detector on your page. It is asking whether the page is useful and original. A reworded generic article is still a generic article.

Real humanizing is an editing process. It is the step that was missing from almost every site that lost in August:

• Start from what you know, not from a prompt. Give the model your notes, your data, the customer questions you actually hear, and the mistakes you have seen. Let it structure; don't let it invent.
• Add what only you can add. A real example, a real number, a photo you took, an opinion you would defend. If you can't add anything, the page probably shouldn't exist.
• Cut the filler. Delete the definition paragraph everyone already has, the “why it matters” section that says nothing, and the conclusion that restates the intro.
• Check every fact. Models are confidently wrong about dates, prices, laws and product details. One wrong fact undermines the rest of the page.
• Read it aloud. If it doesn't sound like a person in your business talking to a customer, rewrite it until it does.
• Publish at a pace you can edit. Ten pages a month that someone genuinely reviewed beat three hundred that nobody did.

Our free AI content detector is useful as an editing aid, not a pass or fail gate. It highlights the passages that read as uniform, generic machine text, which is usually where a human editor has the most to add.

Attribute content to a real expert

Experience and expertise are the parts of E-E-A-T that AI can't fake. That makes clear authorship one of the most useful signals you control. Done properly, it looks like this:

• A real named person on every article, ideally someone who does the work the page describes. For a service business that is usually the owner, a lead technician or a specialist on staff.
• An author bio with verifiable detail: role, years in the field, credentials or licences, and what they have actually done. Link it to a proper author or about page.
• Author markup that matches the page. Google's Article structured data guidance recommends marking up the author as a Person with a name and a URL that identifies them. It should agree with the visible byline.
• An expert reviewer for health, legal and financial content. If the writer isn't the expert, say who reviewed it and when.
• No invented personas. A fake expert is worse than no byline. It is a deception, and the pattern of unverifiable authors across AI-heavy sites is easy to see at scale.

Authorship also matters beyond Google. AI assistants weigh whether a source looks like a real organization with real people behind it when choosing what to cite, and a named, consistent expert across your site, profiles and third-party mentions helps them connect the dots.

Cite real sources and research

The sites that lost rarely cited anything. The ones that held usually did. Citing sources does not fool an algorithm. It is evidence that the writer did the work, and it gives readers a way to check the claims.

• Link to primary sources. Google's own documentation rather than a blog summarizing it; the study rather than a news story about the study; the regulator rather than a competitor's explainer.
• Put numbers next to their source and date. “16.71% of top-10 URLs dropped out of the top 100 (SE Ranking, August 2026)” is useful. “Many sites lost traffic” is not.
• Add your own data where you have it. Anonymized client results, survey responses, job costs, response times. First-party data is the one thing nobody else can copy.
• Keep a sources list at the end so readers and AI systems can see at a glance what the page is built on.

This carries over to AI search too. The Generative Engine Optimization study from Princeton and collaborators found that adding citations, quotations and statistics to content was among the most effective ways to increase how often it was surfaced in generative engine answers, with gains of up to around 40% in their tests. The work that protects you from spam classification also makes you more citable.

If you were hit: a recovery plan

Recovery from a spam update is possible, but it is slow and it requires removing the problem rather than decorating it. As Glenn Gabe put it in his August analysis: “You can recover from spam updates. You just need to address the spam… and then Google will need to see the improvements over time (typically at least several months).”

1. Confirm the timing

In Search Console, compare clicks and impressions for the week before August 18 with the week after August 21. A sharp step down inside that window points at the spam update. A gradual slide, or a drop in late March or late May, points somewhere else.

2. Check for a manual action

Open Security & Manual Actions in Search Console. Algorithmic spam demotions do not show up there, but a manual action for scaled content or thin content sometimes lands around the same time and needs a reconsideration request.

3. Find the page groups that fell

Export page-level data and group by template or folder: blog, location pages, comparison pages, glossary. Usually one or two groups account for most of the loss, and they are usually the groups that were produced fastest.

4. Improve, consolidate or remove

• Improve pages that target a real need and can be made genuinely useful with first-hand detail, a real author and sources.
• Consolidate near-duplicates into one strong page and 301 the rest.
• Remove or noindex pages that exist only to catch a keyword. This is the step most people resist and the one that matters most. A site with a large mass of low-value pages can drag down its good pages too.

5. Change the process, not just the pages

If the workflow that produced the problem is still running, the next update will find it again. Set a publishing pace your team can properly edit, require a named author and sources for every piece, and make human review a gate rather than an optional step.

6. Be patient

Improvements are typically recognized over months, often around a later spam or core update. Keep tracking the affected page groups rather than whole-site traffic so you can see recovery as it starts.

Frequently asked questions

Was August 2026 a core update or a spam update?

A spam update. Google's August 2026 spam update ran from August 18 to August 21, 2026. Google's 2026 core updates were in March and May. Many articles call the August update a core update, but Google's status dashboard lists it as a spam update.

Does Google penalize AI-generated content?

Not for being AI-generated. Google's guidance says appropriate use of AI is not against its policies. What it targets is scaled content abuse: many pages produced mainly to manipulate rankings without adding value, whether written by people, AI or both. In practice, sites that mass-published unedited AI content fit that definition, which is why they were hit hardest.

Will rewriting AI content with a humanizer tool help recovery?

Rarely. Paraphrasing changes the wording, not the value. Google's systems are judging whether a page is original and useful, not whether it trips an AI detector. Recovery comes from adding real experience, data and sources, consolidating duplicates, and removing pages that exist only to rank.

How long does it take to recover from a spam update?

Usually months. After you remove or improve the offending content, Google needs to recrawl and reassess the site over time. Many sites see movement around a later spam or core update rather than immediately.

Do author bios actually affect rankings?

Not as a direct ranking factor, but they matter. Google's guidance asks you to make clear who created the content, and real, verifiable authorship is a strong sign of genuine expertise. Anonymous or invented authors were common among the sites that lost in August.