Skip to main content

AI lead generation: what works, what backfires, and a workflow

AI lead generation means using AI models to find, research, score, and contact potential buyers. Done well, it does the reading a good SDR would do on every account, at a scale no SDR team could match. Done badly, it sends confident, generic emails to thousands of people and burns your domains.

I build AI into lead generation systems every week. This is where it earns its keep and where it does not.

By Garrett Wolfe, founder of OneGTM. Updated .

Key takeaways

  • AI is strongest at reading: classifying companies from their websites, summarising accounts, and triaging replies.
  • It is weakest when it sends on its own. Keep a person approving what goes out, at least until a sequence has proven itself.
  • Use cheap models for classification and stronger ones for copy and judgment calls. The cost difference is large at volume.
  • Most AI lead generation tools wrap the same models. Building your own is realistic if someone technical owns it.
  • Measure AI work the same way as human work: positive replies and meetings by segment, not emails sent.

Where AI helps in lead generation

JobWhat AI doesWhy it works
Building the listReads each company website and answers yes or no questions about your ICPFirmographic filters miss what a company actually does. A model reading the site catches it.
Account researchSummarises recent news, hiring, and product changes into a few lines a rep can useSaves the 10 minutes per account nobody actually spends.
ScoringTurns those answers into a fit score, alongside the data you already haveConsistent judgment across 50,000 accounts, which a human team cannot give.
Signal detectionDecides whether a job post, filing, or news item matches a signal you definedUnstructured text is where the best buying signals hide.
DraftingWrites a first line or a short email from the research and the signalGood drafts, as long as a template carries most of the message.
Reply triageClassifies replies and drafts a response for a person to sendSpeed on positive replies matters more than anything else in the sequence.

Where AI lead generation backfires

  • Fully autonomous sending. Volume goes up, relevance goes down, and the first sign of trouble is a burned domain. More on this in AI SDR tools.
  • Made-up personalisation. A model asked to "personalise" with thin data will invent a detail. Ground every line in a fact you pulled, or do not write it.
  • Whole emails written from scratch. Fully generated emails drift and sound alike. A fixed template with two or three AI-written lines holds up better.
  • No verification step. AI does not make bad data good. Every email still needs verifying before it is used.

An AI lead generation workflow

StepAI doesA person doesTools
1. Source accountsClassifies each company from its websiteWrites the ICP questions and spot checks 50 resultsScrapers, data providers
2. ScoreAnswers ICP questions, outputs a 0 to 100 fit scoreChecks the score distribution against known customersAn LLM plus your CRM data
3. Watch signalsMatches raw events to your signal definitionsPicks the signals and reviews the first matchesSignal tools or your own feeds
4. Find peopleMaps the signal to a personaNothing, if the waterfall is set upEmail finders, verifiers
5. DraftWrites two or three lines inside a fixed templateApproves the first batches, then samplesAn LLM and your sequencer
6. Triage repliesClassifies and drafts a responseSends positive replies and books the meetingAn LLM, Slack, and the CRM

Which model for which job

Classification runs on every record, so it should be cheap: a small, fast model is enough to answer "does this company sell to hospitals" from a homepage. Copy and judgment calls (writing the lines a buyer reads, deciding whether a reply is a soft yes) are worth a stronger model. At OneGTM I run OpenAI first with Claude as the fallback, and pick the model size by task. Per-token prices and cost math are in best LLM for GTM.

AI lead generation tools vs. building your own

Most AI lead generation and AI SDR products combine a contact database, an LLM, and a sequencer behind one interface. That is quick to start and hard to tune: you rarely control the prompts, the data sources, or the sending infrastructure.

Building your own means wiring a data API, a model, and a sequencer together with Claude Code or a workflow tool. It needs someone technical, and it pays off at volume and when your ICP needs judgment a generic tool cannot make. The middle path is having it built once and run by your team, which is what a B2B lead generation agency like OneGTM does.

Measuring AI lead generation

Tag every lead with how it was sourced and every message with which lines AI wrote. Then compare positive reply rate and meetings against your human-written baseline, by segment. If AI-sourced lists book fewer meetings per hundred contacts, fix the ICP questions before you touch the copy.

AI lead generation FAQ

What is AI lead generation?
AI lead generation is the use of AI models to find, research, score, and contact potential buyers: classifying companies from their websites, summarising accounts, detecting buying signals in unstructured text, drafting outreach, and triaging replies.
Does AI lead generation work?
Yes for research, classification, scoring, and reply triage. Fully autonomous sending tends to produce generic messages at high volume, which hurts reply rates and deliverability. The best results come from AI doing the reading and a person approving what goes out.
What are the best AI lead generation tools?
It depends on the job. AI SDR platforms bundle data, an LLM, and sending. Workflow tools like Clay or n8n let you add AI steps to your own process. Teams with a technical owner often build their own with an LLM API, a data provider, and a sequencer.
Can AI replace SDRs for lead generation?
Not fully. AI can take over most of the research and list building and speed up drafting and reply handling. Conversations, judgment on edge cases, and booking qualified meetings still benefit from a person.
Which AI model is best for lead generation?
Use a small, cheap model for high-volume classification and a stronger model for copy and judgment calls such as reading replies. The right split saves a lot at volume without hurting quality.
Will AI-written emails hurt deliverability?
AI text itself is not the problem. Sending too much, to unverified addresses, with near-identical messages is. Keep volume per mailbox modest, verify every address, and use a fixed template with a few AI-written lines.

Want AI built into lead generation without burning your domains?

Related

  • AI SDR alternatives: Custom signal-driven systems to use instead of a packaged AI SDR, with the same published proof.
  • AI SDR tools and AI sales agents: 11x, Artisan, AiSDR, Unify, Qualified and more with published pricing, where AI SDRs fail, and when to build your own.
  • AI for sales: Account research, follow-up, and CRM hygiene handled by AI your reps will use.
  • B2B lead generation agency: Lead gen systems you own instead of one-off lists.
  • Best LLM for GTM: Which AI model OneGTM runs for classification, scoring, research, copy and replies, with per-token prices and cost math.
  • Buying signals guide: The 15 B2B buying signals worth tracking, how to detect each one, and how to turn a signal into outreach.
  • Best lead generation companies: B2B lead generation firms compared by model, with published prices.
  • AI for GTM: How AI GTM works across sales, marketing, and RevOps teams.

Talk to OneGTM about AI lead generation

Tell me what you're building. I reply the same day.

Call with Garrett

Takes about 20 seconds. Times show up right here, and the call is with Garrett directly. See our privacy policy.

Systems we've built for

Rather do it the old way? team@getaob.com LinkedIn