AI Visibility SOPs & Multi-Agent Orchestration (Full Setup) — Transcript

Full transcript

1. 0:00If you're an agency, consultant, or
2. 0:02small marketing team, and you want to
3. 0:04deploy agents to do AI visibility work,
4. 0:07but you need processes, SOPs, and skills
5. 0:11along with the technology to set up
6. 0:13multi-agent orchestration, then this
7. 0:16video is for you. In this video, I'll
8. 0:18give you a tour of how I've created
9. 0:20hundreds of skills for AI visibility and
10. 0:23SEO work. I've connected them all
11. 0:25together, and I put all of that into an
12. 0:27MCP, which then acts as a knowledge
13. 0:29service for Claude code. I then use all
14. 0:32that knowledge to do real work, and then
15. 0:34I built a multi-agent orchestration
16. 0:37setup. I set up an agentic team and have
17. 0:39each one work on separate things
18. 0:41simultaneously. I'll even show you how
19. 0:44AI visibility optimization comes into
20. 0:46play, looking at retrieval optimization
21. 0:49and prompt tracking to figure out
22. 0:51specific fan-out queries that we can
23. 0:53target. So let's jump in. This is
24. 0:55basically a massive shortcut. I've been
25. 0:57doing SEO for 14 years, and I'm building
26. 1:00all of my knowledge and skills in here.
27. 1:02So you might have one agent that does
28. 1:03content refreshes. You might have
29. 1:05another agent that does content
30. 1:07planning. You might have another agent
31. 1:09that writes content specifically for
32. 1:11retrieval, so on and so forth. And
33. 1:13lately, I'm adding the AI visibility
34. 1:15layer where I'm doing research to find
35. 1:18what works and turning those into skills
36. 1:20and knowledge that feeds into the MCP as
37. 1:23well. For example, I applied my content
38. 1:25refresh process here, and I got 10 times
39. 1:28more citations on a project site. So the
40. 1:30MCP is part of the omnipresence platform
41. 1:33that I'm building. It also comes with
42. 1:35prompt tracking and retrieval tools. If
43. 1:37you want to try the MCP out for
44. 1:39yourself, it starts at $29 a month if
45. 1:42you're paying monthly. And there's a
46. 1:447-day no-card free trial. Once you get
47. 1:47an account, you're going to go into the
48. 1:49MCP page here, and you just going to
49. 1:51copy this URL,
50. 1:52api.getomnipresence.com/mcp.
51. 1:56Copy that to your clipboard, and then
52. 1:58you're going to go into Claude whatever
53. 1:59AI you use, and go into connectors, and
54. 2:03you're going to select add custom
55. 2:04connector, call it Omni, and just paste
56. 2:07in that URL there. Once you have it
57. 2:09connected, it'll look like this, and you
58. 2:10can set the permissions to always allow
59. 2:13for all of these different tools that it
60. 2:15comes with. Then you want to hop into a
61. 2:16new session in Claude Code. I highly
62. 2:19recommend using Claude Code because it
63. 2:21excels at multi-agent orchestration.
64. 2:24Then I'm going to give it a workspace to
65. 2:26operate in. So I just created a new one
66. 2:27here called Omni Presence Demo, and I'm
67. 2:30going to select trust this workspace,
68. 2:32and then I'll say connect to the Omni
69. 2:34Presence MCP. We'll just confirm that
70. 2:36it's hooked up and working.
71. 2:38Now it's found my projects, and I'm
72. 2:40going to hit submit on my Omni Presence
73. 2:43project. In the app itself, you can have
74. 2:45multiple projects, and the MCP can jump
75. 2:48between them. So if you're an agency or
76. 2:50a consultant, you can have one project
77. 2:52for every client or brand that you work
78. 2:54with. Now that you're connected, you can
79. 2:56explore the MCP. The best way to do that
80. 2:58is just by asking questions. So I asked,
81. 3:01does the MCP come with lots of processes
82. 3:03and systems to automate the work? It
83. 3:05says, yep, it's not just reference
84. 3:07material. It comes with processes,
85. 3:09reasoning, methodologies, skills, and an
86. 3:12agent layer of workers, systems, and
87. 3:14scripts that can run the work on a
88. 3:15schedule.
89. 3:17And here's a bunch of things that are in
90. 3:18it. So once you're in and you're ready
91. 3:20to try it out for some real work, you
92. 3:22just want to go to the MCP page and copy
93. 3:24this prompt here, set up my
94. 3:25orchestration. This is going to give you
95. 3:27the team of specialists that all work on
96. 3:30different tasks. For example, we might
97. 3:32have a content writer that only writes
98. 3:34content. So we'll have one orchestrator
99. 3:36that is kind of the boss of the other
100. 3:38AI's, and then we'll have a bunch of
101. 3:39specialists like content writers and
102. 3:42internal linkers and all that fun stuff.
103. 3:44So all you're going to do to set that up
104. 3:45is just say set up my orchestration and
105. 3:48paste that into your Claude chat. When
106. 3:50you set up your agent orchestration, you
107. 3:52have the option to run it on one
108. 3:54project, meaning it's one orchestrator
109. 3:56that 's going to work with other session
110. 3:58agents specifically for that project, or
111. 4:00you can even set it up to on several
112. 4:02projects simultaneously, where you
113. 4:04basically have a super orchestrator that
114. 4:06talks to other orchestrators. And then
115. 4:08those other orchestrators are talking to
116. 4:10their own individual specialists for
117. 4:12each project. We're going to go with one
118. 4:14project for now. You can choose your
119. 4:15interface. We're using the cloud desktop
120. 4:18app in this example, so we'll go with
121. 4:19that. Hidden means that they run in the
122. 4:21background and you don't even see them
123. 4:23working. Visible means you can see their
124. 4:25entire chat history and you can talk to
125. 4:27individual session agents as well. So
126. 4:29we'll go with visible. Then you want to
127. 4:31pick a location on your computer. I'm
128. 4:33going to go with / orchestration in my
129. 4:36demo project here. And now it's done its
130. 4:38initial setup. It checked for the
131. 4:40software it needs, such as Node. If you
132. 4:43didn't have this on your computer, it
133. 4:44would install it prior to. It loaded up
134. 4:47all of its templates. It set up some
135. 4:49schedules and systems to help the
136. 4:51orchestrator run sessions autonomously.
137. 4:54It writes some essential files like the
138. 4:57status and the queue. It's automatically
139. 4:59set to wake up every 3 hours and check
140. 5:02in on things. And it's given me some
141. 5:04open to-dos for things that it'll need
142. 5:06access to. And then it's asking to
143. 5:08install some tools. So we'll go with
144. 5:11install the tools. And it's basically
145. 5:12getting all this stuff straight from the
146. 5:14MCP.
147. 5:16It even has a bunch of script examples
148. 5:18which help with automation. And the way
149. 5:20that works is they're in here purely as
150. 5:22an example and they're copied and
151. 5:24customized when you install it. So
152. 5:27you're not installing suspicious or
153. 5:28strange scripts when you run this. It
154. 5:31basically just gives maintained examples
155. 5:33so that it can be set up on your end
156. 5:35without any security risks. So now it's
157. 5:37set up the tools. It's got a watchdog
158. 5:40service which revives dead sessions.
159. 5:42It's got a session opener which can
160. 5:44create its own sessions as needed. This
161. 5:46is useful if it suddenly needs a writer
162. 5:49and one doesn't exist. A secret guard
163. 5:51ensures that you don't send API keys or
164. 5:54anything like that out into the public.
165. 5:56A delegation guard ensures your
166. 5:58orchestrator is actually doing the
167. 6:00delegating and not doing hands-on work.
168. 6:02A doctor and a Codex wrapper, which is
169. 6:04optional if you have a Codex
170. 6:06subscription, you can actually have
171. 6:07Claude code delegate some of the coding
172. 6:10tasks to Codex, which saves on your
173. 6:12Claude subscription and increases your
174. 6:15Codex usage, which is something that I
175. 6:17need to do more often. It didn't set up
176. 6:19some of the optional ones like an alarm
177. 6:21system, which basically alerts you if
178. 6:23stuff goes down, and it didn't set up an
179. 6:25away mode, which is basically where you
180. 6:27can set up a Telegram connection so you
181. 6:28can talk to your orchestrator over
182. 6:30Telegram when you're away from your
183. 6:32computer. And now it's recommending to
184. 6:34create a new session which becomes the
185. 6:35main orchestrator. So we'll just do
186. 6:37that. Now we've got our orchestrator
187. 6:39session pretty much fully set up, and
188. 6:41then we'll give it some stuff to work
189. 6:42on. Let's set up specialist agents. What
190. 6:45do you recommend? So you don't even need
191. 6:47to know what kind of specialist agents
192. 6:49you want.
193. 6:50This will actually recommend a couple to
194. 6:52get started with. So now I'm going to
195. 6:53say give me a content planner and a
196. 6:55content refresher. Both can just create
197. 6:57CSV and markdown files for now.
198. 7:00Depending on the levels of autonomy that
199. 7:02you're comfortable with, you could have
200. 7:03them creating Google Sheets or Docs if
201. 7:06you have it hooked up. You can have them
202. 7:07produce Google Docs on your drive, or
203. 7:10you could even have them connect to
204. 7:12websites and create and upload content
205. 7:14and save them as drafts. You just have
206. 7:17to give them the credentials to be able
207. 7:19to do that. So now it's gone ahead and
208. 7:20set those up. So here's my two
209. 7:23individual sub agents, and I can see
210. 7:25what they're working on and talk to them
211. 7:27individually if I want to. We got one
212. 7:29for a content planner and one for a
213. 7:32content refresher. The best way to think
214. 7:34about this is that these are like little
215. 7:36employees. One of them specializes in
216. 7:39refreshing content, the other one
217. 7:40specializes in planning content. This is
218. 7:43beneficial because then all of their
219. 7:44context and skills are specific to those
220. 7:47tasks and they'll never get their wires
221. 7:49crossed. It also already gave them their
222. 7:52first assignment. The content planner is
223. 7:54going to come up with a couple topics
224. 7:56for us, and the content refresher is
225. 7:58going to propose a couple pages to
226. 8:00refresh. While these are working, I'm
227. 8:01going to talk to my orchestrator and
228. 8:03give it some access to some third party
229. 8:05tools. I've got an API key for Ahrefs
230. 8:08right here, and I've got a Data for SEO
231. 8:11API key right here as well. So, I'm
232. 8:13giving it the file locations instead of
233. 8:15pasting those into the chat because it's
234. 8:17more secure, and this allows my agents
235. 8:19to have tools to work with. And you can
236. 8:21also see the agents are talking to each
237. 8:23other. You can see that they're sending
238. 8:24messages back and forth, and you can
239. 8:27kind of poke in and see what they're
240. 8:28talking about. The orchestrator is
241. 8:30relaying what it's seeing from the sub
242. 8:32agents, but if I want, I can go into the
243. 8:35content refresher sub agent and go talk
244. 8:37to this one individually. And you can
245. 8:39see it's receiving messages from the
246. 8:41orchestrator, and it's just working
247. 8:43autonomously. Now, I'm going to give it
248. 8:44some direction. I'm going to say for the
249. 8:47content planner, propose a content plan
250. 8:49following the Omni content planning
251. 8:51process. That will follow the pre-built
252. 8:53SOP that lives in the Omni Presence MCP,
253. 8:57and you'll get to see what that looks
254. 8:58like in a minute. And then for the
255. 9:00refresher, just pick a random page on my
256. 9:03website and give me the refreshed
257. 9:05markdown file. These are just some
258. 9:06simple tasks so we can see the output
259. 9:09that it creates following the built-in
260. 9:11SOPs without you having to build and
261. 9:14dictate all of these instructions
262. 9:16yourself. So, while those two are
263. 9:17working on their task, we can continue
264. 9:20multitasking and delegating. I'm going
265. 9:22to say let's get another agent going, an
266. 9:24AI visibility optimizer. It's going to
267. 9:26pull fan outs for a specific page, and
268. 9:29it follows Omni processes to optimize
269. 9:31for them, and we'll use Athena HQ
270. 9:34alternatives as a first task. What
271. 9:36that's going to do is pull data out of
272. 9:38the prompt tracking portion of this app.
273. 9:40It's then going to find this prompt that
274. 9:42I'm tracking, which is what are the best
275. 9:44Athena HQ alternatives. If I click on
276. 9:46that, I can see that I'm not getting
277. 9:48cited right now. I can check out Claude,
278. 9:50ChatGPT, and Gemini, and Perplexity and
279. 9:53look at their responses. And the app
280. 9:55also tracks the sources that were cited
281. 9:58and the exact fan outs that were used
282. 10:00for each platform. So, I can drill into
283. 10:02these specific fan outs here. Here are
284. 10:04all the individual fan outs that are
285. 10:06being used by these engines, and I can
286. 10:09even drill into the first fan outs that
287. 10:11were used for each one of them. This is
288. 10:13important because the first fan outs are
289. 10:15generally more research focused and then
290. 10:18later on in the fan out process you'll
291. 10:20see that they start doing site searches
292. 10:22and getting more specific. A site search
293. 10:24is not really something you can optimize
294. 10:26for. So if I drill down into the earlier
295. 10:28fan outs I'll get better ideas of what I
296. 10:30can optimize for. Something I'm noticing
297. 10:32is chat GPT was digging into something
298. 10:35called the executive assistant service
299. 10:38and my page probably doesn't talk about
300. 10:40that. So right there that is an
301. 10:41opportunity of something I can optimize
302. 10:43for. You can also check what position
303. 10:45you're showing up for in these answers,
304. 10:48how often each answer was used and how
305. 10:50many prompts you're tracking that these
306. 10:52show up in. I can also drill into the
307. 10:54sources to see what websites are getting
308. 10:56cited the most often for this specific
309. 10:59prompt. I can see their domains, which
310. 11:01platforms they're being mentioned on,
311. 11:03how often they're cited and then I can
312. 11:06hop into the retrieval view to see how
313. 11:08well optimized I am for that page. So
314. 11:10here's my page on Athena HQ alternatives
315. 11:13and it shows me which fan outs I'm being
316. 11:15cited for and which part of the page is
317. 11:18getting cited. So for example for the
318. 11:20Athena HQ alternatives, which is the
319. 11:22main one, I'm getting cited in my little
320. 11:25intro description there and if I click
321. 11:27on this fan out it'll show me the
322. 11:29specific area of the page most relevant
323. 11:31for that fan out and my individual score
324. 11:33for it. So I can see that I'm pretty
325. 11:35well optimized for the main query there,
326. 11:38but for something like this I am less
327. 11:40optimized. So that gives me a bit of
328. 11:42ammunition for something I can do on my
329. 11:45page to improve my visibility for
330. 11:47specific fan outs and basically just
331. 11:49collect more data over time, which gives
332. 11:51you more fuel for optimization. All
333. 11:53right, so here's my deliverables. We've
334. 11:56got the content plan and I'm just
335. 11:58viewing them in Claude here. So for the
336. 12:01content plan, we have all of these
337. 12:03columns, the URL, the folder, the
338. 12:05content type, status, priority, click
339. 12:08potential, estimate, CPC, tons of data.
340. 12:10So it looks like it correctly followed
341. 12:12the content planning process and it
342. 12:15works as a tracker for when you are
343. 12:17publishing and creating these pages. And
344. 12:20here is my refreshed page, Semrush AI
345. 12:22Visibility Toolkit Alternatives. It's
346. 12:25rewritten the content, so it's taking
347. 12:27the existing content and following my
348. 12:29refresh process, which I've proven on
349. 12:31multiple sites to significantly increase
350. 12:34AI visibility. It basically optimizes
351. 12:36your existing content by improving how
352. 12:38the content is chunked for query
353. 12:40relevance, and then it finds some new
354. 12:43stuff to add to the page to increase the
355. 12:45amount of queries that it can show up
356. 12:47for. And then here's my AI fan out
357. 12:49optimization work for the Athena HQ
358. 12:52Alternatives. You can see that it's
359. 12:54scoring the existing chunks, and it's
360. 12:56given me a step-by-step plan for how to
361. 12:58improve this page. It's even given the
362. 13:01paste ready passages, so I can literally
363. 13:03copy and paste this onto a page, or I
364. 13:05could give this to an agent that has
365. 13:07access to the page to then go and do the
366. 13:09work. So that's how you can set up
367. 13:11multi-agent orchestration and have a
368. 13:13bunch of agents working simultaneously
369. 13:16to do AI search work. And you don't
370. 13:18specifically need my MCP. You can build
371. 13:21your own processes and skills. Claude
372. 13:24Code excels at multi-agent
373. 13:26orchestration, so you could really apply
374. 13:29a similar setup to whatever you're
375. 13:30working on. But basically the main
376. 13:32reason you'd go with my MCP
377. 13:34is because I'm constantly maintaining it
378. 13:37and updating it over time. I 'm
379. 13:38constantly doing new research, figuring
380. 13:41out what works for AI search, turning
381. 13:43those into skills, and putting them into
382. 13:46that MCP. So if you want to learn more
383. 13:48about the research that I'm working on,
384. 13:50those videos are coming up next.