How to Automate Lead Generation (Full Guide)

by the end of this guide you'll know how to generate leads on autopilot for anything. a digital product you sell online, a business generating leads for itself. or someone with no business at all, generating leads and selling them to a business.

it's really just one method. you find who's already paying, you pull the data on them, you build the thing that reaches them, and you place it where they'll see it. the only thing that changes is who ends up with the lead.

by the end of this you'll have:

• an agent that knows exactly who buys your product
• a buyer profile from three months of their own tone
• ad creatives built from that profile, several per angle
• a whop ads campaign that sends leads to your product
• a scanner that finds who has the thing a trade fixes
• a lead pipeline a local business pays for monthly
• all of it on a schedule, with just a couple things kept manual

which agent? grok bot or hermes

both can run everything in this article, and the system is the same either way. the difference is how much setup you want to do.

grok bot is easy to setup. it has its own computer already, with a terminal, a browser and a filesystem, and you talk to it like a chat. so the way this works is you build the whole system as a skill file first, with claude, and then you hand that file to grok bot along with the api keys and the tool access it needs. it reads the skill and runs it. you can build the skill with chatgpt or with grok bot itself, but claude is what i use for it anyway.

that means a casual can run this. you don't set anything up you just give grok bot a file and a few keys and it starts running.

hermes is more technical to start and easier once it's running. it runs on a schedule, which grok bot can do too but hermes is built for it. the hermes setup is just one prompt into claude. you give it the hermes docs and a telegram bot token from @BotFather, and it provisions hermes on railway as a service that stays up. then you give hermes the same skill file, the same api keys and the same tool access, and from there the prompts go to it on telegram.

so the skill file is the main thing. build it with claude, and either agent runs it. every prompt in this article is what goes into that skill file.

from here on i'll say the agent and mean either.

the skill file

a skill file is basically just a document that tells an agent how to do one job. what the steps are, what order they go in, what tools to use, and what to log. you write it once and any agent that reads it can run the job.

you build it with claude (you can use codex too). you paste the prompts from this article into a new chat, tell claude which pipeline you're running, and ask it to turn them into a single skill file with the rules at the top and the steps below. it comes back as a .md file you save and hand to grok bot or hermes.

1turn these into a skill file for [pipeline one / two / three].
2
3- here are the prompts: [paste them]
4- put the operating rules at the top. one table per trade or niche, never read across tables, stop and ask before anything that contacts a real person or spends money, mark anything uncertain as review rather than guessing
5- then the steps in order, each one saying what tool it uses, what it reads, what it writes, and what it hands to the next step
6- list every api key and tool access the agent needs to run this, so i can give them all at once
7- write it so an agent that has never seen this article can run it from the file alone
8- tell me anything in here you think is underspecified and would make an agent guess

which half is yours

there are three types of people this is written for, and each one runs a different part of it.

if you sell a digital product, or want to, you run pipeline one. the lead is yours, the ad is yours, the buyer lands on your page.

if you own a business, you run pipeline two for yourself. you find the people who need what you do, you show them the fix on their own property, and the leads come to you. skip pipeline three if you are the business owner.

if you have no product and no business, you run pipeline two into pipeline three. you generate the leads for a trade in a city, and you sell them to the business that does the job.

pipeline 1, digital leads

this is if you have a product, or an idea for one, you find out exactly who buys it and where they are, you pull everything they've said for three months, and you build the ads from that. (you don't need to worry if you didn't run a single ad before because it's easy to run them with whop ads now)

start with the product

the agent needs to understand the product before it can find the buyer/customer. what it does, what it costs, what problem it solves, and what someone would have to believe about themselves to want it.

so the first prompt is the product, and what comes back is the buyer/customer. who they are demographically, and who they are psychographically. for ex: a 34 year old man in texas is a demographic. a 34 year old man who's tried different things and is depressed none of them worked is a psychographic.

1here's my product: [describe it, or paste the page].
2
3- tell me who buys this. age, situation, what they do, where they are in life
4- tell me who they are underneath that. what they've tried, what they're afraid of, what they'd never admit wanting, what they tell themselves about why it hasn't worked yet
5- tell me where these people are online. the specific subreddits, facebook groups, youtube channels, tiktok accounts, discords, forums. not the category, the actual names
6- for each place, tell me why they're there and what they talk about when they're there
7- tell me which of these places has the buyers and which has people who just talk about it
8- if you're not sure a place is real or active, say so rather than listing it

pull the three months data

now the agent goes to those places and pulls everything from the last three months. posts, comments, replies, the questions people ask, the things they complain about, the words they use. apify has a scraper for every one of those platforms and the agent runs them through the apify mcp.

three months is enough to see what repeats and not so much that it's out of date.

1scrape the last three months from every place you listed as having buyers.
2
3- use the apify mcp. reddit, facebook groups, youtube comments, tiktok comments, whichever scraper fits each place
4- pull posts, top comments and replies. keep the source url, the date and the engagement on each
5- write down every question that comes up more than three times, in their own words, with a count
6- write down every complaint about what they've tried, in their own words, with a count
7- write down the phrases they use to describe the problem, and the phrases they use to describe what they want instead
8- tell me what they've already paid for and what they say about it
9- write it as one profile with the quotes and counts in it, sorted by count
10- mark anything you're guessing rather than quoting

the agent's job is to find what's most common.

the creative plan

the profile becomes a plan before we build out the ads. the agent reads what's in there and works out the angles, which is the complaint or the want with the highest count, and the hooks under each angle, which are the exact phrases from the profile that say it.

then it works out what whop ads needs. whop ads runs on meta, facebook and instagram, so the creatives are vertical video and stills in the three aspect ratios meta uses. the agent knows that from reading whop's docs, and the plan says which creative goes where.

1build the creative plan from the profile.
2
3- take the three complaints or wants with the highest counts and make each one an angle
4- for each angle, write three hooks using the exact phrases from the profile, their words not mine
5- for each hook, say what the video shows and what the still shows. the outcome they described wanting, not the product
6- read whop's ads docs and tell me what formats and sizes meta needs through whop ads
7- for each creative, say which placement it's for. feed, stories, reels
8- every creative ends with the same offer and the same checkout link
9- if an angle doesn't have enough in the profile to write three hooks from, drop it and say why rather than padding it

nine creatives from three angles. enough to learn from, not so many the budget splits into nothing. the agent generates them through replicate or fal, whichever you've given it a key for. right now gpt image 2.5 is the best image model and seedance 2.5 is the best video model, so those are the two it calls, and both come out of the plan without another prompt.

run whop ads

whop ads runs on whop's own agency accounts on meta. you launch from the whop dashboard, not the native ads manager, and the campaigns run on facebook and instagram through accounts that sit at meta's top tier. that means priority in the auction, lower cost per thousand at scale, and far fewer rejections than a fresh account gets. which is really cool.

the pixel is already on your whop page and it attributes from real payment data. so when a conversion is counted, it's because someone paid. whop also forwards conversions to meta's optimisation, so meta learns from the same numbers.

the agent sets this up from the plan. one campaign with three ad sets, one per angle, three creatives in each. start at [placeholder] a day per ad set. the number to watch is cost per sale against the price of your product. if an ad set's cost per sale is above the product price for three days, pause it. the agent pulls spend from the ads dashboard and net revenue from whop every day and puts them side by side, which is the only way to actually see that.

funding can come from your pending whop balance the same day. the ads bring buyers and the buyers fund the ads.

the daily report from the agent is four numbers per ad set. spend, clicks, sales, cost per sale. and which angle is winning and whether anything should pause.

there's nothing to pay to use whop ads. you pay for the spend and that's it.

collect the leads

the ad sends them to your page, and the product is the one you gave the agent in the first prompt. the page says what the ad said, in their own tone.

the agent follows up with the ones who clicked and didn't buy. whop shows you who visited and who checked out.

that's pipeline one. product, buyer, three months of what they say, plan, creatives, ad. the lead was always yours.

pipeline 2, physical leads

this is the version where the lead is for a local business. you find a property or a business with a visible problem from public data, render the fix onto that actual property, get that render in front of the owner, and whoever responds is the lead.

pick the trade

one-word trades everyone understands. solar, roofing, landscaping, pool, awnings, fencing, driveways, signage. the buyer has to already want the thing before the message arrives, which is why the trade has to be one where the property or the business itself tells you they need it. you're not creating demand but just finding it.

some of these are homeowners and some are businesses, and the method doesn't care. a house with a pool is a homeowner. a cafe with no sign is a business. a warehouse with a flat empty roof is a business. a house with an old roof is a homeowner.

what matters is that the thing the trade fixes is visible from public data, so the agent can find it.

so the first prompt is the trade, and what comes back is where the data is.

1i'm generating leads for [trade] in [city].
2
3- tell me what the property or the business looks like when it needs this trade. what's visible, from where
4- tell me which public data source shows that. satellite imagery, street view, listing photos, business listings, permit records, property records. the actual source, and what it returns
5- tell me whether the buyer is a homeowner or a business, and what contact data i'll have for each. address only, or email and phone too
6- tell me what one qualifier i should use. one thing, not a stack
7- tell me what the render needs to show, and which image the render goes into
8- if this trade doesn't have a visible signal in public data, say so rather than inventing one

detect the candidates

the qualifier is just one thing. has the thing the trade fixes. every filter you stack on top cuts the list and adds a guess, and the guesses compound.

the trade will tell you which of these are worth it once the first batch goes out, and that's really valuable information to have.

the data is public and most of it is free. the source depends on the trade, which is why the previous prompt exists. the agent picks the one source the first prompt named and reads that. adding a second source to the qualifier is how the list shrinks.

most imagery apis have a free metadata check that tells you whether imagery exists and how old it is before you pay for the image. the agent runs that first on every address and only pays for the ones that come back recent.

1build the candidate list for [trade] in [city].
2
3- pull every address in [area] from the property records or the business listings, whichever the trade needs, with the owner name where it's public
4- for each one, check whether imagery or a listing photo exists before pulling anything, and drop anything with nothing or with imagery older than three years
5- for the ones that pass, pull the one source we chose and check it for the one qualifier
6- for the ones that pass the qualifier, pull the reference images the render will need
7- write a table with address, owner, contact data, source urls, and a flag for whether the qualifier is clear or borderline
8- mark borderline as review rather than in or out
9- tell me what the batch cost in api calls when you're done

render the fix

the image model puts the fix onto their actual property.

it has to be an edit. the model gets the reference images and it changes only the thing the trade fixes.

several reference frames matter here for the same reason they did in my christmas lights article. one image leaves the model guessing at everything outside it. three or four references and it has enough to work from.

reach the owner

you can do this with three channels. email, sms, postcard. the render is the same in all three, and so is where it points, which is a page built for that one property with the render at the top and a form asking when they'd like a quote.

which channel depends on what you have. for a business you usually have an email and a phone from the listing, so email and sms both work and the agent runs them through whatever sender you've given it. f

or a homeowner you usually have an address and nothing else, so it's post. in my own tests the postcard has outperformed email and sms by a wide margin anyway.

for post, there are multiple postcard companies with api's that you can give your agent access to. the agent sends the batch itself once you've approved it.

1prepare the outreach batch for [city], [trade].
2
3- for each property in the table, pick the channel from what we have. address only means post. email or phone means email or sms as well
4- the render goes with every message. for post it's full bleed on the front, for email it's the image, for sms it's the link to the page
5- one line with it, in plain words, saying this is their property with the fix done and where to see what it takes
6- build one page per property with the render at the top, the address, and a form asking when they'd like a quote. every channel points there
7- for post, the qr corner stays blank in the artwork, i composite the real one in
8- validate every address, email and number before it's in the batch
9- prepare the batch and stop. show me the count per channel and one sample of each

the owner opens the page, sees their place again, fills in a time, and that form submission is timestamped and tied to their address.

capture the response

the qr scan, the form fill, the reply. that's the lead. it has an address, an owner name, a time they want a quote, and a render of the property. the agent logs it to the table the moment it lands, and the table is what you sell in the next pipeline.

an open or a scan without a form fill is still worth logging. someone who looked at their own property with the fix on it and didn't ask for a quote yet is warm, and the second batch follows up with them.

if you're the business owner running this for yourself, this is where it ends. the agent has done the finding, the rendering and the sending, and the only thing left is the conversation which is handled by you.

that's pipeline two. trade, detect, render, reach, capture.

pipeline 3, selling the leads

now the lead goes to the business that does the work, and you get paid for it.

find the businesses

google places, the trade, the city. filter to owner-led companies rather than franchises, because the owner is the one who reads the post, between 15 and 200 reviews, a real address etc.,

the agent pulls the list, and for each one it notes what area they cover.

reach the business

you don't really pitch. the message to the business has a sample of the leads already generated in their area on it. the owner sees finished leads, and the question in their head is how many more of these are there, not whether this works. post if you have the address, email if you have that, do both if you have both.

1prepare the outreach to [trade] companies in [city].
2
3- pull every [trade] company from google places in [city] with 15 to 200 reviews and no franchise markers, and note the area each one covers
4- for each one, pick the three closest leads from the table that haven't been sold yet
5- build a one page pdf with those three: the street and area, the render, and the form fill with the time they asked for
6- write the message in plain words: here are three people near you asking for a [trade] quote this week. here they are
7- the link or the qr goes to the pdf
8- draft one follow-up email for a week later with the same three
9- prepare everything and stop. i send the first message to any owner who replies
10- if a company has fewer than three leads within their coverage area, skip them and say so rather than sending leads across town

price it

per lead or a monthly retainer for the pipeline.

per lead is for the first month. the business pays for what they got, they see the leads close, and they learn what one is worth to them. the number sits against the trade's average ticket.

the retainer is for month two. the business has seen the leads close and would rather pay a flat number for the pipeline. it's exclusivity in a set of zip codes for [placeholder] a month, and the agent keeps the pipeline running in those zips for them alone.

deliver

leads land in a dashboard the business can see. address, owner, render, form fill, timestamp, status.

what keeps them paying is the dashboard filling up without them doing anything, and the first job that closes from it.

what the agent runs

across all three pipelines the agent does the finding, the reading, the profile, the creatives, the property scan, the renders, the page builds, the outreach batches, the sheet, and the daily report. that's most of it.

three things stay with you.

checking the renders before a batch goes out, the first conversation with a business owner, the budget on any ad. the agent recommends, you set it.

what breaks

public imagery is sometimes old in places. the image the agent reads might be three years out, and the thing it sees has been fixed already.

the metadata check tells you the date on some sources and not others, so the first batch in any new area is where you find out how stale it is. look at a few before you trust it.

whop ads rejects creative that makes claims. no invented income numbers, no fake testimonials, nothing that implies a result the buyer will get. the compliance check runs before launch and a rejected ad delays the whole set. so keep this in mind.

business owners usually don't reply to the first message. they reply to the second, or to the follow up a week later. at least two follow-ups is recommended.

where to start

pick one pipeline. if you have a product, pipeline one. if you don't, pipeline two into three.

look at every render before the first batch goes out. the ones that are wrong tell you what to add to the render prompt.

then the second batch, same city, with what you learned.

hope you have fun setting this up.

join here for more value: https://t.me/+pbCBBtUEtu1lZDA1