Apify collects public data, and Claude helps you qualify it and draft a message from a real observation. But a list does not create customers: start with 20 names and personal outreach, not thousands of rows.
You can use Apify to collect public business data and Claude to qualify the list and draft a message from a real observation. But that does not “get customers” by itself. A good system produces 20 suitable people to contact manually and learn from—not thousands of rows and generic messages that look like spam.
What do Apify and Claude do in lead generation?
Apify is a platform for running tools that collect publicly available data, and Claude is a model that can organise that data, read signals, and draft a message. Neither tool proves that a company needs your service, nor gives you permission to contact people. The real signal appears when you review the data, reach out respectfully, and measure responses.
What mistake does lead collection not solve?
We made this mistake ourselves. We collected 109 names for a website-building offer, sent no messages, and eventually deleted the list because that audience was no longer the buyer we wanted to serve. The problem was not that the list was too small; the customer definition was wrong and untested in real conversations.
Do not begin with “how many names can I pull?” Begin with: who has the problem today, and what public evidence suggests they are close to that problem?
How do you define an ICP before running an Actor?
An ideal customer profile (ICP) is a testable definition of the person or business you want to reach. Write a narrow definition that is enough to choose 20 names—not a broad label such as “store owners”.
- Sector: for example, private dental clinics or Shopify stores selling a specific product.
- Location: one city or market at the start.
- Size: an observable sign, such as branch count or a small team shown on the website.
- Signal: public evidence connecting the business to the problem you solve, such as a broken booking page, an unclear offer, or an ad that leads to a page without a clear outcome.
Do not guess whether a company has a Pixel, Conversions API, or a particular ad result from a public page. Write only what you saw, and treat everything else as a conversation question—not a fact in the message.
How do you collect an initial list with Apify?
Apify Store contains Actors that collect public data from sources such as business directories, websites, and company pages. Choose one source that matches your customer, and read its terms and the applicable law before you run it.
- Create an Apify account and choose an Actor suited to the source you need, such as a local business directory or company-website list.
- Set a small scope: one city, one sector, and one search term.
- Collect only the first 20 to 50 results. You do not need a thousand names to learn whether your definition is good.
- Export to CSV or Google Sheets, and keep the source URL beside every name.
Public data is not permission for mass outreach. Use published business contact details only where the law and source terms allow it, and honour opt-out requests immediately.
How do you clean the list before writing any message?
Every row without a clear reason to contact is a row that does not yet deserve a message. Clean the list manually before asking Claude to do anything:
- Remove duplicates and results that are not real businesses or are outside your chosen market.
- Remove any name you cannot connect to a real public signal.
- Record the source, review date, and observation in separate fields.
- Do not add an email address or phone number that is not publicly published, and do not guess or buy it from an unreliable source.
How do you use Claude to qualify a lead and write a personal message?
Claude should not invent a pain point for a company. Give it only documented data: the name, website or page URL, a visible observation, and your service. Ask it to separate fact, possible inference, and a question that still needs an answer.
Use this prompt for each lead:
“Read the data below. Write: 1) one verifiable visible observation, 2) one short question that opens a conversation without a sales pitch, 3) a reason not to send if there is not enough evidence. Do not invent numbers, tools, or results that are not stated.”
The best first message is: a real observation plus one question. It is not generic praise and not a complete sales offer.
Example: “I noticed your booking page asks visitors to choose a service before explaining the difference between the options. Are you seeing repeated questions before people book?”
How do you prioritise without a fake score?
You do not need a 1–10 score to begin. Put each name into one of three decisions:
- Send now: it matches the ICP and has a real public observation.
- Review later: it matches the ICP but does not yet have enough evidence.
- Do not send: it is outside the market or has no respectful reason for a conversation.
This prevents Claude from assigning false precision to incomplete data, and stops you from treating every name as an equal opportunity.
How do you send the first batch and measure it?
Send manually to only 20 people at first, following the platform rules you are using. Record the name, observation, message, and outcome. The first goal is not a sale from one message; it is learning whether the target customer responds to that observation at all.
- If nobody sees the message or reaches the offer, review the channel and contact method.
- If they see it but do not respond, review the signal and first message.
- If they respond but do not have the problem, review the ICP.
- If they reach a real conversation but do not buy, review the offer or price—do not automatically blame the data source.
What does this system not do?
- It does not prove demand. A list provides names; conversations and outcomes prove whether the problem is real.
- It does not justify mass messaging. Adding a name to a template does not make a message personal or permitted.
- It does not read private data. Work with publicly available information and comply with the source terms and laws that apply to you.
- It does not replace human follow-up. Claude speeds up preparation, but you are responsible for the observation, tone, and response.
Frequently asked questions
Does Apify find customers who are ready to buy?
No. Apify can collect public data according to the source and Actor you choose. A lead becomes an opportunity only when it matches your definition, has a real signal, and starts a respectful conversation that proves the problem exists.
Can Claude write a message for every company automatically?
It can prepare drafts from a structured list, but do not send them automatically before review. If the inputs are generic, the messages will be generic too. Review the observation and reason for contact for every person before sending.
Do I need to collect thousands of results?
No. Your first goal is to test the customer definition and message. Twenty personal conversations test a hypothesis more clearly than a thousand names that receive no message.
Can I use extracted emails or phone numbers?
That depends on the country, source terms, and data type. Use publicly published business details only where permitted, do not hide your identity, and offer a clear way to stop contact. This is not legal advice.
What should I do after the first 20 messages?
Review the results before expanding the list. Change one thing only: the customer definition, signal, message, or channel. Then test another small batch. Learning from real responses matters more than adding rows.
Next step
Do not start with thousands of leads. Choose one ICP, collect 20 names with real evidence, and send 20 personal messages. That is how you learn whether you need a better list, a better message, or a clearer offer—instead of burying your time in scraping that never becomes a conversation.