The $50,000 Internal Linking Experiment: Links That Get Clicked Carry More Weight — Transcript

Full transcript

1. 0:00This guy spent $50,000 testing. Does it
2. 0:03matter for SEO if people actually click
3. 0:06links on a page? Can the position of the
4. 0:08link give it more or less weight? So,
5. 0:11would higher up links get more weight?
6. 0:14I'm going to share this guy's post on
7. 0:15Black Hat World. Then I'm going to share
8. 0:17what a Google patent says on this. It is
9. 0:20insane. This Google patent is insane.
10. 0:22This guy's tests are insane. The results
11. 0:25are insane. They confirm things that a
12. 0:27lot of people in SEO thought, but didn't
13. 0:29have direct evidence for. So, this
14. 0:31poster Digital Media Lab says, "I've
15. 0:34been in SEO for the last 9 years. Built
16. 0:36and sold multiple web properties with my
17. 0:39core focus being on content websites.
18. 0:41Ever since AI came out, I've been
19. 0:43dabbling in ways I could leverage its
20. 0:45content creation, analysis, and all
21. 0:47other features to my website's
22. 0:48portfolio. One thing I've realized early
23. 0:51on is that my internal linking process
24. 0:53was hella inefficient. So, I started
25. 0:55figuring out how I can improve it. I
26. 0:57started with a G sheets automation, then
27. 0:59moved to Python and ChatGPT API. The
28. 1:02more I invested and improved internal
29. 1:04linking, the more money I was making
30. 1:06through better ad RPMs and rankings. RPM
31. 1:08means revenue per mill, so the amount of
32. 1:11revenue you get for 1,000 views or
33. 1:13impressions. I've realized I wanted to
34. 1:15niche down to internal linking and build
35. 1:17a tool for it. But to build the best
36. 1:19tool on the market, I needed to base my
37. 1:21ideas on solid data findings. So, for
38. 1:23the last year I've been running so many
39. 1:25different internal linking experiments
40. 1:27on my portfolio of websites. I even went
41. 1:30as far and ran experiments from friends
42. 1:32and websites. In total, 15 websites were
43. 1:35involved. I spent over $50,000 testing
44. 1:39things out. A virtual assistant costs,
45. 1:41content costs, SEO specialist, software,
46. 1:44setting new websites, etc. Here are some
47. 1:47of the most important findings. Get
48. 1:49ready for this. I've seen up to 150%
49. 1:52increase in ad RPMs with improved
50. 1:55internal linking structure and higher
51. 1:57volume of internal links. 10 plus
52. 1:59internal links per page minimum. Across
53. 2:01my portfolio of websites, this resulted
54. 2:04in thousands of dollars for simply
55. 2:06adding more internal links and having
56. 2:08visitors spend more time on my websites.
57. 2:10This was super interesting. I found no
58. 2:13upper limit of internal links that you
59. 2:15can add to the body of the content. I
60. 2:17experienced no ranking decline for pages
61. 2:20and websites where I added over 20 plus
62. 2:22internal links. I started only having
63. 2:25three to four internal links, then got
64. 2:26that number up to 20 plus. My assumption
65. 2:29is that Google will never penalize you
66. 2:31or derank you because you have more
67. 2:33internal links. The more the merrier,
68. 2:35apparently. I still have to break 30
69. 2:37links on a single page, but I just don't
70. 2:40see that being a problem. So, this post
71. 2:42doesn't address page rank dilution,
72. 2:44which means the more links you have on a
73. 2:47page, the less authority each link gets
74. 2:50because the authority is being diluted
75. 2:52among links. But, there's this 2012
76. 2:55experiment from Dan Petrovic, where he
77. 2:57tested if having a backlink from a high
78. 2:59authority page still passes meaningful
79. 3:02ranking value when that page has 4,225
80. 3:06followed outbound links. And get this,
81. 3:09the linked test site jumped from
82. 3:11position 79 to number three within about
83. 3:15three weeks, which showed that even with
84. 3:17thousands of outgoing links, the
85. 3:19backlink still had a strong ranking
86. 3:21effect. The test is called the page rank
87. 3:23split experiment. And all of this flies
88. 3:26directly in the face of what people have
89. 3:28been saying with page rank dilution.
90. 3:30But, we're going to dig into why this
91. 3:31might be the case in a little bit when
92. 3:33we talk about Google's patent on this.
93. 3:35This Black Hat World post continues, and
94. 3:37you're going to love this one. No more
95. 3:39issues with crawled currently not
96. 3:41indexed. Ever since I've taken internal
97. 3:43linking more seriously, I stopped
98. 3:45experiencing this issue. All my content
99. 3:47gets indexed because it usually has at
100. 3:49least five plus links towards and from
101. 3:52that post to other content. I found five
102. 3:54links to be a minimum number of internal
103. 3:56links you need to add to ensure you are
104. 3:59not experiencing indexing issues. This
105. 4:01next part is maybe the most important.
106. 4:03The higher the internal link on a page,
107. 4:06the more value it has. Based on my
108. 4:08findings, internal links placed at the
109. 4:10top carried more value than the ones at
110. 4:13the bottom and influence ranking
111. 4:15significantly more. Internal link clicks
112. 4:18matter so much. Google puts a lot of
113. 4:20value on your internal links being
114. 4:22clicked. I've spent a lot of time
115. 4:24tracking clicks for each internal link
116. 4:26and saw a clear correlation with the
117. 4:29ranking improvement. Then in the
118. 4:30comments of this thread, this poster
119. 4:32answered more. The poster talks about
120. 4:34how the test was performed, said the
121. 4:36$50,000 was spent on all of the tests to
122. 4:39be carried out. There's a lot of costs
123. 4:41involved to carry out a successful test,
124. 4:43starting with people you have to hire to
125. 4:45do some of the work, VAs, write new
126. 4:48content, even though we leverage a lot
127. 4:49of AI here as well, tools, website
128. 4:51development, etc. This poster said
129. 4:54regarding opening links in the same or
130. 4:56new tab, if it matters. The poster said,
131. 4:59"I haven't noticed any ranking
132. 5:00difference between opening internal
133. 5:02links in the same or new tab. My
134. 5:04personal preference is to keep the same
135. 5:06tab." How about this? Does it matter if
136. 5:08links are on topically similar pages?
137. 5:11The poster said, "All my experiments
138. 5:13showed that internal linking is best
139. 5:15done by similarity. Putting restraints
140. 5:18based on category or a tag doesn't make
141. 5:20sense when you can build internal links
142. 5:22site wide and improve more content by
143. 5:24having a wider pool of pages to link to.
144. 5:26So, he means having internal links
145. 5:28should be chosen based on how
146. 5:29semantically or topically similar the
147. 5:32pages are, not based on whether they
148. 5:34share the same category or tag. And
149. 5:36then, the placement of the links. The
150. 5:38poster says, "When I said internal links
151. 5:40at the top carry more value, what I mean
152. 5:43to say is that they don't need to could
153. 5:45at the top of the article, like in the
154. 5:46introduction, but at the top relative to
155. 5:49the other internal links. I am always
156. 5:51numbering the internal links. The first
157. 5:53one that I place is number one, the
158. 5:55second one I place is number two on that
159. 5:57page, and so on. But number one can be
160. 5:59in the middle of the actual content if
161. 6:01there are no other internal links above
162. 6:03that one. So it is completely okay how
163. 6:05you do it. So that's what this person's
164. 6:07$50,000 tests showed about internal
165. 6:10links. But how about Google's own
166. 6:12patent? This patent shows they care
167. 6:15about links on sites being clicked. The
168. 6:17patent is called ranking documents based
169. 6:19on user behavior and or feature data.
170. 6:22And it explicitly describes collecting
171. 6:25actual user behavior around links. In
172. 6:27the patent, Google describes assigning
173. 6:29link weights based on the resulting
174. 6:31probability that a link will be
175. 6:33selected. This is a quote from the
176. 6:35patent. "Assigning a weight to the
177. 6:37particular link, where the weight
178. 6:39reflects the probability that the link
179. 6:41will be selected." So links that get
180. 6:43clicked carry more weights. But how is
181. 6:45Google going to get this data? You might
182. 6:47be thinking. The patent says, "The user
183. 6:50behavior data might be obtained from a
184. 6:52web browser or a browser assistant
185. 6:54associated with clients." Immediately
186. 6:56after this, it says, "The web browser or
187. 6:59browser assistant may record data
188. 7:01concerning the documents accessed by the
189. 7:03user and the links within the documents,
190. 7:05if any, the user selected." So they get
191. 7:08it from Chrome. They get the click data
192. 7:10from Chrome, just like they do with
193. 7:12their SERPs. Or maybe they get it from
194. 7:14Google Analytics. But it's very clear
195. 7:16that they care and they're trying to
196. 7:18measure whether links on a page are
197. 7:20clicked. So there's a concrete example
198. 7:23in this patent. If page W links to X, Y,
199. 7:26and Z, and users click W to X twice, and
200. 7:30W to Z once, those are recorded as
201. 7:33actual positive click instances. Then
202. 7:36these real clicks are used to train a
203. 7:38model predicting which links are more
204. 7:41likely to be clicked. So, in Google's
205. 7:43patented design, they can observe clicks
206. 7:45on links on ordinary third-party
207. 7:48websites, even if that site itself
208. 7:50wasn't being accessed with Chrome, had
209. 7:52no Google Analytics or special tracking
210. 7:54code, just because of this model. So,
211. 7:56the patent uses these observed clicks as
212. 7:59training data, then builds this model to
213. 8:02predict how likely a link is to be
214. 8:04clicked based on features like position,
215. 8:06anchor text, contacts, source page,
216. 8:09target page, and whether the links are
217. 8:11on the same domain. This is a quote from
218. 8:13the patent. "Model applying may assign
219. 8:16weights to links based on the dynamic
220. 8:18model generated by model generating. A
221. 8:21link's weight may reflect the
222. 8:22probability that the link will be
223. 8:24selected." So, really, it seems like
224. 8:26links that get clicks matter so much
225. 8:29more, even for pages that have thousands
226. 8:32of links. So, link juice actually isn't
227. 8:34spread evenly among links on a page. And
228. 8:37this would explain Dan Petrovic's 4,000
229. 8:39link experiment where a link still has
230. 8:41value. Maybe Dan clicked it using
231. 8:43Chrome. Maybe other people clicked it.
232. 8:45Maybe the position said that it was
233. 8:46valuable. Literally, in Dan Petrovic's
234. 8:49experiment, he says, "Is it worth
235. 8:51getting a link from an authoritative
236. 8:52domain and/or high-page rank page which
237. 8:55has thousands of outgoing links on it
238. 8:57already?" Yes. Back to this Google
239. 8:59patent, it's worth noting this one is
240. 9:01expired. And as a lot of people know,
241. 9:03just because Google has a patent doesn't
242. 9:05mean they're going to use it. But, we
243. 9:07know for a fact that Google cares a lot
244. 9:10about clicks with their Navboost system.
245. 9:12Google measures clicks in the search
246. 9:13engine results pages. Google measures
247. 9:16pogo-sticking, so clicking on a page,
248. 9:18then going back to the search engine
249. 9:20results pages because you're
250. 9:21dissatisfied with the page that you
251. 9:23clicked. Google measures all this stuff
252. 9:25to see if people like the results.
253. 9:27Yesterday's episode was about how Google
254. 9:29admits they don't understand the content
255. 9:31on a page. They fake understanding it by
256. 9:34using behavioral signals like clicks.
257. 9:36That's episode 1,162.
258. 9:39Google admits they fake understanding
259. 9:41your content. So, it stands to reason
260. 9:43Google would also value link clicks on a
261. 9:45page. So, whether it's an internal link
262. 9:48or it's an outbound link, it looks like
263. 9:50it matters whether or not it gets
264. 9:53clicked. That's going to improve the SEO
265. 9:56of the page that the link is on because
266. 9:58it shows people are interacting with
267. 9:59that content. Then the pages that are
268. 10:02linked to receive more authority if
269. 10:04their links are clicked. But hey, what
270. 10:06do you think? Have you seen this
271. 10:08yourself? Have you observed that it
272. 10:10matters more for SEO if your links are
273. 10:12clicked? Let me know in the comments. I
274. 10:14really like all this stuff. This is so
275. 10:16fun. And if you want to know how to find
276. 10:17keywords where people actually want to
277. 10:19click because they're looking to achieve
278. 10:21something and how to make content that
279. 10:23gets people to click, my SEO course
280. 10:26shows you that. The course is designed
281. 10:27around finding keywords where people are
282. 10:29looking to buy something, use something,
283. 10:31or make a discovery call. When they get
284. 10:33to the pages that address their search
285. 10:35intent, they click to get the thing that
286. 10:36they want. The course shows you exactly
287. 10:38how to make these pages, how to
288. 10:40structure your site for these pages, how
289. 10:42to find these keywords, how to do an SEO
290. 10:44technical audit, how to build links,
291. 10:46examples in every niche. Everything is
292. 10:48in this course. If you haven't checked
293. 10:50it out yet, check it out. You're going
294. 10:52to love it. You can get it at
295. 10:53compactkeywords.com.
296. 10:55That's everything that I got for you on
297. 10:58this episode of the show. This is
298. 11:01episode 1,163
299. 11:05of the podcast. 1,163
300. 11:08days in a row doing this show, going
301. 11:11strong. If you watched this episode on
302. 11:13YouTube, thank you so much for watching.
303. 11:16If you listened on Spotify or Apple
304. 11:18Podcasts, thank you so much for
305. 11:19listening. And I will talk to you again.
306. 11:22When am I When am I talking to you
307. 11:24again? I won't going to I'm kind of
308. 11:25forgetting. Oh my gosh, it's tomorrow. I
309. 11:27will talk to you again tomorrow. Bye
310. 11:31now.