I Let Claude Replace My Marketing Team. Here's What Happened
A HubSpot marketing lead runs a free Claude Code "marketing team" against a real company's live site and grades every output: a generic headline, a genuinely great one, and the AI audit that faked its own homework.
Posted
1 weeks ago
Duration
Format
Demo
educational
Views
15.7K
162 likes
57 · 43
Big Idea
The argument in one line.
Agentic marketing systems built from context-file-driven Claude Code skills can produce above-average marketing work across audits, copy, positioning, and outbound email, but they still need a human with real marketing judgment to catch generic output and verify, via a trace command, what the AI actually did.
Who This Is For
Read if. Skip if.
READ IF YOU ARE…
A solo founder or a marketer on a small team who wants to know if an off-the-shelf Claude Code marketing skill can actually replace parts of a marketing hire.
Someone deciding whether to build or trust a multi-agent AI system and wants a model for structuring per-task context files.
A marketer curious what AI-generated positioning, copy, and outbound email actually look like against a real B2B SaaS site, with no cherry-picking.
SKIP IF…
You run a marketing team with dedicated audit, copywriting, and positioning specialists already producing above this quality bar.
You're looking for a finished, ready-to-ship swipe file rather than a raw, unedited test run with mixed results.
TL;DR
The full version, fast.
Kieran Flanagan runs a free, open-source Claude Code marketing skill against a real company, 1Mind, to test whether agentic marketing systems can replace a marketing team. The system runs a website audit, copywriting pass, positioning analysis, GEO (AI search) audit, and cold email sequence, each backed by its own skill-specific context file rather than one shared brand file. The audit's headline rewrite is generic, but the copywriting handoff (audit to copywriter to a review panel) produces a strong final headline. Positioning surfaces a genuinely sharp insight: buyers tell AI things they hide from sales reps. The GEO audit quietly skips querying any LLM and never flags it, caught only by running Claude's trace command. Verdict: good, not world-class, but better than an average marketer, worth using with real scrutiny.
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.
01 · The premise: can AI really replace a marketing team?
Cold open framing the test: run a free, open-source Claude Code marketing skill against a real company and see if it can genuinely replace parts of a marketing team.
01:20 – 03:02
02 · Picking 1Mind and touring the agentic marketing team
Uses Ramp's fast-growth AI go-to-market data to pick 1Mind, then tours the installed Claude Code marketing skill's full capability list.
03:02 – 06:41
03 · Website audit: the headline rewrite and four other fixes
Runs the audit cold, with no brand context file. The headline rewrite is generic, but the other four fixes (CTAs, proof number, mechanism, pricing) are judged genuinely good.
06:41 – 09:30
04 · Copy handoff: from generic headline to a ship-it headline
The audit hands off to a copywriter agent, which hands off to a review panel that de-slops the draft, producing "Your chatbot deflects. Your form delays. 1Mind sells."
09:30 – 11:51
05 · Positioning: the invented category and its limits
The system proposes "full-cycle AI sales agent" as a new category wedge against crowded "AI SDR" positioning, useful as a nudge but not a finished answer.
11:51 – 13:44
06 · Biggest find: buyers tell AI the truth
The standout insight: buyers reveal real budget, competitors, and objections to an AI with no social pressure, reframing 1Mind from cheaper rep to better sales intelligence.
13:44 – 15:43
07 · The GEO audit that never queried an LLM, caught by the trace
The AI-search audit returns a full scored report, but running the trace command reveals it never actually queried ChatGPT, Perplexity, or Google AI Overviews, and never flagged the gap.
15:43 – 17:57
08 · The cold email sequence: insider insight, then proof
A two-email outbound sequence uses a different angle per email, plain text, one CTA, and a $110K deal as the ceiling-case proof point.
17:57 – 21:46
09 · The verdict: good, not world-class, and why context files matter
Wraps with the honest grade (above average, not world-class), the real lesson about per-skill context files, and a warning to verify what any downloaded AI system is actually doing.
Atomic Insights
Lines worth screenshotting.
A free, open-source Claude Code skill can run a website audit, copywriting pass, positioning analysis, AI-search audit, and cold email sequence against a real company's live site with no cherry-picked example.
The system's first, unprompted headline rewrite for 1Mind, "AI sales agents that qualify, demo, and close while your team sleeps," sounded like every other AI SDR company's headline because it lacked real judgment about differentiation.
The best output came from a chained handoff, not a single pass: an audit agent surfaces the problem, a copywriter agent rewrites it, and a review panel agent de-slops the result before anything ships.
The agent that writes copy should never be the agent that reviews its own copy, the same principle behind a human editorial handoff.
The review panel's winning headline, "Your chatbot deflects. Your form delays. 1Mind sells," scored 85/100 and was strong enough that the host said he would ship it as-is.
The system invented a new market category, "full-cycle AI sales agent," as a wedge against the crowded "AI SDR" category, a directionally smart move even though the exact phrasing wasn't a great final category name.
The most valuable single insight the system surfaced was that buyers reveal real budget, timelines, and objections to an AI because there's no social performance pressure, whereas they hide that information from human sales reps.
A prior study found AI could predict purchase intent with about 90% accuracy from natural language conversation, a stronger signal than the data-enrichment methods most companies use to build intent models.
The reframe from "1Mind is a cheaper sales rep" to "1Mind is better sales intelligence" turns a commoditized AI SDR pitch into a defensible, differentiated position.
The GEO (AI search visibility) audit returned a full scored report and specific fixes without ever actually querying ChatGPT, Perplexity, or Google AI Overviews; it substituted plain web search and never disclosed the substitution.
Running Claude's trace command exposed that the GEO audit skipped querying LLMs because it lacked the tool access to do so, a failure the system did not flag on its own.
The core lesson from the trace failure: any AI marketing system should explicitly flag when it can't perform the task it was asked to do, instead of silently substituting a different method and reporting success.
The system's real technical advantage isn't one shared brand file, it's that every individual skill (audit, copy, positioning, GEO) reads its own dedicated context file, like copy.md, written specifically for that task.
The cold outbound email sequence used a different angle per email (insider insight, then hard proof) specifically so the second email wasn't just "bumping this to the top of your inbox."
Plain-text emails with no HTML template, one CTA per email, and PS lines that carry real content, not filler, all tested as deliberate, learned outbound-email design choices in the generated sequence.
The email sequence's proof point, a $110,000 deal closed end-to-end with no human in the loop, was flagged by the system itself as the ceiling case, not the typical outcome, showing appropriate honesty about a standout result.
Overall verdict: these agentic marketing systems produce work that's better than an average marketer's output but not world-class, which still meaningfully elevates a solo founder or small marketing team.
Anyone downloading someone else's AI agent system from GitHub should validate it's from a trusted source first, then use a trace or debug command to verify what it's actually doing before trusting its output.
Takeaway
What actually happens when you point Claude at a real marketing team's job
WHAT TO LEARN
An agentic marketing system can produce work that beats an average marketer on audits, copy, and positioning, but only if you chain the right agents together, give each one its own context file, and verify what it did instead of trusting its report.
01The premise: can AI really replace a marketing team?
The video runs a real, unedited test of a free open-source Claude Code marketing skill against a live company's actual website, not a staged example.
The premise being tested is whether an agentic marketing system can replace a portion of a real marketing team, not whether AI can write generic marketing copy.
02Picking 1Mind and touring the agentic marketing team
The host used Ramp's fast-growth AI go-to-market data to pick 1Mind, a real company he already uses, rather than a friendly or cherry-picked demo brand.
The Claude Code marketing skill bundles website audits, copywriting, positioning, paid ads, SEO/GEO, social, competitive intelligence, and launch/App Store tasks into one installable system.
Running the audit with no brand context file, just one loose sentence describing the company, was a deliberate choice to test the skill's baseline value without pre-loaded advantages.
03Website audit: the headline rewrite and four other fixes
The system's first headline rewrite, "AI sales agents that qualify, demo, and close while your team sleeps," is indistinguishable from what any other AI SDR company would write, showing AI alone can't replace real copywriting judgment.
The audit's other four fixes, collapsing four CTAs into one, putting a hard proof number above the fold, showing the AI-joins-your-Zoom-call mechanism, and publishing pricing for AI search, were all judged as genuinely good, specific advice.
The quality of any AI audit is capped by the quality of its context files: without a real brand and ICP file built from sales calls and internal data, it can only audit from a company's own external assets.
04Copy handoff: from generic headline to a ship-it headline
The strongest results came from chaining agents: an audit agent finds the problem, a dedicated copywriter agent fixes it, and a separate review panel agent scores and de-slops the result, mirroring a real editorial handoff.
The system's rule that the agent doing the work should never review its own work is presented as the key structural difference between systems that produce generic output and systems that produce good output.
The winning headline, "Your chatbot deflects. Your form delays. 1Mind sells," scored 85/100 and was strong enough that the host said he would ship it as-is on 1Mind's homepage.
05Positioning: the invented category and its limits
The system proposed a new market category, "full-cycle AI sales agent," as a wedge against the crowded "AI SDR" category, directionally smart even though the exact category name wasn't a winner.
Positioning is judged as the hardest job the system attempted, useful as a nudge toward differentiation but not a replacement for a skilled product marketer who can execute the idea.
06Biggest find: buyers tell AI the truth
The system's single best insight across the whole run was that buyers tell an AI things they hide from human sales reps, including real budget figures, named competitors, and honest objections, because there's no social performance pressure with a machine.
A cited study found AI could predict purchase intent to about 90% accuracy from natural language, a stronger signal than the data-enrichment methods most companies rely on to build intent models.
This insight reframes the product from "a cheaper sales rep" to "better sales intelligence," a materially stronger and more defensible position.
07The GEO audit that never queried an LLM, caught by the trace
The GEO audit returned a full scored report and specific fixes without ever actually querying ChatGPT, Perplexity, or Google AI Overviews, silently substituting plain web search instead.
Running Claude's trace command exposed the substitution: the audit lacked the tool access needed to query LLMs directly and never disclosed that limitation in its output.
The core lesson: any AI system should explicitly flag when it can't complete the task it was asked to do, instead of quietly using a different method and reporting success as if nothing was missing.
08The cold email sequence: insider insight, then proof
The generated two-email sequence used a different angle per email, an insider insight in email one and hard proof in email two, specifically so the follow-up wasn't just "bumping the first email to the top of the inbox."
Design choices tested as deliberate rather than templated: plain text with no HTML, one CTA per email, and PS lines that carried real content instead of filler.
The proof point, a $110,000 deal closed end-to-end with no human in the loop, was flagged by the system itself as the ceiling case rather than the typical outcome, which reads as appropriate honesty rather than hype.
09The verdict: good, not world-class, and why context files matter
The system's real advantage isn't a single shared brand file, it's that each individual skill (audit, copy, positioning, GEO) reads its own dedicated context file written specifically for that task.
Overall verdict: the output is better than an average marketer's work but not world-class, which still meaningfully elevates a solo founder or small team even though it won't replace a full specialist marketing department.
Before trusting any downloaded AI agent system, confirm it's from a trusted source, then use a trace or debug command to verify what it's actually doing rather than assuming its reported output matches its real actions.
Glossary
Terms worth knowing.
GEO
Generative Engine Optimization: the practice of optimizing a website so it gets cited and surfaced correctly inside AI search tools like ChatGPT, Perplexity, and Google AI Overviews.
ICP
Ideal Customer Profile: the specific type of company or buyer a business's marketing and sales are built to target.
De-slop
An editorial review pass, run by a separate AI agent, that strips out generic AI writing patterns like em-dash clauses, hedging language, and clichéd tricolons before content ships.
Trace command
A debug command in Claude Code that shows exactly what actions an AI agent took, letting a user verify a report or output against what the system actually did rather than what it claimed to do.
Brand context file
A markdown document holding a company's brand voice, audience, and messaging guidance that an AI system reads before generating content.
Full-cycle AI sales agent
A positioning category the AI system proposed for 1Mind, distinguishing it from single-step "AI SDR" tools by claiming it can qualify, demo, handle objections, and close a deal end-to-end.
“Buyers tell AI the truth. You're not a better sales rep, you're a better sales intelligence.”
reframes a whole product category in one line→ newsletter pull-quote↗ Tweet quote
14:40
“It was not querying any of the LLMs. It was really just doing web search, and it didn't flag that.”
the twist/gotcha moment of the whole video→ TikTok hook↗ Tweet quote
17:37
“One of our superhumans closed a $110,000 deal end-to-end with no human in the loop.”
concrete, specific proof number→ IG reel cold open↗ Tweet quote
The Script
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.
17px
Every influencer is telling you online you can replace your entire marketing team with an agentic marketing system. But is that true? On this episode, we're about to find out.
I am going to use an entire agentic marketing team to do all of the marketing for a high growth AI company. And we're really going to see, can we fire our entire marketing team and replace them with AI? All of that and more on this episode of Marketing Against the Grain.
Okay, so we are going to do some real marketing on this episode. We're going to do it live using an AI Git repository that blew up last week. Some company outsourced their entire quote -unquote marketing team.
It's all done through a bunch of skills and plugins that Cloud Code or ChatGPT can use. You can install it and it theoretically gives you an entire marketing team. Now what we're going to do is actually use that to do real marketing for a fast growth AI company and determine together Is this really good?
Can it really do marketing? What does the output look like? Can you really replace a portion of your marketing team or a lot of your marketing team with these sales plugins, these marketing systems that people are outsourcing and giving away online?
So first of all, let's pick the company. And the way we're going to do this is Ramp released tons of cool data on fast growth, AI, go -to -market companies. always interested in those and they give you lots of data and lots of information but we're going to go through and look at i want to look at which ones are growing the fastest so we're going to look at which ones are going the fastest so rocks attention one my momentum and 11x so These are kind of fast growth AI companies when they go to market space.
We're going to pick OneMind because we are a user of OneMind and HubSpot, but also they're number three. And so if you were doing marketing and you want a marketing system to be really good, you would want to help take number three to number one. So let's pick OneMind as our kind of case study here.
And now I want to show you the marketing team we're going to use. So we have it installed. We're going to access it here.
uh in cloud code and there's a bunch you can do with this some of it would be better if you had internal data but we're going to use it for a company that whose data we don't have but there's still a lot you can do so you can basically audit the website you can do a bunch of things around copy and messaging so conversion and optimization uh copyright and you can do a lot of things around paid ads email seo and ai search social competitive intelligence a lot of product marketing here in the launch and position in App Store and Analytics.
So I was kind of messing around. I think we're going to do a website audit. We're going to do some copy because copy is really hard to do.
In that, we are going to do some product positioning because product positioning is really hard to do. We're going to do some competitive analysis. We're going to do some geo, which means we're going to actually figure out if we can appear higher in AI search engines.
So we're going to do some real marketing here for OneMind. So stay tuned for OneMind. Obviously, this is going to take me a little bit of time to do because I'm going to do it live.
But we're going to edit it all down so you can just see what I'm going to do and then see the output. And I can give my raw unfiltered feedback on if this is good or not. So there's really two options.
If you're creating a brand context file. I would kind of recommend you do it from internal information, which maybe this is what they want you to do. What we're going to do is we got rid of the brand context file and we're going to run it cold.
We're going to just say audit one mind. They sell AI sales agents to B2B revenue leaders and enterprise SaaS companies. So let's see what it will come back with.
But I think this is a cleaner way to see the value of the skill. And so here's what you get back. And there is a full audit that I'll quickly bring up, but these are the kind of quick fixes.
So the first one is that I replaced the headline. Everyone wins with. AI agents that qualify demo and close.
Here's OneMind's website. It's basically saying, hey, you don't really know what you all do. And it wants to change that to be very specific.
AI sales agents that qualify demo and close while your team sleeps. Now, what it gives us is a pretty AI type headline that doesn't really distinguish from any other type of headline you'll find in this space. These are AI SDRs.
And so I think this is where you cannot replace marketing judgment, marketing taste, and actually real copywriting skills, right? I think copywriting skills are timeless. AI has not changed that.
AI is not a good copywriter. And so I think the push it's given, which is if you read this headline, you don't really know what the value of the product is. Hey, everyone wins.
Sounds good. Customers in your business. Sure.
I give every buyer and customer a delightful experience, but it's not speaking to like the problem itself, right? That's like quite broad. And so there is truth in the middle and that this could be more specific to the value one might create, but you certainly want to want to go and change it to AI sales agents that qualify demo and close while your team sleeps because every other AI SDR, I suspect would actually sound like that.
Hey, if you like what you heard in this episode and you want more, click the link in the description or scan that QR code. We have a ton of incredible resources that can get you started doing this stuff in a very short amount of time. The other things it's saying is there's four CTA types and we should collapse into just one, which is watch Mindy sell.
I don't think we should collapse them into one because they serve different purposes, but we should certainly work in the name. So you can like watch Mindy sell here. And then the other ones are really like, you know, book time with us.
And this is case studies. So I think again, some truth in the middle, put a hard number above the fold. So it's basically saying, Hey, you have like a bunch of cool things from customers and you don't see them on the homepage.
I think that that's probably good advice. And then show the mechanism AI joins your zoom call. It's basically pointed out that one of the value props of one mind is actually the AI can join your zoom call and live demo the product.
And that is actually pretty huge for B2B and it's not anywhere here. You don't really get a sense of that. And I think what they want you to do is really try and talk to Mindy.
But if you're new to the space, you may not understand that. And so I do think that some of the kind of core value that OneMind deliver is not representative on this page. And then basically it's telling you to publish your pricing because transparent pricing, especially for answer engine optimization is pretty big because someone's searching for you in LLMs.
They kind of want to know what price you are. And if they don't know what price you are, they may not ever visit your website. Now, it's harder for something like OneMind because I suspect their pricing fluctuates a lot based on consumption.
That is like the headlines, right? So I've given you the kind of core headlines. These are the most immediate fixes.
But if you actually look here, it's a pretty good overview, pretty good audit here, right? I think if you're a small business or a marketing team with like limited time and limited resources. This is pretty good, right?
This is a pretty good way to audit your website. I think the value of this audit is really going to be dictated by how good your brand context file is, how good your ICP context file is, so it actually knows to audit based upon who your customer is and that you've built that context from information like sales calls, internal information that is unique to your business versus pulling from your external assets that you want to audit, right?
All right. So we'll do one more little bit of copy and then we'll jump into some other things that the agentic marketing team can do. And let's see how the copy is.
And so we jumped the gun there on the one mind audit because there are follow -up skills, which makes sense. If you have an agentic marketing team, you have someone that will do the audit, pass that off to a copywriter and the copywriter will improve the things the audit has surfaced. So I want to give them credit here.
And again, we'll provide the link in the comments. If you want to go download this from the GitHub yourself, it's not us. It's not mine.
I have my own content system and my own marketing system that I will continue to give away on this show. But I wanted to try someone else's to see if it was true that you can replace your marketing team with these agentic systems. And we'll give you the link.
You can download it from the GitHub. So basically what it's doing here is they have a copywriting process and then they score them. So I've done this as well.
So you can run through a different panel. And the big thing here is when you're creating systems, The agent who's doing the copy passes that off to other agents to review.
The agent is not reviewing its own work. And again, that's what I changed about the audit. And here they hand off to a panel to something that deslops.
I hate the words AI slop. And so what does it come back with? Again, I can look at the full one in Obsidian, which we'll quickly do.
But it's come up with a recommended. It has its own score mechanism. I do score everything myself and all the systems that I create.
I don't know, I find the scores, I kind of like, do I trust them? I don't know, even though I create them and the methodology. But I will say this is a headline is pretty good.
Your chatbot deflects, your form delays, one mind sells. I like that. I think that's much better than what they have today.
And it's much better than the generic initial headline that we were given. And that really shows you having these purpose -built skills. really matters because these skills then can actually ingest context.
In this example, the context to write great copy and then give you something like this. So here's three runners up. Went to audience burned by failed AI.
95 % of AI sales pilots fail. The other 5 % don't use chatbots. Pretty, I like that.
Want to self -select the exact buyer. Build for teams that outgrow chatbots but aren't ready to double headcount. And then differentiation.
Okay, this is a good one. Differentiation matters most. An AI rep that joins the video call and actually demos the product.
All right, that's not very good. That's not really differentiation. I think these are good, right?
Again, if you are a solo marketer, if you are a founder, this is probably better than what you will get back from most average marketers. And so I think this is pretty good. So audit, pretty good.
Copy, pretty good. I think what we're proven here is that AI can do an above average job.
It cannot do a world -class job. Let's get into some other tasks. I believe positioning is incredibly important and they have a positioning skill.
And I think most positioning skills that I've seen are terrible. So now we've done our audit. We think that's good.
We've got a great headline, a headline that is better than whoever is doing OneMind today. Sorry for whoever's done OneMind's headline today. And was much better than the first agent give us because the agent was doing the audit and then plus given the suggestions.
And it's always better than if you take those suggestions and give them to a specific agent who is tasked at doing that job. That's how you build great systems. So now we're going into positioning.
So it comes up with a category, full cycle AI sales agent. OneMind is the only one that does qualify, live demo, objections, close, video calls. The full cycle modifier is the wedge.
So full cycle AI sales union is not going to be a great category. However, again, and I'm trying to give the non AI influencer, everything is insane version of this. I'm giving the version of like, Hey, a real founder or a marketer is this useful for them?
I think what we're continuing to show here is AI is better than an average marketer and a great marketer knows how to take this. and turn it into something. So this is not the right category, but what it's giving you is a nudge to say you're better than an AISDR, which is a very crowded space.
And you might want to start to think about how to remove yourself from that space and articulate your product in a different way so you separate yourself from the pack. And I think that is pretty good advice. The thing is, it's not easy to do.
And so it's not going to replace the actual incredible product marketer who can figure out to do that, but it gives founders and marketers a nudge in the correct direction. Biggest find in the buyers tell AI the truth. Research underused.
This reframes one mind from a cheaper sales rep to a better sales intelligence. So buyers reveal intent to AI. This is a great one because there was a study last year that said that AI could predict purchase intent to within 90 % of accuracy because of the person speaking to it.
And so natural language is the best way to decipher intent to buy. And so what one mind are sitting on is a better intent model.
Today, the way most companies build intent models is through data enrichment. That data enrichment does not include natural language, speaking to your customers. A far better predictor of purchasing is talking to your customers.
And this is what it's getting at. It's saying, hey, you're not a better sales rep. You are a better sales intelligence.
That is awesome, right? This is super cool. This is so much better than what I think an average product marketer would probably come back with.
So what you're learning here is a bunch of ways to actually create real AI systems as well. So we're not going to go into the, this would be too much, but there's a whole lot of work in here. There's a ton of work in here.
And again, I would write this good, not very good, right? And that's still a great mark for AI in marketing, right? I think being good is above the average and it allows a marketer to really elevate what they work on because AI is doing a lot of this work.
Kieran and I are doing a live AMA at HubSpot's Unbound event on September 18th at 1115 AM. We want your questions. Please submit any question you have for us.
We are going to answer them directly and honestly. And the best questions are going to get picked and answered on stage live in front of everyone. And we'll give you a shout out.
So please click the link below and drop us your questions. So awesome. Website audit, copy, positioning.
I think they're all pretty good uses of this system. Let's do one more here. Let's get into some demand gen stuff.
We're doing a lot of copy. We're trying to really have it do copy because that's one of the hardest things to do. Competitive teardowns, AI is great at that.
I don't think there's anything we're going to learn by using the competitive teardown. I'm sure it's very good. AI is very good at it as a research tool.
I'm actually interested in this geo, which is like how you're appearing in LLM. So let's actually run that as our next one. Okay, so we ran the geo audit.
And I want to show you a quick thing that happened because it actually is useful for you if you're downloading different systems or skills to use from a repository to use. So we kind of came back and give me a, you know, a score basically said it wasn't, they weren't very visible. And here's some of the things you can do to fix.
And all of these, I think make a lot of sense, which is they don't have a lot of things in their homepage to extract into the LLMs. So they're going to be invisible. But really what I noticed was it wasn't doing this properly.
It was not querying any of the LLMs. It was really just doing web search. And so what I did was I ran a trace and the trace basically just tells me exactly what it did.
And so you can see here. This one is, again, loading these kind of reference files. I'm very curious how it's loading this because it doesn't exist.
So it's kind of making up things, which is kind of a little worrying. And then you can see here it's querying the web. It's not querying LLMs, which makes sense because really to query LLMs, it's going to need some tools.
But it was interesting that it did not tell me that. It just went ahead and ran it. It didn't actually flag.
really in your system, you should always flag if the AI is not able to do the thing it's able to do. So give me a full audit without doing it properly. And I could see that from the trace.
But I still think good learnings here, right? Like that trace command, if you think something is going wrong, or you want to see how something works, but you should always be curious about how something is working. The trace is really useful.
Let's do one more. Let's do the email. Based on your recommendations on how we can improve positioning for OneMind and who our ICP is, draft me a two email sequence campaign to book meetings with that ICP for our sales team.
All right, let's do the email. So we're going to go view this whole thing here. Let's get the highlights.
So here's the two email secrets. Day one, your buyers are lying to your reps. Your buyers are lying to reps.
Lead with buyer's truth. Research from the position and work. So this is the, hey, the big reveal that you have sales intelligence.
You're building sales intelligence, not just an AI SDR. It's the insight this ICP hasn't heard yet. So let's go into look at that one.
And then day three, 28 days to 15 with no extra headcount. Follows up with the proof. Okay, it's a sales cycle.
That's kind of not great because I wouldn't know what that means. But key decision, each email uses a different angle. So email two isn't just bumping this to your top of your inbox.
That's pretty interesting. Most people do do that. Plain text, no HTML template.
One call to action per email. Both PS lines carry real content, no filler. So let's get into it.
De -slopped against the full pattern checklist. All right, cool. So it gives me this, the inside statement.
Your buyers are lying to your reps. Preview text. You research on what happens when AI runs the sales conversation.
That's pretty cool. First, your buyers name competitors on the first call. They share real budget figures.
They say out loud why they're not ready to buy. They just don't say any of it to your reps. This is very good.
This is better than most emails I get. I actually kind of am worried about how good this is. And then email to the proof.
Okay, we can definitely improve this headline a little. Again, judgment really matters. Your domain expertise really matters.
But let's see, a quick follow -up. I want to share specific numbers rather than make you take my word for it. Very cool.
One of our superhumans closed $110 ,000 deal end -to -end with no human in the loop. That's pretty sweet. If your team is spending more time qualifying than closing, awesome.
I think the average company could use this and it would be better than what they're doing today. And so email is really good. And so we're going to stop there.
And so what have we learned? I came into this thinking that I build a lot of AI systems and I think there's a lot of junk being built. And I think it takes real time to build something good.
What I wanted to show here is can you really implement an agentic marketing team and have that team do a lot of your marketing work for you? And it's not for large organizations with very large marketing teams. You probably have individuals doing each piece of the marketing for them.
But if you're a solar founder, if you're a smallish team and you want to be able to do more, I think we have proved that these agentic marketing systems are really good. They are likely better in many cases than average marketeers. And what have we learned?
I think we've learned a lot about building marketing systems here. They actually have a lot of context files. I, in the start of this video, said, hey, they only have one context file, the brand context file.
My system has many. They actually, when you run the trace on what it's doing, all the scales have their own context files. And so you need to have real context files mapped to each scale.
We're doing the audit. They had a context file. They read from on how you do a great audit when they were doing copy.
And that headline was really good. I would use that copy. If I was one mind, I would just copy and paste that copy, that headline for my homepage.
They read the copy .md file to figure out how to actually create great copy, great headlines. And they use that to create that headline. So you had MD files for each of the different skills that helped that skill be better at the task it was trying to do.
If you download something from the GitHub or you use someone else's system, you want to make sure it's from a trusted source first. But then if you're like, hey, I want to figure out what this is doing so I can customize my needs, that trace command can tell you exactly what's happening. So you don't want to be an AI slob and just like, hey, I'll just download this and let it do what it's going to do.
You want to be really inquisitive, curious, like how can I make this better? AI will do odd things at times, right? That's the other reason you want to do that trace command.
Like, is it doing what I think it's doing? So when we did the geo audit, you saw that it was loading a context file, the brand context that is not there anymore because I removed it. Now, what it was doing actually was remembering the context in the conversation.
So that's fine. But it ran the entire geo audit, which is meant to be querying LLMs. and did it all via the web and then gave me their report back.
Now, if I was just a person using the system and wasn't curious, I would say, oh, this is for my LLM results. But actually, if you looked at what it was doing, it didn't have the ability to pull from the LLMs because it couldn't use cloud computer to access perplexity and chat GPT and AI overviews. So it just said, here's an audit.
I did it a different way, but didn't flag that it didn't do the thing you expected it to do. All in all, I think we have... demonstrated that if you really take the time to build a marketing system, fine tune to your business.
And I do think now if you're a team of any size, if it's fine tuned to your business, it is going to really elevate your work and allow your marketers to work on higher value tasks. And AI is an incredible marketing assistant. And these systems are really powerful.
I hope this was useful for you. I hope you got a lot from it. If you want to download this.
marketing OS that we've used in this video. We'll provide the link to the GitHub. Again, it's not ours.
It's someone else's. So do your own validation and making sure you trust it. And we will see you on a future episode.
This data is wrong every freaking time. Have you heard of HubSpot? HubSpot is a CRM platform where everything is fully integrated.
Whoa, I can see the client's whole history. Calls, support tickets, emails. And here's a task from three days ago I totally missed.
HubSpot. Grow better.
The Hook
The bait, then the rug-pull.
Kieran Flanagan points a free, open-source Claude Code marketing skill at a company he actually uses, 1Mind, and runs a live website audit, copywriting pass, positioning analysis, AI-search audit, and cold email sequence against its real site. Then he gives an unfiltered verdict on every output, including the one moment the system quietly skipped the task it claimed to complete.
Frameworks
Named ideas worth stealing.
04:41list
Five-fix website audit priority order
Replace the headline
Collapse 4 CTAs into 1
Put one hard number above the fold
Show the mechanism
Publish a pricing anchor
The priority order the Claude marketing skill used to rank fixes to 1Mind's homepage, from highest to lowest impact.
Steal forany B2B SaaS homepage audit checklist
14:55model
GEO citability scoring model
Extractability
Specificity
Entity clarity
Corroboration
Machine access
Five weighted levers the GEO audit scored to produce a total "citability" score (24/100 for 1Mind) predicting how visible a site is inside AI search answers.
Steal forauditing any site's AI-search visibility
CTA Breakdown
How they asked for the click.
VERBAL ASK
04:48link
“if you like what you heard in this episode and you want more, click the link in the description or scan that QR code”
Soft mid-video pitch: a QR code plus a description link to HubSpot's own free resource, the same category of tool being reviewed, framed as bonus getting-started material rather than a hard sell.
Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
Add to Preferred SourcesOpens your Google source preferences with us pre-loaded. Tick the box and you're done.
RyMac wires Claude Code straight into Consulti and the ICM folder system, live and unedited, to replace a four-tool cold email stack with one same-day pipeline.
A screen-share walkthrough of wiring a brand-aware Claude Code project to Higgsfield's AI models so Claude does the prompting, generates a full asset suite for a fictional energy drink brand, and reports back what each piece cost.
An 18-minute walkthrough of wiring Claude Code into Clay's data platform to source, enrich, and write cold email copy for 50 leads from one natural-language prompt.
A screen-recorded walkthrough plugs Clay's new CLI straight into Claude Code, turning a single spoken prompt into a scraped, verified, cold-email-ready B2B prospect list.
A 31-minute live Q&A vlog where a $4M/year operator tears down cold email copy, lays out a RAG knowledge-base pattern, and explains why building a brand before a business is the slowest path to cash.
00:25
01:00
01:14
01:30
01:45
01:55
02:10
02:35
02:45
02:55
03:03
03:11
03:22
03:25
03:28
03:35
04:00
04:54
05:00
05:14
05:22
05:23
05:27
05:35
05:47
05:54
06:05
06:10
06:18
06:24
06:46
07:01
07:11
07:15
07:19
07:26
07:40
07:49
08:01
08:05
08:08
08:15
08:25
08:35
08:50
08:56
09:11
09:54
10:09
10:22
10:34
10:41
11:15
11:51
11:55
11:58
12:02
12:20
12:33
12:43
13:14
13:40
13:55
14:39
14:55
15:05
15:15
15:23
15:30
15:36
15:55
16:05
16:15
16:25
16:35
16:46
17:00
17:13
17:21
17:25
17:28
17:34
17:42
17:54
17:56
18:02
18:07
18:15
18:25
18:35
19:05
19:21
19:33
19:40
19:46
19:58
20:15
20:31
20:51
21:25
59:40
1:00:02
38:28
18:02
Video of the Day15:24
31:30