



I cold pitch coliving founders, retreat operators, wellness coaches and the like. These people are good at what they do and oftentimes have no idea why their marketing isn't working.
Reaching them through social outreach is just not going to happen. And while, cold outreach has a reputation for being spray-and-pray, I have been detrmined to make it the opposite of that. So I built something in Claude code with no skills whatsoever
Basically, it's a local Streamlit dashboard that runs the whole acquisition pipeline, start to finish.
Using coliving as an example, it finds the lead by starting with a region/country selector that pulls cities with coliving spaces from Nomad List. I pick a city, get a grid of one-click search links via a Google operator search, Airbnb, Booking.com, Maps, Instagram, Facebook, coliving.com.
From here I save it to a leads queue. Name, city, website, source. All stored locally.
When I have e few in there I do a mini audit of their site and socials. I feed the tool everything I can find:
* Raw HTML source of their homepage (Ctrl+U, paste the whole thing). BeautifulSoup strips the JS bundles, inline styles, SVG blobs and base64 images before it goes to Claude, so the token count stays reasonable.
* Then I voice note to Claude my own manual notes from browsing the site. What could be communicated better, what is the imagery like, is there a disconnect anywhere etc etc. These observations get weighted heavily — they override Claude's HTML read if there's overlap.
* I paste in their Instagram bio and recent captions. How active they are. What they post about. Do they have structure. Is there an obvious strategy?
* I paste in Google Tag Assistant output. What's actually firing at runtime. GA4, Meta Pixel, GTM, Hotjar. Or nothing. That's usually the finding!
* It also has an upload to view screenshots if there's a layout or formatting issue I want Claude to see.
Then I hit 'run audit' and it generates a personalised pitch in one pass.
The audit covers: SEO gaps, tracking setup, copy quality, messaging consistency between homepage and social, missing conversion elements. Then it splits cleanly into findings and a pitch.
The pitch is editable in a text area. I can tweak it before logging. Word count and character count live beneath it so I know what I'm sending.
The whole thing goes to Notion with one click via either "Pitch sent" or "Pitch pending" and it saves the business name, contact details, channel, ICP segment, findings, and pitch text.
It also sets a follow-up date automatically and the form clears for the next lead.
There's a chat panel underneath where I can drill into specific findings, ask for three alternative opening lines, get the pitch shortened, or ask what the single biggest gap is between how they present on Instagram vs the website.
So far I have completed 35 audits. Got 18 responses, booked eight 45-min calls and landed two clients.
After the actual call, I have a separate tab in the dashboard where I paste the Otter/Fireflies transcript, add any notes, paste in the original pitch for context and Claude extracts: what we discussed, what's working, what's broken (with root cause and impact), and three priorities in order.
Then it fills a duplicated Notion template page which is a summary of my findings and a bespoke Loom video with my my paid offer attached. All of the blocks are mapped by heading structure, callout colours, bullet positions. Date, Loom embed URL, all of it.
The hardest part was keeping my voice consistent. The first dozen versions kept generating shitloads of bloody em dashes, words like "genuinely", phrases like "lands well", you know the drill. It also had this awful habit of predicting which operators in a market will succeed and which won't which makes me sound like a prophet at best and a prick at worst.
The fix was a hardcoded constant injected into every generation prompt, before whatever the config says. That eliminated all the negative-to-positive reframing and AI filler words.
Cold pitching has stopped being something I dread. In fact, it is kinda fun. Every lead gets a specific audit, not a template. I know what's broken before I send anything. The pitch references real things and I often get feedback like "how did you know that!"
People have really responded to the specificity.
So, that's whole thing. And yeah, just a solopreneur trying hard and grinding. Trying to use AI to get the best out of myself without changing how I communicate.
I have no intention on becoming an SaaS founder. I just wanted something for me. Thanks for reading this far!
I cold pitch coliving founders, retreat operators, wellness coaches and the like. These people are good at what they do and oftentimes have no idea why their marketing isn't working.
Reaching them through social outreach is just not going to happen. And while, cold outreach has a reputation for being spray-and-pray, I have been detrmined to make it the opposite of that. So I built something in Claude code with no skills whatsoever
Basically, it's a local Streamlit dashboard that runs the whole acquisition pipeline, start to finish.
Using coliving as an example, it finds the lead by starting with a region/country selector that pulls cities with coliving spaces from Nomad List. I pick a city, get a grid of one-click search links via a Google operator search, Airbnb, Booking.com, Maps, Instagram, Facebook, coliving.com.
From here I save it to a leads queue. Name, city, website, source. All stored locally.
When I have e few in there I do a mini audit of their site and socials. I feed the tool everything I can find:
* Raw HTML source of their homepage (Ctrl+U, paste the whole thing). BeautifulSoup strips the JS bundles, inline styles, SVG blobs and base64 images before it goes to Claude, so the token count stays reasonable.
* Then I voice note to Claude my own manual notes from browsing the site. What could be communicated better, what is the imagery like, is there a disconnect anywhere etc etc. These observations get weighted heavily — they override Claude's HTML read if there's overlap.
* I paste in their Instagram bio and recent captions. How active they are. What they post about. Do they have structure. Is there an obvious strategy?
* I paste in Google Tag Assistant output. What's actually firing at runtime. GA4, Meta Pixel, GTM, Hotjar. Or nothing. That's usually the finding!
* It also has an upload to view screenshots if there's a layout or formatting issue I want Claude to see.
Then I hit 'run audit' and it generates a personalised pitch in one pass.
The audit covers: SEO gaps, tracking setup, copy quality, messaging consistency between homepage and social, missing conversion elements. Then it splits cleanly into findings and a pitch.
The pitch is editable in a text area. I can tweak it before logging. Word count and character count live beneath it so I know what I'm sending.
The whole thing goes to Notion with one click via either "Pitch sent" or "Pitch pending" and it saves the business name, contact details, channel, ICP segment, findings, and pitch text.
It also sets a follow-up date automatically and the form clears for the next lead.
There's a chat panel underneath where I can drill into specific findings, ask for three alternative opening lines, get the pitch shortened, or ask what the single biggest gap is between how they present on Instagram vs the website.
So far I have completed 35 audits. Got 18 responses, booked eight 45-min calls and landed two clients.
After the actual call, I have a separate tab in the dashboard where I paste the Otter/Fireflies transcript, add any notes, paste in the original pitch for context and Claude extracts: what we discussed, what's working, what's broken (with root cause and impact), and three priorities in order.
Then it fills a duplicated Notion template page which is a summary of my findings and a bespoke Loom video with my my paid offer attached. All of the blocks are mapped by heading structure, callout colours, bullet positions. Date, Loom embed URL, all of it.
The hardest part was keeping my voice consistent. The first dozen versions kept generating shitloads of bloody em dashes, words like "genuinely", phrases like "lands well", you know the drill. It also had this awful habit of predicting which operators in a market will succeed and which won't which makes me sound like a prophet at best and a prick at worst.
The fix was a hardcoded constant injected into every generation prompt, before whatever the config says. That eliminated all the negative-to-positive reframing and AI filler words.
Cold pitching has stopped being something I dread. In fact, it is kinda fun. Every lead gets a specific audit, not a template. I know what's broken before I send anything. The pitch references real things and I often get feedback like "how did you know that!"
People have really responded to the specificity.
So, that's whole thing. And yeah, just a solopreneur trying hard and grinding. Trying to use AI to get the best out of myself without changing how I communicate.
I have no intention on becoming an SaaS founder. I just wanted something for me. Thanks for reading this far!
I cold pitch coliving founders, retreat operators, wellness coaches and the like. These people are good at what they do and oftentimes have no idea why their marketing isn't working.
Reaching them through social outreach is just not going to happen. And while, cold outreach has a reputation for being spray-and-pray, I have been detrmined to make it the opposite of that. So I built something in Claude code with no skills whatsoever
Basically, it's a local Streamlit dashboard that runs the whole acquisition pipeline, start to finish.
Using coliving as an example, it finds the lead by starting with a region/country selector that pulls cities with coliving spaces from Nomad List. I pick a city, get a grid of one-click search links via a Google operator search, Airbnb, Booking.com, Maps, Instagram, Facebook, coliving.com.
From here I save it to a leads queue. Name, city, website, source. All stored locally.
When I have e few in there I do a mini audit of their site and socials. I feed the tool everything I can find:
* Raw HTML source of their homepage (Ctrl+U, paste the whole thing). BeautifulSoup strips the JS bundles, inline styles, SVG blobs and base64 images before it goes to Claude, so the token count stays reasonable.
* Then I voice note to Claude my own manual notes from browsing the site. What could be communicated better, what is the imagery like, is there a disconnect anywhere etc etc. These observations get weighted heavily — they override Claude's HTML read if there's overlap.
* I paste in their Instagram bio and recent captions. How active they are. What they post about. Do they have structure. Is there an obvious strategy?
* I paste in Google Tag Assistant output. What's actually firing at runtime. GA4, Meta Pixel, GTM, Hotjar. Or nothing. That's usually the finding!
* It also has an upload to view screenshots if there's a layout or formatting issue I want Claude to see.
Then I hit 'run audit' and it generates a personalised pitch in one pass.
The audit covers: SEO gaps, tracking setup, copy quality, messaging consistency between homepage and social, missing conversion elements. Then it splits cleanly into findings and a pitch.
The pitch is editable in a text area. I can tweak it before logging. Word count and character count live beneath it so I know what I'm sending.
The whole thing goes to Notion with one click via either "Pitch sent" or "Pitch pending" and it saves the business name, contact details, channel, ICP segment, findings, and pitch text.
It also sets a follow-up date automatically and the form clears for the next lead.
There's a chat panel underneath where I can drill into specific findings, ask for three alternative opening lines, get the pitch shortened, or ask what the single biggest gap is between how they present on Instagram vs the website.
So far I have completed 35 audits. Got 18 responses, booked eight 45-min calls and landed two clients.
After the actual call, I have a separate tab in the dashboard where I paste the Otter/Fireflies transcript, add any notes, paste in the original pitch for context and Claude extracts: what we discussed, what's working, what's broken (with root cause and impact), and three priorities in order.
Then it fills a duplicated Notion template page which is a summary of my findings and a bespoke Loom video with my my paid offer attached. All of the blocks are mapped by heading structure, callout colours, bullet positions. Date, Loom embed URL, all of it.
The hardest part was keeping my voice consistent. The first dozen versions kept generating shitloads of bloody em dashes, words like "genuinely", phrases like "lands well", you know the drill. It also had this awful habit of predicting which operators in a market will succeed and which won't which makes me sound like a prophet at best and a prick at worst.
The fix was a hardcoded constant injected into every generation prompt, before whatever the config says. That eliminated all the negative-to-positive reframing and AI filler words.
Cold pitching has stopped being something I dread. In fact, it is kinda fun. Every lead gets a specific audit, not a template. I know what's broken before I send anything. The pitch references real things and I often get feedback like "how did you know that!"
People have really responded to the specificity.
So, that's whole thing. And yeah, just a solopreneur trying hard and grinding. Trying to use AI to get the best out of myself without changing how I communicate.
I have no intention on becoming an SaaS founder. I just wanted something for me. Thanks for reading this far!
I cold pitch coliving founders, retreat operators, wellness coaches and the like. These people are good at what they do and oftentimes have no idea why their marketing isn't working.
The overall goal was to stay true my voice but utilize Claude to speed up the process.
Waiting for inbound leads is not fun. Endless content creation is lame. And while, cold outreach has a reputation for being spray-and-pray, I have been determined to make it the opposite of that. So I built something in Claude code with no skills whatsoever.
Basically, it's a local Streamlit dashboard that runs the whole acquisition pipeline, start to finish.
Using coliving as an example, it finds the lead by starting with a region/country selector that pulls cities with coliving spaces from Nomad List. I pick a city, get a grid of one-click search links via a Google operator search, Airbnb, Booking.com, Maps, Instagram, Facebook, coliving.com.
From here I save it to a leads queue. Name, city, website, source. All stored locally.
When I have e few in there I do a mini audit of their site and socials. I feed the tool everything I can find:
Then I hit 'run audit' and it generates a personalised pitch in one pass.
The audit covers: SEO gaps, tracking setup, copy quality, messaging consistency between homepage and social, missing conversion elements. Then it splits cleanly into findings and a pitch.
The pitch is editable in a text area. I can tweak it before logging. Word count and character count live beneath it so I know what I'm sending.
The whole thing goes to Notion with one click via either "Pitch sent" or "Pitch pending" and it saves the business name, contact details, channel, ICP segment, findings, and pitch text.
It also sets a follow-up date automatically and the form clears for the next lead.
There's a chat panel underneath where I can drill into specific findings, ask for three alternative opening lines, get the pitch shortened, or ask what the single biggest gap is between how they present on Instagram vs the website.
So far I have completed 35 audits. Got 18 responses, booked eight 45-min calls and landed two clients.
After the actual call, I have a separate tab in the dashboard where I paste the Otter/Fireflies transcript, add any notes, paste in the original pitch for context and Claude extracts: what we discussed, what's working, what's broken (with root cause and impact), and three priorities in order.
Then it fills a duplicated Notion template page which is a summary of my findings and a bespoke Loom video with my my paid offer attached. All of the blocks are mapped by heading structure, callout colours, bullet positions. Date, Loom embed URL, all of it.
The hardest part was keeping my voice consistent. The first dozen versions kept generating shitloads of bloody em dashes, words like "genuinely", phrases like "lands well", you know the drill. It also had this awful habit of predicting which operators in a market will succeed and which won't which makes me sound like a prophet at best and a prick at worst.
The fix was a hardcoded constant injected into every generation prompt, before whatever the config says. That eliminated all the negative-to-positive reframing and pesky filler words.
Cold pitching has stopped being something I dread. In fact, it is kinda fun. Every lead gets a specific audit, not a template. I know what's broken before I send anything. The pitch references real things and I often get feedback like "how did you know that!"
People have really responded to the specificity.
So, that's whole thing. And yeah, just a solopreneur trying hard and grinding. Trying to use Claude to get the best out of myself without changing how I communicate.
I have no intention on becoming an SaaS founder. I just wanted something for me. Thanks for reading this far!
I cold pitch coliving founders, retreat operators, wellness coaches and the like. These people are good at what they do and oftentimes have no idea why their marketing isn't working.
The overall goal was to stay true my voice but utilize AI to speed up the process.
Waiting for inbound leads is not fun. And while, cold outreach has a reputation for being spray-and-pray, I have been determined to make it the opposite of that. So I built something in Claude code with no skills whatsoever.
Basically, it's a local Streamlit dashboard that runs the whole acquisition pipeline, start to finish.
Using coliving as an example, it finds the lead by starting with a region/country selector that pulls cities with coliving spaces from Nomad List. I pick a city, get a grid of one-click search links via a Google operator search, Airbnb, Booking.com, Maps, Instagram, Facebook, coliving.com.
From here I save it to a leads queue. Name, city, website, source. All stored locally.
When I have e few in there I do a mini audit of their site and socials. I feed the tool everything I can find:
Then I hit 'run audit' and it generates a personalised pitch in one pass.
The audit covers: SEO gaps, tracking setup, copy quality, messaging consistency between homepage and social, missing conversion elements. Then it splits cleanly into findings and a pitch.
The pitch is editable in a text area. I can tweak it before logging. Word count and character count live beneath it so I know what I'm sending.
The whole thing goes to Notion with one click via either "Pitch sent" or "Pitch pending" and it saves the business name, contact details, channel, ICP segment, findings, and pitch text.
It also sets a follow-up date automatically and the form clears for the next lead.
There's a chat panel underneath where I can drill into specific findings, ask for three alternative opening lines, get the pitch shortened, or ask what the single biggest gap is between how they present on Instagram vs the website.
So far I have completed 35 audits. Got 18 responses, booked eight 45-min calls and landed two clients.
After the actual call, I have a separate tab in the dashboard where I paste the Otter/Fireflies transcript, add any notes, paste in the original pitch for context and Claude extracts: what we discussed, what's working, what's broken (with root cause and impact), and three priorities in order.
Then it fills a duplicated Notion template page which is a summary of my findings and a bespoke Loom video with my my paid offer attached. All of the blocks are mapped by heading structure, callout colours, bullet positions. Date, Loom embed URL, all of it.
The hardest part was keeping my voice consistent. The first dozen versions kept generating shitloads of bloody em dashes, words like "genuinely", phrases like "lands well", you know the drill. It also had this awful habit of predicting which operators in a market will succeed and which won't which makes me sound like a prophet at best and a prick at worst.
The fix was a hardcoded constant injected into every generation prompt, before whatever the config says. That eliminated all the negative-to-positive reframing and AI filler words.
Cold pitching has stopped being something I dread. In fact, it is kinda fun. Every lead gets a specific audit, not a template. I know what's broken before I send anything. The pitch references real things and I often get feedback like "how did you know that!"
People have really responded to the specificity.
So, that's whole thing. And yeah, just a solopreneur trying hard and grinding. Trying to use AI to get the best out of myself without changing how I communicate.
I have no intention on becoming an SaaS founder. I just wanted something for me. Thanks for reading this far!
The emphasis here is 100% on invent.
For my next magic trick I will retrofit the justification around the tactic... abracadabra.
It is happening right at this very moment. Someone has 'found' a citation loophole, or a crawlable schema trick that gets picked up by whatever AI is currently hoovering the web, and within about forty eight hours there's a service offering on hidden in a question on a Reddit post.
Another fucking reason to pepper unsuspecting victims with a thing they didn't know existed until right now.
And I'm not ruling out that some of these 'tools' I see getting flogged might be useful. But does it actually move the needle much? Is it that important to materialize on Grokipedia or whatever we're calling it this week?
Here's something that's starting to irritate the fuck out of me..
There's an entire cottage industry forming around getting mentioned by AI. Schema markup this, structured data that, optimise your content so the machines will quote you like a primary source in someone's ChatGPT conversation. Everyone's very confident about how it works.
Oh so specific and sellable.
I reckon nobody actually knows how any of this works. The people selling the playbook don't know. The people buying it don't know. The AI companies themselves issue documentation that contradicts what their own systems demonstrably do.
We are all milling around stargazing at this unstoppable force that is AI, trying to get a handle on who knows what.
The competitor who's getting cited by Claude probably isn't doing anything you're not doing. They got there first, or they got lucky. Who knows. They may just have inbound links from a domain that aged well. You cannot buy your way into network effects that already calcified three years ago.
Just keep building. Keep grinding away as there is no substitute for turning over every rock and working hard. Word of mouth doesn't care if ChatGPT knows your name. A client who calls because someone they trust told them to call is not sitting there cross-referencing your AI visibility score.
I'm probably going to keep ignoring all of this and see what happens.
My guess is that what happens is roughly the same thing that happens when you ignore every other visibility fad, which is either nothing, or something unrelated.