The argument in one line.
Pairing Claude Code with a conversational chatbot replaces four manual steps in B2B lead gen -- research, outreach, qualification, and sales prep -- with a single pipeline that runs while you sleep.
Read if. Skip if.
READ IF YOU ARE…
- You run B2B outbound and spend hours per week finding prospects, enriching data, and writing personalized cold emails.
- You have a landing page with a static name-and-email form and no qualification before leads hit your CRM.
- You already use Claude Code for dev work and want to extend it into go-to-market workflows.
- You manage a small sales team and want automated pre-call briefs instead of manual research before every meeting.
SKIP IF…
- You need a production-grade system with GDPR compliance and robust error handling -- this is a live demo at crash-course fidelity.
- You want strategy; this is a specific tool implementation walkthrough (Cursor, Landbot, Google Sheets, Calendly).
The full version, fast.
Claude Code acts as the orchestration layer on both ends of the funnel: it researches and emails prospects outbound, then post-capture it reads a Google Sheet, tiers leads into hot/nurture/newsletter, and builds markdown sales briefs. The connective tissue is a claude.md file that stores the ICP, scoring rules, and research playbook -- Claude reads it automatically every session so the pipeline runs identically each time. Landbot handles the middle: qualifying every landing page visitor in conversation and pushing scores, transcripts, and structured data to the sheet. The result is a complete loop from cold prospect to pre-briefed sales call that requires no human intervention until someone is ready to close.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:37
01 · Intro: promise + system overview
Hook claim delivered, full pipeline described in one breath.
00:37 – 01:16
02 · Hybrid outbound/inbound architecture
Visual overview of the two-sided funnel: Claude Code on the edges, Landbot in the middle.
01:16 – 03:23
03 · Workspace setup: Cursor + claude.md
Live Cursor IDE walkthrough -- folder structure, claude.md contents, launching Claude via terminal, Claudia bypass tip.
03:23 – 04:56
04 · Outbound: prospect research via web
Single prompt: find 8 European B2B prospects matching ICP in claude.md, follow research flow, save CSV. Live execution.
04:56 – 06:34
05 · Reviewing prospects and drafting outreach
CSV shows 8 prospects. Next prompt: write personalized 80-word outreach emails using each company insight field, one CTA.
06:34 – 07:24
06 · Inbound: Landbot on the landing page
Demo of Landbot chatbot on landing page. Contrast with static forms: chatbots qualify in real time.
07:24 – 09:24
07 · Landbot builder + AI agent setup
Behind-the-scenes of the visual builder -- AI agent instructions, UTM hidden fields, storing responses in variables.
09:24 – 10:55
08 · Live qualification demo + Calendly
Live chatbot walkthrough: name, email, team size, timeline, prior tools. AI task block scores and writes brief. Calendly embedded.
10:55 – 12:40
09 · CRM output: Google Sheets + transcript
Landbot pushes each qualified lead -- name, email, score, transcript, sales brief -- to Google Sheets.
12:40 – 13:26
10 · Embed options + AI-generated CSS
Launch options: live chat, full-page embed, popup, shareable link. Claude-generated CSS and JS for custom styling.
13:26 – 14:35
11 · Scoring live leads from Google Sheet
Claude reads Google Sheet URL, pulls score column, categorizes: 3 hot, 1 nurture, 1 newsletter, 1 unscored.
14:35 – 16:26
12 · Building markdown sales briefs
For each hot lead: Claude visits website, reads transcript, generates brief -- company one-liner, pain points, talking points, validating question.
16:26 – 17:49
13 · Flow analytics + weekly review loop
Landbot Flow Analytics shows drop-off. Export PNG to Claude for diagnosis. Weekly workflow: paste Sheet URL, get briefs for hot leads.
17:49 – 18:26
14 · Wrap-up and CTA
Full system recap: research to capture to qualification to optimization. Plug for companion strategy video.
Lines worth screenshotting.
- A claude.md file in the project folder is permanent context -- Claude reads it every session, so the pipeline runs identically without re-prompting the ICP or scoring rules.
- Prospect research and personalized outreach are one Claude Code session: find ICPs, export to CSV, then draft emails referencing each company insight.
- A chatbot qualifies in real time -- you get team size, timeline, prior tools, and a score before any human touches the lead.
- Claude Code reads a live Google Sheet URL directly; no dedicated CRM integration is needed to score and segment leads.
- Sales prep briefs built from chatbot transcripts plus web research eliminate pre-call manual research for the sales team.
- Sharing Landbot flow analytics as a PNG with Claude weekly surfaces drop-off diagnoses without additional tooling.
- Full pipelines like this take under an hour to build but replace days of manual prospect research and qualification.
- Using Claudia (alternate Claude Code launcher) bypasses permission prompts for automation-heavy workflows.
- UTM hidden fields in Landbot let you personalize the chatbot opening message based on traffic source.
- An AI task block inside Landbot can write a sales brief automatically from the conversation before the lead hits your CRM.
One file is all the context Claude needs.
WHAT TO LEARN
A single claude.md file storing your ICP, scoring rules, and playbook turns Claude Code from a chatbot into a repeatable pipeline that runs the same way every time.
- A claude.md file in your project folder is permanent context -- Claude reads it automatically, so you never re-explain who your customer is or how scoring works across sessions.
- Prospect research and personalized outreach are one Claude Code session: find ICPs, export to CSV, then draft emails referencing each company insight in a single prompt chain.
- Chatbots qualify in real time where static forms cannot -- team size, timeline, prior tools, and intent signals are captured before any human touches the lead.
- Claude Code reads a live Google Sheet URL directly; no dedicated CRM integration is needed to tier leads into hot, nurture, and newsletter categories.
- Sales prep briefs built from chatbot transcripts plus web research eliminate pre-call manual research -- the salesperson opens a markdown file, not a browser.
- Landbot flow analytics exported as a PNG and shared with Claude weekly surfaces drop-off diagnoses without additional tooling or dashboards.
- Full pipelines like this take under an hour to build but replace days of manual prospect research, email writing, and lead qualification across a typical B2B sales cycle.
Terms worth knowing.
- claude.md
- A markdown file placed in a Claude Code project folder that acts as persistent session context. Claude reads it automatically on every run, storing ICP definitions, scoring rules, and playbooks without needing re-prompting.
- ICP (Ideal Customer Profile)
- A detailed description of the company type, size, geography, and role that makes the best-fit buyer. Used here as the filter Claude applies when researching prospects.
- Lead scoring
- Assigning a numeric or categorical score to a lead based on qualifying criteria to triage them into sales-ready, nurture, or newsletter tiers.
- UTM hidden fields
- URL parameters that Landbot reads silently to personalize the chatbot opening message based on traffic source, without asking the visitor.
- AI task block
- A Landbot flow component that calls an AI model mid-conversation to perform a discrete task -- such as writing a sales brief from collected conversation data -- before the lead is saved to the CRM.
- Claudia
- An alternative launcher for Claude Code that bypasses standard permission prompts, useful for automation workflows requiring unattended operation.
Things they pointed at.
10:55toolGoogle Sheets
Lines you could clip.
00:45
“That's like driving a sports car to the mailbox.”
01:13
“This isn't a skill. This is actual work getting done.”
03:14
“It's like you're the orchestrator, not the worker.”
06:02
“There's no template, mail merge, or anything like that.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphoranalogy
00:00Cloud Code can now do your lead generation for you on repeat, but only if you build the right skills and system. I mean, actually, find your prospects, write outreach, qualify them, build a capture system, basically all of it.
00:14In fact, we've used this exact setup in Lambda to build lead generation pipelines in under an hour. Systems that used to take days of manual work of outreach writing, tool configuration. All now in a couple of hours.
00:27So in this video, I'm gonna walk you through the entire setup so you can have this lead generation system running in Cloud Code while you focus on closing those leads. Let's get into it. When most people hear AI for lead gen, they might think like writing better emails or cleaning up a list.
00:44That's like driving a sports car to the mailbox. What I'm about to show you is a complete fun. Outbound and inbound running as one system.
00:52Cloud Code handles the outbound, finding prospects, writing personalized emails, driving traffic to our landing page. Landbot captures that traffic along with every organic visitor and qualifies them in conversation.
01:07Then Cloud Code takes over again to score the leads and build sales prep briefs on the hot ones. This isn't a skill.
01:15This is actual work getting done. Let me show you the setup. Okay.
01:19So here is the workspace. Basically, ignore this CSV file folder here because I used it for just to do some testing. Basically, you just have a folder, in this case, called lead gen demo, and you have this Claude dot m d file created.
01:32This file basically teaches Claude about your lead gen playbook. If I double click on it, you will see here information like the ideal customer profile, the tools, the research flow, how you work, the playbook.
01:45This file was done by Claude, but I just gave it all this information. Now side note, as you will see, I'm using Cursor here.
01:52Cursor is an app used basically by developers. I'm just using it as the interface. I'm not a developer, so I find it to be more user friendly than having all this all laying around.
02:04But to use it in cursor, basically, I have here the folder, the lead gen folder with the claud dot m d file that is just here that we open it. To open it, the terminal here, just double click on the folder and open an integrated terminal. Here, you will type claud and it would open up.
02:20Another thing is that I launch claud using claud here, typing claud. If I type Claudia instead of claud, it will launch a claud version that will bypass permissions, and it hasn't I haven't had any issues just yet.
02:33So going back to the Claude dot m d file. And this file, you write it once. You have it in the folder, and Claude will use it as context to basically do the tasks that you propose to it.
02:45You will not have to write the ideal customer profile each time you're talking to Claude. Claude will automatically read that file when you give him a task and will have this context.
02:57I won't have to reexplain the ideal customer profiles, the scoring rules. Claude just knows it based on this Claude dot m d file in the folder.
03:06And in practice, as I mentioned, I might have plenty of different tabs open. Claude doing in one, the research, on the other one, the outreach. You can have as many as you want.
03:16It's like . Okay. So this is the infrastructure.
03:21Now let me show you what is actually possible. Let's build this lead gen pipeline with Cloud Code. So let's start with outbound.
03:28We're gonna ask Cloud Code to find eight prospects that match our ICP that is found inside the Cloud dot MD file. We're gonna tell him, find me eight prospects matching the ICP in cloud.md, follow the research flow defined in cloud.md, and save the results in a CSV file.
03:45So I'm gonna click on send, it's gonna start working. As you can see, I'll research eight European b two b slash prospects matching the ICP that is found inside the cloud dot m d file as I mentioned.
03:57So cloud, as you can see, is searching in the web. It's pulling candidates, companies, visiting their websites.
04:04It's already telling me I have one candidate from Berlin. So it's basically going all over the web and looking for those ICPs, those leads.
04:12Okay. So now it's writing the CSV file with the information plot gathered.
04:17Great. So as you can see, it did the test. So it saved eight prospects to CSV prospect batch with the date on it and also logged the running activity dot m d.
04:26So it created a new dot m d file with a little bit deactivity of it. And, basically, the list is these eight ICPs.
04:33It also flagged a couple of things. I already did a test before the video, so it flagged that Kestrel was already in your existing prospects CSV file, so I swapped it for content. Five of the eight have no emails.
04:46In this case, I would recommend connected to maybe a tool like Apollo or Clay to enrich those leads and and find those emails, or you can obviously at the end try and find it manually, but it's fine. Great. So if we have a look at the file, we're gonna open up the finder.
05:01And in CSV, you'll see this new batch today. And if we look here, you will see the eight prospects.
05:07Now with the website, the contact name, the role, the LinkedIn, which is important if we wanna do some outreach there. The email is in blank. That's fine for now.
05:16And one sentence inside. Okay. So for each prospect in the CSV file that you created, write a short personalized outreach message using the one sentence inside field that created before.
05:27Keep it under 80 words, one CTA, book a fifteen minute call, or go to lambot.ao/demo. In there, we will have a chatbot built with Lambot to capture also organic leads and score them and save this information into this outreach dot CSV file, one row per prospect. So we click on send.
05:46It's gonna do the work, create another CSV file with this personalized outreach message. Let's see.
05:53So it's reading one file. Great. The one that it just created with these eight prospects, and it's thinking.
06:00So Claud is reading each prospect's insights and writing a personalized message for them. There's no template, mail merge, or anything like that.
06:10So it's writing the CSV file as we speak. It updated activity dot m d, which is fine for context, and it's done.
06:18Eight messages saved to CSV outreach batch. So if we open up Finder again, we will see it's this one here, and it created the outreach message, the subject for each one of the eight prospects that found before. Each message is personalized for the company.
06:34So now I've got outbound running, which is good. Personalized emails driving prospects to my landing page. But those prospects aren't the only people landing there.
06:43I've also got organic traffic, paid coming in. So both or all of the groups need to get captured the moment they hit the page. Let me show you how.
06:52So this is an example of a page that I was referring to, the lambo.ao/ demo page, for example, or a landing page. In here, a lot of traffic comes in.
07:02It can come from paid ads, search, organic traffic from the emails. So in here, we have this chatbot here or AI agent that basically guides the user, qualifies it.
07:13I'm gonna say, what can I help you with? I'll be very brief. Let's see what happens, but lead gen chatbot, and it's gonna work its magic and ask questions.
07:23Now what I wanna show you is that, obviously, when I refresh, a lot of companies get into this trap where instead of having a chatbot, they might have a form. I mean, if it's a long form, good luck because no one will fill that up. But if it's just name and email, you submit it.
07:38Okay. Fine. You get that information, but there's no qualification in time.
07:42And that's what's good about chatbots that you're able to qualify, ask questions. It's a conversation. So you get this qualification in time, and you can even book a call.
07:52We're gonna look at this now when I show you the bot that I built, but you can actually book a call in here in the actual conversation. So I already have a a chatbot created here with LANDBOD. This is a side by side view.
08:05Here, you have the final result of this chatbot, and this is the builder, the LANDBOD builder. Let me walk you through the bot. I used Cloud Code to build this chatbot here just to give me an outline of the different flows and how it could look like, and I just build it in the Copilot.
08:20It was fairly quick, the Lumbert Copilot. So, basically, the bot starts, and we have an ask a question. So, hey.
08:26Thanks for stopping by. What brings you here today? There are UTM terms you can use in the URLs to affect the actual chatbot that you have on the page.
08:34This is a bit more advanced, but it's called hidden fields. Just plugging it in here so you understand that if I change this and the UTM is, for example, carrots, we'll say here, interested in carrots. So you're able also to personalize the experience, just the first experience.
08:52And this interested in, the actual chatbot, the AI agent will know about it. So if it's for an outreach email, you might have this UTM term differently, and you get this personalized experience everywhere.
09:05So just wanted to plug that in here. So going back, uh, we have this ask a question as as I mentioned, and we're gonna save whatever the user says here in this first message variable. Once the user types so what brings here today, I'm gonna say, I want to book a call with the sales team.
09:23Gonna click on send. It's gonna jump into an AI agent. Now this AI agent has its instructions, context, its different conversation phases, script paths.
09:33This was done also by Claude. It helped me prompt this, and it has the qualifying questions. So the objective, team size, timeline, etcetera.
09:42So it's gonna store all this information under all these variables here that we're gonna later use to save. So happy to help you with that before I can can I get to the right person, full name, and email address? I'm gonna type NickNick@test.com.
09:56This is a test email. You could have in the instructions to test email, like, they're not real, whatever.
10:03You can have the agent to check for that. So it says, okay. Thanks, Nick.
10:07So it's gonna ask some quick questions, the qualifying questions. Again, in the instructions, you could tell the AI agent, in this case, to ask them one by one. I didn't specify that.
10:17That's fine for me. This way, we go quicker as well. So I'm gonna say one to 10.
10:21What's your expected timeline within thirty days? Have you used any similar tools before?
10:27Yes. I'm gonna say Drift. And are you looking for a solution right now?
10:32I'm just researching. Okay. So now the agent will gather all that information and store it in this different net field.
10:38So name, email, team size, timeline, etcetera, and it's gonna also create a score.
10:45So as you can see, it has two outputs, fully qualified or user refused or skips. In this case, I was fully qualified because I did answer all the questions. Now it went through an AI task block.
10:55This AI task block basically writes a short sales prep brief for our sales team that is stored under the variable sales brief with all the information that the AI agent gathered. This is going to go connected to a Google Sheets. Now this Google Sheets is this one here, and if we look, we get a new line in on this Google Sheets.
11:15I'm using Google Sheets as the CRM in this case. Obviously, you might wanna use your own CRM, whether that's HubSpot, Pipedrive. We do have plenty of native integrations in Lambo as you can see here, or you can connect it to Zapier to connect it to your own custom CRM.
11:31Or if that custom CRM has API capabilities with webhook block, you can connect basically to anything and send that information over there. So as you can see, I also send the transcript of the conversation for context, and here is the sales brief that the AI task block created.
11:48Obviously, you might make it look a bit better this this Google Sheets. I'm not a Google Sheet expert, but that's fine. We got the information.
11:55So now, basically, it's asking, no? Thanks at name. Would you like to book a call with our sales team?
12:00So I'm gonna click on yes for this case. If a user might not wanna do that, then there's the goodbye message so you would book a meeting. And here is the Calendly integration, which would just select a date and and a time.
12:13I'm just gonna skip for now, and then it would just go through the red path, event canceled. No problem. We'll reach out to you via email.
12:19Thanks. Now think that this full page chatbot could actually be on your landing page embedded. It could be in the form of a live chat.
12:28Once you're done with with the chatbot, if you click on launch, you can decide here how you want this to be embedded. It's a live chat, full page embed, or pop up, or you can share it as a link, which would be, like, a full page bot. Now in terms of design, you can also here, if you go into design, you can design the actual chatbot, and whether that's with a design template or, like, changing the background, the font, or we have the custom code that if you're actually using Cloud Code, this is very powerful.
12:58You can add CSS written by Cloud. You just copy paste it here or custom scripts with JavaScript.
13:04With Clod, you can do amazing stuff, honestly. I'm just using here the the basic template, but you can do a bunch of incredible stuff in terms of design. So now once the chatbot is done and embedded on the page, every visitor, whether that's outbound, organic, paid search, whatever, is captured and prequalified by this Lambert Floy.
13:23Now the leads are in. What does Cloud Code do now? Let me show you.
13:27So Lambert gave us a basically a full list of qualified leads in this Google Sheets, and this is where Claude Co. Does. It's sort of like second shift.
13:37Two things. One is score the fresh leads from this Google Sheets doc. So, basically, if we go back to cursor, I'm gonna open up an integrated terminal, Claudia.
13:48We're gonna tell Claude to read the Google Sheet from this URL, the Google Docs. Make sure that, uh, the share settings are are open to view or editor. This way Claude can go in and and look for the information.
14:00And based on the score column, tell me how many you would land as hot leads to nurture. Now it's gonna go into Google Docs. It's fetching all that information from the Google Sheets document, and it's gonna based on the score column that Lambda scores, they chat the conversation, it's gonna tell me which ones land as a hot lead, nurture, or should go to our newsletter.
14:21As you can see that based on the score, this one should be considered hot leads. This one's to nurture based on the scoring. So hot leads three, nurture one, and newsletter one.
14:30And there is one unscored. Score cell is blank. We can double check that.
14:35As I mentioned, three hot leads. What does Cloud Code do now? Obviously, Claude has more information than the one gathered from the Lambda chatbot.
14:45So, basically, what we're gonna tell Claude now for those three hot leads is that to build a proper sales brief, follow the Richmond flow from the cloud dot m d file, output one markdown file for lead name brief, include company one liner, one j told the bot probable pain points, three talking points, and one validating question to ask them.
15:07This is really useful for the sales team. So gonna click on send, and it's gonna start creating that file for those three hot leads. It's gonna go into the their website.
15:17Now it flagged that two of those three leads is the ones that I tested before. So, obviously, not gonna be able to test domain or find names because it watches the test. And it's really cool that Claude was able to understand that.
15:29So it's gonna look for that this one, Ana Garcia, go into fintechflow.io web search. It's asking me basically questions.
15:37No? So a few ways to proceed. Which one do you want?
15:42Build Breeze from the transcript data only. Treat these as frictional personas for the demo. You pull me to the real hot leads.
15:48So it's asking me these questions. So we're gonna just go with option one as as Claude was recommending. So I just actually typed one, and and that's it.
15:57But now it's gonna build a brief from the bot transcript data. Great. So it created this dot m d file for Ana Garcia based on the chat transcript and information from the Internet, and it's gonna do the same with Nick example, which is we can actually see it here.
16:13So sales brief, email score submitted, what they told the bot, probably paying points, free talking points, and validating questions. Great.
16:21This is very useful for sales. At this point, you know, we got a full system, outbound, inbound, qualification, scoring, all in one pipeline.
16:29All point of this is that it gets smarter over time. And I wanna show you a feature from Lambert. Obviously, you could use Claude here, and you should use Claude here to ask it questions.
16:39Is this working? Is this not share information? But we do have Insight Analyze in Lambda.
16:46We have Flow Analytics. You can actually see in the chatbot where users are dropping. Obviously, this chatbot is just a test.
16:53It doesn't have that much information. But you can see, for example, here, two users took this path. That's a 100% from the previous block.
17:00You can see how many users drop, where they drop, and you can export this as a PNG exporting flow. And you can share this with Claude, and Claude will sort of, like, help you out with the Lambda flow and what you should do in each step, why do you think they're dropping. So have it sort of like as this this companion.
17:20You can also go into Claude each week. Let's say it's Monday morning. The chatbot has been working the whole weekend.
17:28We go here. We type Claudia or Claude in your case, depends, and you tell him just go to Google Sheets.
17:35You paste the URL, and it would go into Google Sheets and basically look at the information and just write those sales briefs for those hot leads based on the scoring of the of the Lambert box. This system, this funnel for us has been working extremely well. I mean, just super easy, super fast, and it's it's just amazing.
17:54So now you've got the full system from research to capture to qualification to optimization. And if you wanna go deeper into the AI side of lead generation, I recommend you watch the video up here where my colleague, Irene, basically walks you through how to use Cloud Code as a copilot for lead generation.
18:14Strategies, prompts, workflows that are basically working for us on autopilot and that I'm pretty sure will be very helpful for you as well. That's basically it.
18:23I'm Nick for Lamad, and I'll see you in the next one.
The Hook
The bait, then the rug-pull.
Nick from Landbot opens with a blunt claim: Claude Code can run your entire lead gen pipeline on repeat. Not write better emails -- find, qualify, and brief your leads while you focus on closing.
Named ideas worth stealing.
00:37model
Hybrid Lead Gen Pipeline
- Outbound (Claude Code: research + email)
- Capture (Landbot: conversational qualification)
- Post-capture (Claude Code: scoring + briefs)
Three-stage model where Claude handles the research-heavy edges and Landbot handles the human-facing middle.
Steal forAny B2B SaaS with a landing page and an outbound motion
01:36concept
claude.md Playbook File
- ICP definition
- Scoring rules
- Research flow
- Outreach playbook
A markdown file Claude reads automatically every session -- no re-prompting required.
Steal forAny repeatable Claude Code workflow
13:56list
Lead Tiering
- Hot (fully qualified, ready to call)
- Nurture (interested, not ready)
- Newsletter (low intent)
Three-tier classification applied by Claude Code after reading the Landbot score column.
Steal forAny lead scoring system where a chatbot or form assigns a numeric score
How they asked for the click.
VERBAL ASK
18:01next-video
“I recommend you watch the video up here where my colleague Irene basically walks you through how to use Claude Code as a copilot for lead generation.”
Soft end-screen plug. No subscribe ask, no product push.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
