Who the August 2026 spam update hit, and what you can do about it

Google's August spam update ran from the 18th to the 21st, after several weekends of ranking swings that looked like live testing. Spam updates usually go after links or site reputation abuse. Google told Search Engine Roundtable this one did neither. So what did it go after?

Nine days after it finished, the losers share one common trait: sites that published a lot of AI-written pages fast. If you publish AI-assisted pages, here is who lost, who won, why and what you can do going forward.

Who lost: the sites that bulk-published AI pages

One site with 169,000 pages was adding 500 to 1,000 a day. Its Google traffic went to zero. Its owner called the pace "gradual".

Guru99, which runs scaled affiliate listicles, halved again: 108,000 to 50,000 clicks a month. In 2021 it did 14 million.

The one that got the most attention was browserstack.com. It published more than 2,000 AI articles in a single day and lost around 70% of its traffic. It has 23,000 referring domains and 300,000 branded searches a month.

Who won: short pages that help someone buy

SaaS sites with short, bottom-of-funnel pages grew straight through the update. This matches every update since the helpful content system. Pages that exist to refer a buyer to something useful keep ranking, because Google and the AI answer engines still have to send that buyer somewhere.

The other winners were sites that had been wrongly flagged earlier. Several of those spiked back to where they were.

How Google decides: a provisional score, then your visitors

Google cannot tell whether a page was written by a model. What it can do is assign a provisional site quality score based on what the site looks like, rank the pages for a while, and then watch how people respond. If visitors bounce and leave quickly, the score drops and the pages go with it.

The conclusion is simple: AI-assisted content that people actually read persists, because the page dwell metrics confirm its value.

The one tell Google does not have to guess at

Text copied out of ChatGPT, Claude or Gemini often carries characters you cannot see: zero-width spaces, spaces that only look like spaces, tag characters, variation selectors. They survive every paste and end up in your HTML, where anyone parsing the page can count them. If you publish AI-assisted content at any scale, that is the cheapest possible signal to hand out. It is also the easiest one to stop.

If you got hit: this is what you should do

The best approach right now is initially to remove and 301 redirect AI published pages that do not add a lot of value for your readers. When you evaluate a page, think: "Would I read this?". If the answer is no, remove it. For pages that are ok, make sure that the page can be reached with a maximum of 3 links from your homepage.

If you want the update read against your own numbers, wire Search Console into Claude over MCP and ask it to list every page whose clicks fell by more than half between August 17 and August 22. That is your checklist.