AI lead generation: what it can and can't do

Finding is where AI saves the most time, because the alternative is reading hundreds of posts or pages by hand. A model can read a public post and tell you whether the author is asking for a recommendation, complaining, or selling. That's the job HearBuyers does on X and Reddit.

It does well at reading large volumes of text and sorting by meaning rather than keywords. It does badly at understanding your market's slang without examples, and at knowing what you consider a good fit.

Finding also depends on what the tool can see. A tool that reads public posts sees public posts. It can't see private messages, closed groups or anything that isn't published, and Reddit's public data access is changing between 31 October 2026 and March 2027, so coverage may shift. Ask any vendor what they can read today, not what they could read last year.

Enrichment adds details to a name: employer, title, email. Some tools combine many data providers and try them in turn. Clay's home page, for instance, describes enrichment from 150+ providers (checked 17 October 2026). Apollo bundles contact data with sequences and charges in credits (its pricing page, 17 October 2026).

Enrichment is good at saving lookup work. It is bad at being right. Data goes out of date as people change jobs, and a model asked to fill a gap may guess. If a tool returns an email, ask how it was verified. If it can't say, treat the email as unconfirmed.

Scoring turns judgment into a number. Good scoring tells you why. Bad scoring gives a bare 87 that you can't argue with.

It does well at applying the same standard to hundreds of items without tiring. It does badly when the score is trained on the wrong signal. A model that rewards keyword matches will score a job ad as high as a buyer.

HearBuyers gives each public post a 0 to 100 score for buying intent and highlights the sentence that triggered it, with a one-line reason. You can try that on any sentence in the buying intent scorer. Here are three samples:

• Sample, score 90. "Looking for a scheduling tool for a six-person clinic. We switch at the start of next month." A specific job, a size, a date.
• Sample, score 40. "What scheduling tools do people use for small teams?" General interest.
• Sample, score 5. "We just launched our scheduling tool! Try it free." A vendor's announcement.

Drafting is the most tempting part and the most dangerous. A model can write a polite reply in seconds. It can also write one that gets you removed from a community, or a message that sounds like every other automated message in the inbox.

It does well at a first draft you can edit, written in a consistent voice. It does badly at real personalisation. If the draft mentions something specific about the person, check that it's true. Models invent details, and a wrong detail in a cold message is worse than none.

Some tools send for you. Buska's Scale plan lists an AI agent that drafts outreach, and its pricing table says you approve it (its page, 17 October 2026). Whether to let anything send without a human is your decision, and we recommend against it. HearBuyers never posts or sends on its own: a person edits, approves and posts every reply.

• You have to decide your ideal customer. A model can propose one, but the choice is yours.
• It writes what it infers from your website, so it can't know your offer's real strengths. If the site is vague, the replies will be too.
• It doesn't replace relationships. The first reply gets you a conversation, and what happens after depends on how you handle it.
• It can't guarantee compliance. Communities have rules about promotion and platforms have rules about automation, and you are responsible for following them.
• It can't make a weak offer work. Better targeting of a product people don't want only finds more people who say no.

1. Mass automated messages teach people and platforms to ignore you. On Reddit, Rule 2 of the site rules asks for authentic participation and bans spam and content manipulation.
2. Wrong names and addresses cost trust.
3. If you can't see why someone scored high, you can't improve the filter.
4. Access to data on social platforms changes, so a tool built on one source can lose it.
5. If you stop reading the posts yourself, you lose the instinct that tells you the tool is wrong.

Ask these before you pay:

1. Which sources does it read, and which are public posts versus purchased databases?
2. Does each score come with a reason and the evidence behind it?
3. Who sends? Does a person approve before anything leaves?
4. How does it handle community rules and platform rules?
5. Is it priced per seat, per credit or per result? What happens if it finds nothing?
6. What can I export, and can I leave?

Our guide to the best AI lead generation tools compares named products against these questions, and the best AI SDR tools list focuses on tools that write and send.

1. Write down the sentence a buyer would post when they need you, using the real wording from your last five customers.
2. Run a tool on that sentence for one week and read every result it flags.
3. Count how many were real. If fewer than a third were, tighten the filter before adding volume.
4. Reply by hand to the best ones. Learn what works before you automate any step.

For the general craft of finding buyers, see prospecting and sales signals.

It's worth it when it saves reading time and keeps a human on the final step. It isn't worth it when it promises leads without showing you where they came from.

Yes, for public posts. A tool can read posts, score the ones that ask for what you sell and draft a reply. It can't read private messages or posts that were never public.

Some tools aim to. Whether that works depends on your market and your tolerance for errors. This guide's view is that AI works best as an assistant for research and drafting, with a person deciding what goes out.

Only if you accept the risk of errors going out under your name. Many teams choose to approve every message.

Pick one source where your buyers ask for help and write the sentence they'd post. Then test it against five real posts by hand. If you can't tell which ones are buyers, a tool won't be able to either.