Как создать ИИ маркетолога: парсит, разбирает конкурентов и создаёт мощный контент | ИИ Сотрудник — Transcript
Full transcript
- 0:00In my opinion, a marketer should do
- 0:01three things. First, they should
- 0:03collect data. Second, they should
- 0:05analyze this data. And third, they
- 0:07should create content based on this
- 0:08analysis, based on the data generated.
- 0:11In this video, I'll show you how I
- 0:13assembled my own marketer who collects
- 0:15data. For example, he analyzed or
- 0:17conducted marketing research in the
- 0:20electrical services niche in St.
- 0:22Petersburg and gave me this score,
- 0:24which completely describes what's
- 0:26happening in the electrical services
- 0:28market in St. Petersburg, meaning who
- 0:31provides what services, who the leaders
- 0:33are, and which companies. He analyzed
- 0:36not just a website, but also Avito,
- 0:38which seems impossible to analyze. He
- 0:41analyzed photos of Proruns and analyzed
- 0:44Yandex services. He analyzed over 4,000
- 0:47different ads and, accordingly, came to
- 0:51me with this report. The second thing
- 0:54he should do is, he's collected the
- 0:56data, he's completed the first part of
- 0:58the work, he can perform this part of
- 1:00the work based on a request we send him
- 1:03in a Telegram chat. All my agents work
- 1:05through Telegram. It's easier for me to
- 1:07interact with them this way. It's code,
- 1:10it's antigravity, it's code, it doesn't
- 1:12matter. So, all my agents work through
- 1:14Telegram, not through a terminal. It's
- 1:16inconvenient for me to interact with
- 1:18them through a terminal, for example,
- 1:21from a phone. Okay? So, he's done the
- 1:23research for me. So, he found the data
- 1:25I needed. The second thing he should do
- 1:28is conduct a competitor analysis. What
- 1:31did he do here? He came to me with this
- 1:34report as a competitor analysis. We
- 1:37took several YouTube videos and looked
- 1:40at what was in those YouTube videos.
- 1:42Why were these YouTube videos, or SHO
- 1:45videos, TikTok videos, Instagram ReS
- 1:47videos, why did they get so many views?
- 1:49So, he analyzed the hook, he looked at
- 1:52the structure, he looked at what drives
- 1:55audience retention, what strong wording
- 1:58was used in this particular video. We
- 2:01analyzed literally several videos. The
- 2:04more videos, the more, well, data for
- 2:06analysis you have, the better,
- 2:09accordingly, he will do this work.
- 2:11That's it. So, we now know what works
- 2:14in our niche, in this case, the
- 2:16electrical niche. And thirdly, we must
- 2:19now create content based on what works,
- 2:22based on the data we have. And the
- 2:25third task, the third task was
- 2:26precisely content creation. We asked
- 2:29our marketer to create content based on
- 2:33his analysis and, accordingly, on the
- 2:37data we have. In the end, he came to us
- 2:40with these three posts, because he was
- 2:43asked to create three posts; They might
- 2:45ask you to create 10 posts that you can
- 2:48simply copy and paste and publish
- 2:50somewhere on Instagram, Telegram, maybe
- 2:53VK, or any other convenient platform
- 2:56you like, even Max. We did this for
- 2:59text content. The same can be done
- 3:01using this as a basis: making videos
- 3:04out of this—not videos, but carousels
- 3:07, generating images. We can also create
- 3:11a script for short videos from this. So
- 3:14, you'll have a constant stream of
- 3:16content generation. And here I'll show
- 3:19you how to do this using simple queries
- 3:21. All of this, in theory, should work
- 3:24within the system. And I'll show you
- 3:26how to do it within the system. For
- 3:27example, you could have your agent do a
- 3:30market analysis on Mondays, do an
- 3:32analysis of your competitors 'market on
- 3:35Tuesdays, or maybe even on Wednesdays,
- 3:38and then bring you content for the
- 3:40following week on Fridays. You set all
- 3:43of this up in a scheduled posting, and
- 3:46it's all created in the scheduled
- 3:48posting in just a few minutes. So, the
- 3:51main idea of this video is to show you
- 3:53an AI marketer, to build an AI marketer
- 3:56together with you, who will collect
- 3:58data, analyze this data, and create
- 4:00content for you based on this data.
- 4:02Let's go. Okay. How do we build such a
- 4:03marketer? As always, we have two
- 4:05options. The first option is that you
- 4:08build this marketer together with me
- 4:10now, meaning I'll show you the tools
- 4:12we'll be using. The second option is
- 4:15that you download these seven files in
- 4:17the first pinned message—how many?
- 4:20That's seven files, which you download
- 4:23directly from this Telegram channel.
- 4:25And you either manually send these
- 4:28files to your agent working on CLDcode,
- 4:32MS, Codex, Antigravity, or, if you use
- 4:35an app like Codex, an app like clkD,
- 4:38well, the official app from Claude,
- 4:41then you simply send this entire folder
- 4:44with all this documentation here in the
- 4:47chat and ask him to build a similar
- 4:50system, a similar structure. It works
- 4:53in both places. It doesn't matter where
- 4:56you use your agents. Even if you use
- 4:58them through the terminal, you simply
- 5:00upload all these materials to the
- 5:02appropriate documentation or folder. If
- 5:05it's complicated, or if you don't
- 5:07understand how to do it or implement it
- 5:10somewhere, you immediately ask the
- 5:13agent what instructions you need to
- 5:15upload. It will tell you, you upload
- 5:17them, and that's it. After that, it
- 5:19implements everything the same way I do
- 5:21. That is, there are several files that
- 5:25you send it, and it uses these files,
- 5:28these instructions, they are verified,
- 5:31meaning they work in my agents, and we
- 5:34installed them on Codex, ID Code, and
- 5:36Antigravity. And they understand what
- 5:39these files are, how to use them, and
- 5:42how to implement them for everything to
- 5:45work. So, you're downloading from the
- 5:47first pinned message now. I'm now
- 5:48showing you how this structure works.
- 5:51Here's a step-by-step breakdown in this
- 5:53video. Okay, for our marketer to work
- 5:55properly, they need to perform three
- 5:57steps. The first step is data
- 5:58collection. How will we collect the
- 6:00data? Of course, we can train or
- 6:03instruct our agents to simply collect
- 6:06data, meaning they go online and search
- 6:09for specific data. I have a similar
- 6:12structure in place, but to be honest,
- 6:14it's not entirely ideal. It's not
- 6:17parsing, because most content creators,
- 6:20or most services, or most databases,
- 6:24especially Avito, Viberis, large
- 6:27resources where it's desirable to pull
- 6:31in specific data, impose restrictions.
- 6:35The restriction is that you have to
- 6:38register with Avito, or even Viber. And
- 6:42they impose endless restrictions
- 6:44everywhere, which most agents can't get
- 6:47around. If you need to find some
- 6:49superficial information, like news or
- 6:52news digests, then your marketers, oh,
- 6:55your agents, will handle it without a
- 6:58problem. But if you need to do any
- 7:01serious work, serious social media
- 7:03parsing, parsing, parsing, analyzing
- 7:06platforms like Avito, or more in-depth
- 7:09parsing, uh, based on some data, then
- 7:12it's best to do all of this through the
- 7:15Firecoll service. I have a separate
- 7:17video about how to connect Firecoll to
- 7:20your agent. I won't go into that now.
- 7:23The advantage of Firecoll is that it
- 7:27gives you about a thousand or fifteen
- 7:30hundred free tokens daily per month,
- 7:34which you can use to train your Agent
- 7:38to parse any platform. The advantage,
- 7:41perhaps, of Firecoll is that it's one
- 7:43of the services that bypasses all these
- 7:45platform restrictions. So, we'll ask it
- 7:49now. I already have Firecoll connected
- 7:51to my agent. This was done using one of
- 7:54these instructions that are written
- 7:57here. These are the parsing
- 7:59instructions, the market parsing
- 8:01instructions, the full instructions. I
- 8:03sent these instructions to my agent,
- 8:05and he installed everything himself, he
- 8:08connected himself, he set up the
- 8:11connection himself, yes, the free plan,
- 8:13about 1,000 requests a month, and he
- 8:16hasn't had any problems now. Let's, you
- 8:19know what we'll do? Let's do some
- 8:21parsing. That is, we'll ask him to
- 8:23parse an ad on Avito. And I most likely
- 8:27won't be able to access Avito myself
- 8:28because I have a VPN enabled right now
- 8:30and it will block me. That is, avita.ru
- 8:33will block me, but we'll ask...Yes, you
- 8:37have a VPN enabled. So, you see, Avito
- 8:40is imposing some restrictions. We'll
- 8:43ask him to parse. Oh, I don't know,
- 8:45let's ask some electricians in St.
- 8:47Petersburg, so he'll find an ad for
- 8:49those same electricians. Why do we need
- 8:51to parse electricians? We can choose
- 8:53any niche. And my main goal is to show
- 8:57you that for a tool, for a marketer who
- 9:00will be doing work every week, every
- 9:03day, or at our request, it can handle
- 9:06any parsing without a problem,
- 9:08including Avito parsing. You can give
- 9:11it complex tasks, or you can simply ask
- 9:13it to parse the market, because it will
- 9:16already have the information, skill,
- 9:18and experience on how to parse the
- 9:20market. That is, it has complete
- 9:22instructions on how it should parse the
- 9:25market for your specific situation. You
- 9:28can, of course, modify this prompt; you
- 9:31can ask it to modify it specifically
- 9:33for you, for your niche. I'll use its
- 9:35classic prompt. I log in to my agent,
- 9:38who already has access to Firecoll. The
- 9:41only thing I've changed here is the
- 9:42folder where it should save the report.
- 9:44So, let's send this request, and in the
- 9:46meantime, let's familiarize ourselves
- 9:48with the prompt I sent it. Conduct
- 9:50market research in the electrician
- 9:52services niche in St. Petersburg. Use
- 9:54Fcrawl to analyze websites, catalogs,
- 9:58and ads. Generate a detailed Markda
- 10:00report and save it to the Dump folder
- 10:02in Obsidian. Report structure: Key
- 10:04segments include minor repairs,
- 10:05electrical installation, turnkey
- 10:07electrical services, electrical panels,
- 10:09smart home, and commercial properties.
- 10:12Targets for each segment include
- 10:13minimum price and average price. This
- 10:15is, roughly speaking, a general report,
- 10:17a general analysis, not necessarily for
- 10:20Avito. So, if we asked it to do the
- 10:23same here, but for Avito, and after
- 10:26this analysis, we'd ask it to do this,
- 10:30we could also look at specific
- 10:32platforms to see what it can do. Right
- 10:36now, we're starting by asking it to
- 10:38create a full report, a global report,
- 10:40roughly speaking, covering the entire
- 10:43internet, but with regional
- 10:44implications. That is, electrician
- 10:46services in St. Petersburg. Now, in 5-7
- 10:49minutes, I'm pausing this video, in 5-7
- 10:53minutes he'll do this work, come back
- 10:56to us with a report, and we'll review
- 10:59it in detail. The agent hasn't
- 11:01responded to me yet; he's probably
- 11:02getting ready to write me something on
- 11:04Telegram, but I saw this file in the "
- 11:06dump" folder. Look, here it is. Dump,
- 11:08Electric, SPB Market Search. Let's take
- 11:12a look. So, you see, he immediately
- 11:14linked this note to the other notes. He
- 11:17wrote that he does this using FCW
- 11:20plusdel Avitonaliz. Oh, so he took the
- 11:23data from the Avito analysis from the
- 11:26previous note, that is, when I was
- 11:28analyzing Avito, uh, on June 4th, it
- 11:31turns out, on the same topic, he pulled
- 11:33these notes from there too. I honestly
- 11:36don't remember that I analyzed Avito
- 11:38anymore. So, he's saying there are
- 11:404,000 ads in St. Petersburg on Avito
- 11:42alone. And the structure of the players
- 11:45, that is, the multiservice Landremont,
- 11:48that is, he talks about who is in this
- 11:51niche, aggregators and an army of
- 11:53private sellers on Avito. There. That
- 11:56is, he has completely written it down,
- 11:58that is, he most likely also parsed
- 11:59Avito. He should indicate a list of
- 12:01sources for the main segments at the
- 12:03bottom, but if he hasn't, you can ask
- 12:06him to do so. So, look at what he did.
- 12:08Here's a report. The main segments:
- 12:10minor repairs, electrical installation,
- 12:12turnkey electrical work, control panels
- 12:14, competition, where is the lowest,
- 12:16average, and highest and, accordingly,
- 12:18the average bill. And who is working
- 12:20and, accordingly, where they look for
- 12:23clients for companies with permits.
- 12:25Minor repairs, the minimum price is
- 12:27200-350 rubles per operation. Socket
- 12:30and switch in a finished place.
- 12:32Callouts are free for most. Rough
- 12:34electrical installation. And, turnkey
- 12:37electrical work, apartment. Control
- 12:39panels, the minimum price for a simple
- 12:41control panel. Here is 1,300. Smart
- 12:43home, minimum price per element—that
- 12:46is, I've already analyzed the Proiru
- 12:48specialists and scenarios above. It's a
- 12:50good thing, by the way, that we didn't
- 12:52ask him to analyze Avito exclusively.
- 12:54He immediately took a large sample,
- 12:56including private contractors from
- 12:58Proiru, and I don't know what kind of
- 13:01keyword this is. Oh, what kind of
- 13:02website is this? Ah, some kind of
- 13:04specialized, maximum-price website,
- 13:06accordingly, here. He took the entire
- 13:09internet in St. Petersburg, and
- 13:12completely, ah, ads on sites like
- 13:15Proiru, Avito, and probably some other
- 13:18sites. Analysis of competitors'
- 13:20packaging. I found Lenremon
- 13:22Multiservice—by the way, it's also a
- 13:25service that allows you to find private
- 13:28specialists. The only thing is, I think
- 13:31they operate centrally. A specialized
- 13:33company, two electricians, that is,
- 13:35proi.ru, Yandex services, aggregators,
- 13:37a website, perfect packaging, high
- 13:39trust in the platform, price,
- 13:41accordingly, so much for the services
- 13:43participants. So, well, it's a pretty
- 13:45good report, yes. And everything we
- 13:48asked him to do, by the way, he wrote
- 13:52to me on Telegram. Ah, with 1988 lines.
- 13:57All the sections according to your
- 13:58structure, respectively, market
- 13:59such-and-such. So here he is, ah, a
- 14:02brief summary so we don't have to dive
- 14:06too deep, so we can see what he
- 14:09actually did. Top 10 ideas for future
- 14:11content based on the identified pain
- 14:13points. So, he immediately pulled out
- 14:15the pain points and wrote down ideas
- 14:18for our future content. If we're an
- 14:20electrician, what do we need, and how
- 14:23do we attract an audience, respectively
- 14:25. Four FireCroll queries, and
- 14:28competitor Avito scripts parsed from
- 14:32existing data. Prices are current as of
- 14:35July 2026. But listen, yes, our
- 14:37e-marketer's eyes, they didn't let him
- 14:39down, so he did an excellent job with
- 14:40the parsing. So now we have data on
- 14:43competitors. We can either enter this
- 14:46data manually or in the handouts found
- 14:50in the first pinned message. Here's a
- 14:53guide on how to train an agent to turn
- 14:56this into a rhythm. A good agent is one
- 14:59you assign a task to, and they complete
- 15:01it. An excellent agent is one you
- 15:03assign a task to once, and after that,
- 15:05they complete it every so often. Here's
- 15:08a guide on how you can turn a good
- 15:11agent into an excellent agent, so they
- 15:14can send you these kinds of reports
- 15:16every Monday, or reports on the areas
- 15:19you need, or use Chrome—maybe once a
- 15:22week, once a month, or something else,
- 15:25or at some other time. In other words,
- 15:28they can collect data for you without
- 15:30your request. They can simply ask you
- 15:32for it, or they can come to you with
- 15:35ready-made data and say, "Hey, I've
- 15:37done the work. Go ahead and check it
- 15:39and see what we can do with this data."
- 15:42And then we have to work with this data
- 15:44. It would be foolish to start turning
- 15:46this data into content now. We must
- 15:48first look at what type of content is
- 15:50popular, that is, what's popular in the
- 15:52market, what's already out there. That
- 15:54is, we shouldn't just continue to pull
- 15:57data, but rather look at what's
- 16:00happening in our market specifically.
- 16:03Basically, we need to help our agent
- 16:08with oversight, so they understand, not
- 16:14what data, but how to create content
- 16:19that goes viral. What do we do for this
- 16:22? The easiest way is to turn on TikTok,
- 16:26YouTube, Shorts, whatever—that is,
- 16:29any social media. And we look at what's
- 16:32already out there. What hooks are
- 16:35popular videos using, what locations
- 16:38are popular videos using, what type of
- 16:41content—reels, long videos—what the
- 16:46author says, why these videos are
- 16:48getting views. So, we need to analyze
- 16:51what's already out there. So, the
- 16:54second step is analysis. It's not an
- 16:56analysis of the data we've collected;
- 16:59he's already done the analysis—
- 17:02analysis to convey this data in a
- 17:05format that the audience, roughly
- 17:07speaking, will accept. So, now we need
- 17:11to find some viral videos on our topic.
- 17:14In this case, we've chosen electricians
- 17:16. So, we need to find viral videos on
- 17:18the topic of electricians to help the
- 17:21agent create content that's already
- 17:23resonated. See? I'll open my TikTok now
- 17:27, type in some keyword like "
- 17:29electricians," and I'll find a few
- 17:32videos and show you how I'll send these
- 17:35videos to my agents for them to analyze
- 17:38. Although, I think we could probably
- 17:40find something on YouTube, too. Let's
- 17:42go to YouTube and type in the word "
- 17:44electrician." Yes, here we're no longer
- 17:48limited to just electrician. So,
- 17:51electrician, or electrician. Here we're
- 17:54no longer limited to location. Here
- 17:57we're looking for everything that
- 17:59catches our eye and will adapt it to
- 18:01St. Petersburg. Let's see. 2.5 million
- 18:04views. Well, this is just about
- 18:06assembling a junction box. Okay, an
- 18:08electrician should know that. Uh, so we
- 18:11can look at, yeah, that format. This
- 18:13guy, by the way, I've met him a few
- 18:16times on YouTube, he comments on
- 18:18certain videos, but that's probably not
- 18:20of interest to us yet. Let's do this.
- 18:23It's him again. So, you see, this guy
- 18:25has flooded the entire internet. We
- 18:27need some kind of content, where...Well
- 18:31, we can take this guy. Let's take a
- 18:35look. Yeah, I'll take this video. It
- 18:37looks like someone is explaining
- 18:39something and I'm sending it to my
- 18:41agent. I can send it to this agent, but
- 18:44in my marketing department, for example
- 18:47, it's distributed among several agents
- 18:50. I have Antigravity. It's better at
- 18:53extracting the main meanings from
- 18:55videos. I'm sending him a link to this
- 18:58video and, accordingly, I'll wait for
- 19:01him to turn on, activate, and give me
- 19:03an in-depth analysis of this video. Yes
- 19:05, I'm still looking for a different
- 19:08type of content. That is, we can
- 19:10probably give you long-form content,
- 19:13because there are a lot of people on
- 19:16YouTube who make great content. For
- 19:19example, five years ago, ah, all the
- 19:21stages of electrical installation in an
- 19:23apartment, well, this is a very long
- 19:25video at 2.12 hours, I think. Of course
- 19:27, I should have prepared in advance.
- 19:29Well, let's also watch a twenty-minute
- 19:31video, we'll use that. Uh, how is an
- 19:35electrician's apartment set up? Well, I
- 19:37don't think this is an electrician.
- 19:38Okay, the agent has already sent me a
- 19:40transcript of one of the videos. Okay,
- 19:43let's move on. Electrical installation
- 19:46in an apartment can only be done after
- 19:47alignment. He also immediately added
- 19:50this video to my YouTube. Okay. And
- 19:53let's, well, I think two videos are
- 19:55quite enough. We go to the folder in
- 19:57Obsidian where all these materials are
- 20:00saved, that is, all these eyes. Here,
- 20:02ah, electrical, installation,
- 20:06everything's okay. The essence, the
- 20:09context, and, accordingly, the words "
- 20:10video transcription" are here. All
- 20:12these notes are saved exactly as I need
- 20:14them. Since electrical is not my niche.
- 20:17Let's try it. So, let's return to our
- 20:19agent. So, what do I do? Let's send the
- 20:22agent a task right away. So, what's
- 20:24going on here? I'm not asking it to
- 20:26analyze the video now, because
- 20:27antigravity is responsible for analysis
- 20:29, and it's precisely configured for
- 20:31that. I'm asking our marketer, based on
- 20:33the video analysis done by "Antigravity
- 20:36"(meaning he went out, pulled out all
- 20:38the transcripts, all the data, and
- 20:40created a brief structure), to analyze
- 20:43these videos as the chief marketer and,
- 20:45accordingly, extract the entire
- 20:47structure, all the ideas, all the hooks
- 20:50, everything that works, because these
- 20:52videos have garnered a large number of
- 20:55views. I sent him two videos. Two
- 20:57videos isn't much. After he does all
- 21:00this, I'll ask him to, with the data we
- 21:03received and the analysis that works,
- 21:06that is, we'll, roughly speaking, begin
- 21:10to combine all this for our future
- 21:13content. Let's wait until he does this
- 21:15analysis of the two videos. Of course,
- 21:17it's better not to rely on two videos
- 21:19alone. It's better to take 10, 15, 20,
- 21:2130 videos, for example, and I regularly
- 21:24upload all my content for the last two
- 21:26years to my marketers who create
- 21:28content for my Telegram channel, along
- 21:31with all my text content for the last
- 21:33year and a half to two years, so they
- 21:36understand how I write. For my agent,
- 21:39who creates the scripts for my videos,
- 21:42I upload all my statistics and data
- 21:45from YouTube, plus I transcribe them
- 21:48for the last two years so they can
- 21:50write the structure of my future videos
- 21:54for me. I can just show you; if you're
- 21:56interested, message me. And we can
- 21:58implement something similar there for
- 22:00anyone. So, I'll just show you the
- 22:03structure of my videos while he does
- 22:05the analysis. Here are the ideas, oh,
- 22:08ube videos from my channel. And here,
- 22:10accordingly, are some fresh ideas, no,
- 22:13development, here's the channel
- 22:15development strategy. And, accordingly,
- 22:18here's this thing where the analysis of
- 22:21all my videos didn't give you. And when
- 22:24it's not brainstorming ideas for my new
- 22:27videos, it focuses on these metrics. It
- 22:30calculates how many views each video
- 22:33gets, how long each video is, which
- 22:36networks, and why they perform better.
- 22:40And based on subsequent videos—it's
- 22:42constantly learning, and based on
- 22:44subsequent videos, it gives me better,
- 22:46better, best, roughly speaking, content
- 22:48. I had to switch agents because this
- 22:50one ran out of content. I'd asked them
- 22:52too many questions before filming the
- 22:55video. So, all my agents are
- 22:57interchangeable, which is convenient.
- 22:59So, there's an agent based on CLC code,
- 23:01an agent based on antigravity code, and
- 23:03the advantage of these agents is that
- 23:05they share a single database. So, they
- 23:08all have the same data. So, they can
- 23:11access obsidians without any problems.
- 23:15And I'm a VZA, meaning I replace it. So
- 23:17if something stops somewhere, my work
- 23:20will never stop because I have other
- 23:22agents that can pick it up. And all
- 23:25these agents have their own systems.
- 23:28For example, they have some things of
- 23:30their own, of course, but most of their
- 23:32solutions overlap. So, I have an agent
- 23:34architecture, and each one looks at the
- 23:37architecture of the other and picks up
- 23:39the skills it needs from there. It
- 23:42works really well. And thanks to this,
- 23:45I never have any downtime. So, if my
- 23:47agent stops, I can easily turn on
- 23:49another agent, and it continues the
- 23:51first one's work because the entire
- 23:53system is here. I have separate
- 23:55instructions for assembling such agents
- 23:58. And, accordingly, all the files for
- 24:00assembling and training these agents
- 24:02are in the first pinned message. I
- 24:04already showed you this. Good. So, this
- 24:06one picked up the work first and,
- 24:08accordingly, completed the task. So,
- 24:11look what it did. Ah, I analyzed both
- 24:13videos like a marketer. So, he analyzed
- 24:16both of these videos. Right now, for
- 24:18the sake of demonstration, I'm throwing
- 24:20everything into the "dump" folder
- 24:21because I'm not an electrician, I don't
- 24:23need this data. If I wanted to combine
- 24:25them and do something with them, I
- 24:27would, of course, create a separate
- 24:29folder. And I would save all these
- 24:30analyses there, as I showed you with
- 24:32the YouTube example, so that I could
- 24:34compile the entire structure in the
- 24:36future. So, good. So, he, ah, fully
- 24:39described why each video worked. He
- 24:43explained the idea of the video, he
- 24:45chose the hooks, he chose the structure
- 24:47, he described, ah, what causes,
- 24:49accordingly, the retention of this
- 24:51video, the strong wording that was used
- 24:53, the emotions that were used in this
- 24:55video. The same thing applies to the
- 24:57second video. So, he was able to
- 24:59analyze these videos without any
- 25:01problem. And now we have data that
- 25:03allows us to—we have the data that we
- 25:07collected. We have an analysis of what
- 25:10works. And now we can easily create
- 25:13content or do whatever we want with
- 25:15this data and this analysis. Since
- 25:17we're hiring a marketer, and this will
- 25:20be an e-marketer, the e-marketer must,
- 25:22accordingly, create content. Therefore,
- 25:25the third task is content creation. All
- 25:28these agents, they work precisely
- 25:30according to these files, where all the
- 25:32instructions are fully written out on
- 25:34how each of these agents should perform
- 25:36this or that task. Same thing. We don't
- 25:39have to manually send them links to the
- 25:41videos we like, although I still do
- 25:43that manually, because the easiest way
- 25:44for me to simply forward the videos
- 25:46that resonate with me to my antigravity
- 25:48agent is to simply forward them to my
- 25:49antigravity agent. Antigravity will
- 25:51then save them, and other agents will
- 25:53pick them up and be able to work with
- 25:55these materials. Because, I repeat,
- 25:58their entire database is in one place.
- 26:01Very convenient. I use the dump purely
- 26:04to demonstrate various topics. Good. We
- 26:07have the data, we have, uh, everything
- 26:09that resonates. Let's create some kind
- 26:11of content. I don't know, let him write
- 26:14me a post, or three versions of a post
- 26:17based on all this data and all this
- 26:20content analyzed. So, I'm sending him
- 26:23the following prompt. Let's do it this
- 26:25way. And I need to tell him to put all
- 26:27this in the dump folder, otherwise
- 26:29he'll throw it in with my Telegram
- 26:31posts, throw all this information, and
- 26:33then they'll get mixed up. I won't
- 26:35delete it, as always. So, the dump
- 26:37folder asked him to do this work,
- 26:39because, I repeat, he'll throw all this
- 26:41at me. So, use the material that's
- 26:43already in memory. Sources, the latest
- 26:46market research report. And the latest
- 26:48analysis of a video competitor. Choose
- 26:50the most promising topic. Your task is
- 26:52to prepare a post for Telegram for
- 26:54publication. You can ask him to do more
- 26:56than just a post for Telegram; you can
- 26:59ask him to do some kind of carousel for
- 27:01Instagram, that is, do anything. In
- 27:03this case, he asked for a post for
- 27:05Telegram. So, explain for each topic.
- 27:07So, it's completely described here what
- 27:09he needs to do. The same will be
- 27:11described in your instructions, which
- 27:14you'll download. And the same here in
- 27:17the prompts. He'll know exactly what he
- 27:19needs to do. He just won't need to
- 27:22write such a long prompt, such a big
- 27:24request. But since my topic isn't
- 27:27electrical, I'd rather let him do it
- 27:30correctly in the video. And when I
- 27:33assign tasks, I mean, for content
- 27:35creation, for example, I don't assign
- 27:37tasks at all for Telegram, because they
- 27:40create the content for me themselves.
- 27:43So, they write posts for me. Here's my
- 27:46Telegram in my style, here's Telegram,
- 27:48and here are ideas for text posts. Each
- 27:51of their agents, respectively, throws
- 27:54in ideas here, uh, every few days.
- 27:57Claude wrote the code based on, uh,
- 27:59what worked for me. So, you see, he's
- 28:02already analyzed my content, analyzed
- 28:05the spare videos that I have, and,
- 28:07accordingly, creates ready-made text
- 28:10content. Either for Instagram or
- 28:12Threads. They do the same for Instagram
- 28:15. I get ready-made ideas for carousels
- 28:18every day, and I tweak them. Maybe I
- 28:22choose different colors, maybe I choose
- 28:24different headlines, but they already
- 28:27know what I need. And, accordingly, my
- 28:29e-marketers create content for me.
- 28:32Every day they bring me a carousel so I
- 28:34can publish these carousels on social
- 28:37media. There are prompts here that you
- 28:39can copy, go somewhere, I don’t know,
- 28:41to the GPT chat. Let me show you. Here,
- 28:44I think, yesterday I manually sent them
- 28:46. Prompts to the GPT chat. And you can
- 28:48do the same thing using the a key. So.
- 28:52Ideas for something else. Here’s the
- 28:55carousel, see? Here’s yesterday’s
- 28:56carousel that I published. I uploaded
- 28:59the prompts here and, accordingly, got
- 29:02a ready-made carousel. The prompts were
- 29:05pre-made in my Obsidian. I just had to
- 29:09copy and paste them, or message the
- 29:11agent. I'd run out of budget on Kiap.
- 29:15It's the service we use to generate
- 29:17carousels. And I was too lazy to
- 29:20replenish it. I'd simply log into the
- 29:23GPT chat on my phone in my free time
- 29:26and send these requests, receive these
- 29:28images, save them, and publish them on
- 29:31Instagram. Okay, our agent has already
- 29:34completed the job. Let's see what they
- 29:39did there. So, we asked them to save
- 29:41all this information to the dump folder
- 29:44. Dump. Here's the post. Ah, Telegram
- 29:47posts for the audience. A socket from
- 29:50300 rubles turns into 15,000. How do
- 29:52you calculate an estimate? And how do
- 29:54you read an electrician's estimate? Why
- 29:56doesn't a normal electrician bother you
- 29:59with plastering? He's based on content.
- 30:02Aluminum fretwork panel, which is
- 30:04quietly waiting for a fire. Well,
- 30:05basically, not bad. And now the text
- 30:07posts themselves. Why this post in
- 30:09particular? So, he's selling us an idea
- 30:11. Post number two. And why this
- 30:13particular post? Post number three. And
- 30:15, accordingly, why this particular post
- 30:17? Everything we asked him to do was
- 30:20based on the materials, the sources we
- 30:24sent him. So, look, yes, our marketer
- 30:26found the data, analyzed the
- 30:28competitors, created the best content,
- 30:31and created the content. All of this
- 30:34setup, all of it, took us literally
- 30:36half an hour because we have ready-made
- 30:39materials, ready-made instructions,
- 30:41which you can download as the first
- 30:44pinned post. And then it's just routine
- 30:46. So, we tell him to do the same thing
- 30:49once a week. Once a week, he searches
- 30:51for data, and during the week, we send
- 30:54him some links for him to analyze and
- 30:56video content. And, ah, once a week, he
- 30:58creates not three, for example, but 10
- 31:01text posts for us, or seven text posts,
- 31:03and we simply copy, paste, copy, paste,
- 31:05and publish these text posts on our
- 31:07Instagram. And the marketer works. If
- 31:10we also connect some N8N system, we can
- 31:14automate content publishing and connect
- 31:18some chatpoints and attract clients
- 31:22through these automations. In other
- 31:25words, the AA marketer works without us
- 31:28. Our main task is simply setting them
- 31:31up initially. Guys, I hope this video
- 31:33was useful to you. You will definitely
- 31:35build yourself such a marketer for your
- 31:37business. I'm not saying you need to
- 31:40fire your existing marketers, but this
- 31:42one will work for you, it will work
- 31:44from standard subscriptions, from chp,
- 31:46antigravity, claude, the herm code
- 31:48works on all systems. The main thing is
- 31:51to just start building it, and then you
- 31:54can gradually delegate such tasks and
- 31:56increase the number of employees in
- 31:58your project, in your business. This
- 32:00was Grizuevich. Be sure to like and
- 32:02subscribe to the channel. Bye.
About this transcript
This page contains the full transcript of Как создать ИИ маркетолога: парсит, разбирает конкурентов и создаёт мощный контент | ИИ Сотрудник by Игорь Зуевич, generated from the public captions YouTube serves with the video. The transcript has 5,076 words across 764 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.
What you can do with it
Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.
Free YouTube transcript tool
YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.