Google's September 2026 Spam Update Is Here (And It Came With a Secret AI Spam Detector) — Transcript

Full transcript

1. 0:00Google's September 2026 spam algorithm
2. 0:03update has been unleashed and it is
3. 0:05registering a 9.4 out of 10 in intensity
4. 0:09on Seamrush's volatility tracker and it
5. 0:12is coinciding with this viral paper from
6. 0:15Google about an anti-PAM system called
7. 0:18safe. So, I'm going to share what we
8. 0:20know for sure, details on this safe
9. 0:23system from Google, community reactions
10. 0:25to the update and what to do if you see
11. 0:28volatility yourself in Google Search
12. 0:30Console. The update dropped a few days
13. 0:33ago on September 24th. This is what a
14. 0:35spam update is for people who don't
15. 0:37know. This is in Google's documentation.
16. 0:39While Google's automated systems to
17. 0:41detect search spam are constantly
18. 0:43operating, we occasionally make notable
19. 0:46improvements to how they work. When we
20. 0:48do, we refer to this as a spam update
21. 0:51and share when they happen on our list
22. 0:53of Google search ranking updates. This
23. 0:55particular spam update is the fourth of
24. 0:58the year, 1 month after the August spam
25. 1:00update, which was crazy. That update
26. 1:03annihilated so many AI sites. AI site
27. 1:06owner after AI site owner reported a
28. 1:09huge decline in organic traffic. And
29. 1:12everybody agreed that last month's
30. 1:14update was specifically targeting sites
31. 1:17that were scaling SEO content. Whether
32. 1:19they were using AI or programmatic SEO
33. 1:23done improperly, there is some unusual
34. 1:25behavior with this update that's worth
35. 1:27mentioning. Barry Schwarz, the founder
36. 1:29of Search Engine Roundtable, said this
37. 1:31to John Mueller. John Mueller is
38. 1:33Google's search advocate, who's
39. 1:35basically the face of Google search. He
40. 1:37gets his information directly from
41. 1:39Google engineers and answers questions
42. 1:41about Google's algorithms. Barry Schwarz
43. 1:44said, "Are we sure this should say two
44. 1:46weeks and not two days? The previous
45. 1:48three spam updates were all about 2
46. 1:51days. Core updates are 2 weeksish. Was
47. 1:54that a typo?" And John said, "Yes, it's
48. 1:56likely to take longer than some of the
49. 1:59previous ones. Some of the previous spam
50. 2:01updates." So core updates take around 2
51. 2:03weeks to roll out. Spam updates roll out
52. 2:06real fast. But this one is different.
53. 2:08And John Mueller said this is
54. 2:09intentional. So while this update just
55. 2:12started, we could really see continuing
56. 2:14volatility over the next couple of
57. 2:16weeks. I already saw volatility in two
58. 2:18very different sites, both of which are
59. 2:20not scaled content. So I think this is
60. 2:23really affecting a wide range of sites.
61. 2:25One of the sites is a bottomfunnel SAS
62. 2:27site. That one declined and then
63. 2:30rebounded back to where it was fast. And
64. 2:32when I looked at that, I said, "What is
65. 2:34happening? Oh no, this looks bad." ad
66. 2:36and then it just rebounded right away
67. 2:38because bottom ofunnel at bottom of
68. 2:39funnel is very stable and I've seen this
69. 2:41happen with different updates to this
70. 2:43bottom offunnel site a bunch of times.
71. 2:45An update will come, there will be
72. 2:47volatility, it'll rebound back to where
73. 2:49it was. Then I had a top offunnel blog
74. 2:52site that continues to experience
75. 2:54volatility. It had a decline at the same
76. 2:56time just a few days ago as the bottom
77. 2:59offunnel site. It rebounded back to
78. 3:01where it was. But unlike the bottom of
79. 3:04funnel site, this one continues to
80. 3:06experience volatility. And as I said at
81. 3:08the beginning, this update is
82. 3:10registering an insane 9.4 out of 10
83. 3:13spike on Seamrush's volatility tracker.
84. 3:16Seamrush says it's a Google quake. At
85. 3:19the same time this update drops, this
86. 3:21paper from Google about a new AI spam
87. 3:24detector called SAFE starts going viral.
88. 3:27This is an article now on search engine
89. 3:29journal from Roger Monty from September
90. 3:3125. Roger writes, "Google published
91. 3:34details of its new anti-PAM system
92. 3:36designed to detect AI content. This is
93. 3:39why it may be part of Google's spam
94. 3:41update. Google has published a research
95. 3:43paper about detecting spam that mimics a
96. 3:46human manual review that catches content
97. 3:48that violates the quote unquote spirit
98. 3:51of policy violations and platform
99. 3:53guidelines. The system is called scaled
100. 3:56abuse forensic examiner safe and it is
101. 3:59expressly designed to identify AI
102. 4:02generated content. It discloses that it
103. 4:04has a synthetic AI video and channel
104. 4:06abuse design but it also has a content
105. 4:09analysis component and uses multimodal
106. 4:12semantic embeddings. It references the
107. 4:14problem of nonhuman engagement patterns.
108. 4:17So the paper says the proliferation of
109. 4:19botn nets and coordinated adversarial
110. 4:22campaigns necessitates robust methods
111. 4:25for identifying nonhuman engagement
112. 4:27patterns. So detecting AI spam. The
113. 4:29entire paper leans towards video, but
114. 4:32the paper repeatedly uses phrases such
115. 4:35as synthetic media, synthetic content,
116. 4:38adversarial synthetic media, and
117. 4:40multimodal signals. The paper never
118. 4:43identifies safe as exclusively a video
119. 4:46detection system and that says something
120. 4:49because the paper discusses analysis of
121. 4:51generative content artifacts
122. 4:53particularly its use of multimodal
123. 4:55semantic embeddings which means that
124. 4:57this could apply to text. This could
125. 4:59apply to SEO content. This is Google's
126. 5:01second system identified in 2026 that is
127. 5:04designed to catch AI generated spam. The
128. 5:07previously identified system is called
129. 5:09scalable cluster termination system. And
130. 5:12actually, I made an episode about that,
131. 5:13episode 1,083 of the podcast. Google's
132. 5:16new AI spam detector could wipe out
133. 5:19entire SEO networks. If you want to see
134. 5:21how Google is designing systems to catch
135. 5:24AI spam or scaled content abuse, check
136. 5:28out that episode. One of the things that
137. 5:30you'll learn, for example, is that
138. 5:31Google isn't going to just look at a
139. 5:33single document or a single page. It's
140. 5:35going to look at all the pages on a
141. 5:37site. Then it might see how does this
142. 5:39site compare against other sites in
143. 5:41Google Search Console? How does this
144. 5:42site compare against sites with similar
145. 5:44patterns? Oh, maybe patterns can be
146. 5:46picked out that identifies sites as
147. 5:49using scaled content. And then they can
148. 5:51go through another layer of detecting
149. 5:53AI. So this article on search engine
150. 5:55journal continues. The fact that Google
151. 5:57is devoting resources to catching AI
152. 5:59slop shows that Google is concerned
153. 6:02about AI spam content. These systems may
154. 6:05be a component of the September spam
155. 6:07update. Now, it is important to note
156. 6:09that if you look at the document
157. 6:11properties on this safe paper that was
158. 6:14published by Google, the document
159. 6:16properties say created May 28th. So,
160. 6:18this system may have been in effect for
161. 6:21the June spam update and the August spam
162. 6:23update as well. The Search Engine
163. 6:25Journal paper continues, "AI enables
164. 6:27abusive networks to mass-produce
165. 6:29synthetic content while systematically
166. 6:31tweaking it to evade traditional
167. 6:33detection systems. Humans can find
168. 6:36coordinated spam networks by examining
169. 6:38relationships, behavior, content, and
170. 6:41even zoom out to examine infrastructure,
171. 6:43but manual inspections don't scale fast
172. 6:46enough to catch up with a massive scale
173. 6:49of AI generated slop." This is a direct
174. 6:51quote now from Google's paper.
175. 6:53Traditional forensic workflows which
176. 6:55rely heavily on manual pattern
177. 6:57recognition and metadata analysis are
178. 7:00illquipped to handle this volume. The
179. 7:02synthetic gap, the time between the
180. 7:05emergence of a new generative attack
181. 7:07vector and the deployment of a counter
182. 7:09measure remains a critical
183. 7:11vulnerability. And so this search engine
184. 7:13journal article says, "The goal of safe
185. 7:16is to catch content that may not match
186. 7:19an existing rule or known violation
187. 7:21pattern, but still violates the intent
188. 7:24of the policy or platform guideline and
189. 7:27can go undetected by traditional
190. 7:29classifiers and fine-tuned violation
191. 7:32detection models." And then there's this
192. 7:34section, the safe system has been
193. 7:35deployed. The research paper is very
194. 7:38secretive. It's only three pages long
195. 7:41and mentions having tested the system
196. 7:43but does not share the results of the
197. 7:45tests. That is highly unusual and points
198. 7:48to how Google is keeping the public in
199. 7:50the dark about safe. But it does share
200. 7:53that the system has been deployed. The
201. 7:55paper says early deployment results
202. 7:58indicate that safe significantly
203. 8:00accelerates the identification of novel
204. 8:02synthetic threats reducing forensic
205. 8:05investigation time compared to human in
206. 8:07the loop workflows. There are three
207. 8:10technical foundations of safe one. It
208. 8:12looks for coordinated behavior that is
209. 8:14different from normal human activity. It
210. 8:17analyzes patterns including timing. So,
211. 8:19bursts of activity, infrastructure,
212. 8:22hosting behavior, and other shared
213. 8:24signals, including fake user behavior
214. 8:27signals. Two, it uses specialized AI
215. 8:29agents to divide forensic work into
216. 8:32separate tasks with an orchestrator
217. 8:34agent, the root agent, that manages the
218. 8:36other agents and makes a final call. If
219. 8:39you ever designed your own systems using
220. 8:42AI agents, you can imagine this very
221. 8:44clearly. The agents are talking to each
222. 8:46other. The managing agent makes the
223. 8:48final call. And three, it analyzes the
224. 8:51meaning and context of content,
225. 8:53including multimodal analysis to
226. 8:56identify spirit of policy violations.
227. 8:59The takeaway from this is some in the
228. 9:01SEO community believe that Google is
229. 9:04using AI content detection to identify
230. 9:06spam. This research paper shows that
231. 9:08what Google is doing goes way beyond
232. 9:11that. Google is using systems that go
233. 9:14beyond simple AI content detection and
234. 9:16now has a system that goes beyond
235. 9:19traditional classifiers. SAFE behaves
236. 9:21like a human forensic investigative team
237. 9:24using specialized AI agents to analyze
238. 9:27content behavior infrastructure and
239. 9:30content producer relationships to
240. 9:33identify synthetic abuse networks. That
241. 9:36is from Roger Monty at Search Engine
242. 9:38Journal. If you want to go deeper on the
243. 9:40consequences of using AI for SEO, if
244. 9:43there are consequences and what might
245. 9:45they possibly be, or what would Google
246. 9:47even do if it is detecting AI in your
247. 9:50content? Episode 1,153
248. 9:53of this podcast, Google can detect your
249. 9:55AI content. Here's what happens next.
250. 9:58That will answer a lot of questions that
251. 10:00you have. We're going to finish this
252. 10:02episode with community reactions about
253. 10:04this September spam update and then what
254. 10:07to do if you are impacted. This first
255. 10:09one is from M. Um Khaled anx who said,
256. 10:13"Spam updates are the ones that move the
257. 10:16map pack. Fake listings and keyword
258. 10:18stuffed business names get swept out.
259. 10:20Shops that kept their Google business
260. 10:22profile name clean usually come out
261. 10:24ahead." This is from Jennifer on X.
262. 10:27Jennifer said, "Is Google going crazy?
263. 10:30Two spam updates in less than a month.
264. 10:32Then Jane boutique SEO on X said, "Here
265. 10:35we go again. Buckle up." Citizen Zoo on
266. 10:38X said, "Yeah, I expected this. This
267. 10:40year spam updates seem more frequent and
268. 10:43a lot faster to complete, but not this
269. 10:45one, cuz this one is taking 2 weeks to
270. 10:47roll out." Because of the timing with
271. 10:49this one, some people think that this
272. 10:51update is going back and catching spam
273. 10:53that didn't get caught in the August
274. 10:55spam update. Three more community
275. 10:57reactions. These are from the SEO
276. 10:59subreddit. Somebody said, "We can just
277. 11:01expect these to roll out every month
278. 11:03now." Another person said, "I got great
279. 11:05numbers yesterday and nothing today.
280. 11:07What is going on?" Somebody else said,
281. 11:09"If it's programmatic SEO, my guess is
282. 11:12it's sets where only the city swaps and
283. 11:14the rows are the same go first. Pages
284. 11:17with something computed on top, rank in
285. 11:20the city, changed since last month,
286. 11:21might hold up better." So this person is
287. 11:24theorizing that programmatic SEO pages
288. 11:27where very little is changed between the
289. 11:30pages, especially for you'll see people
290. 11:33targeting keywords like SEO agency New
291. 11:35York, SEO agency Philadelphia, SEO
292. 11:38agency Los Angeles, and the only thing
293. 11:40that's changed on the page is the city
294. 11:42name. So the person is suspecting that
295. 11:45those will get hit. that already goes
296. 11:46against Google's spam policies where you
297. 11:49don't change the meaningful content on
298. 11:51your pages while targeting different
299. 11:52keywords. And so this person is saying
300. 11:54if you actually are making decent
301. 11:56changes, you might hold up better. Now,
302. 11:58if you are hit, this is my
303. 11:59recommendation. If you are hit and you
304. 12:01see in Google Search Console, you see
305. 12:03crazy volatility. I recommend wait for
306. 12:05the roll out to complete before doing
307. 12:07anything drastic. Being reactive could
308. 12:10do you more harm than good. But always
309. 12:12always go over your most important
310. 12:15pages, the pages that bring conversions
311. 12:17to decide. Just ask yourself these two
312. 12:20questions. Is the relevance there. Do I
313. 12:22have the keyword in the page title? URL
314. 12:24slug H1 beginning of the first sentence.
315. 12:26And then am I answering search intent
316. 12:29right away? I recommend always doing
317. 12:31that. And if you are going after bottom
318. 12:33offunnel keywords, you're just going to
319. 12:35be a lot more stable. There might be
320. 12:36volatility, but every time this happens
321. 12:38to me, I rebound. I have a course on
322. 12:41doing bottom offunnel SEO. This is from
323. 12:43EPE Kanahara, the rank and rent master
324. 12:46who said this exact thing that bottom
325. 12:48offunnel SEO is more stable sites that I
326. 12:51built over 11 years ago and how steady
327. 12:53the rankings have been.
328. 12:54>> Dude, it's the bottom of the funnel.
329. 12:55It's the bottom of the funnel, man.
330. 12:57Bottom of the funnel doesn't die.
331. 12:59>> Exactly. Yeah. And I'm I'm I took your
332. 13:01compact keyword course as well. Great
333. 13:04program, especially for the price.
334. 13:06>> Amazing stuff.
335. 13:07>> Yeah. I'm a big believer in bottom the
336. 13:09funnel keywords as well. Um, it has it's
337. 13:11the sole reason I've survived with SEO
338. 13:14with all these different sites. Thank
339. 13:16you for that, EP. If you want to save
340. 13:17years learning how to do bottom offunnel
341. 13:19search engine optimization where you are
342. 13:21specifically finding keywords where
343. 13:23people are searching for what you offer,
344. 13:26these people are in heat. They are
345. 13:27looking to convert. They want to become
346. 13:29users. They want to become customers.
347. 13:31They want to call you up to see if you
348. 13:33can fit what they're looking for. My SEO
349. 13:36course compact keywords shows how to
350. 13:38find these keywords, how to target them
351. 13:39with conversion-based SEO landing pages,
352. 13:42not top offunnel blog posts. How to
353. 13:44structure your site for these pages, how
354. 13:46to build backlinks, how to do an SEO
355. 13:48technical audit, a content audit,
356. 13:50examples in every niche, Google Docs,
357. 13:52Google Sheets, so much is in this
358. 13:54course. Visit compactkeywords.com.
359. 13:56You will love it and we will be
360. 13:58monitoring more of the September spam
361. 14:01update in the days and weeks ahead.
362. 14:03That's everything that I got for you on
363. 14:05this episode of the show. This is
364. 14:07episode 1,180
365. 14:09of the podcast. 1,180
366. 14:12days in a row doing this podcast. Let's
367. 14:15go. If you watch this on YouTube, thank
368. 14:17you so much for watching. If you
369. 14:19listened on Spotify or Apple Podcasts,
370. 14:21thank you so much for listening and I
371. 14:23will talk to you again tomorrow. Bye
372. 14:26now.