The short answer to how to automate lead generation with AI is six steps: write down who you sell to, turn it into a search, score every match, find emails only for the fits, let an agent write from research, and put the search on a weekly schedule. Most answers stop at the writing, which leaves the hard part, deciding who should get a message at all, done by hand.
This guide walks each step in Oneprofile with the screens, the settings, and the credit price of each. Once it is set up, the chain repeats every week and asks you for one approval. The search half has its own page, Find leads, if you want the filters in depth.
To follow along on your own market, describe your buyer in a free workspace and check the match count before you spend a credit. A free workspace has 200 credits a month, 3 tables, and 100 rows per table.
How does AI lead generation work?
Lead generation has always been two jobs: finding people who might buy, and starting a conversation with them. AI changes who does the reading in between.
Three things happen, in this order:
Search. A plain-language description of your buyer becomes filters: a role, an industry, a location, a headcount range. Data sources supply the people who match.
Judgment. A model reads each match against your ideal customer profile and decides how well it fits, with reasons you can check.
Writing. Another model writes each message from what is known about that person and their company.
Automation is what strings the three together and repeats them. Any one of them, done by hand, is fine at twenty leads.
At two hundred a week, the step people drop is judgment. It is also the step that decides whether the rest works, and we will come back to it.
How to automate lead generation with AI: find and research
The first three steps build the list and decide who on it is worth writing to. Expect about twenty minutes, most of it spent reading what the AI drafted about your own company.
Write your ideal customer into business context
The search, the scorer, and the agent that writes your messages all read it. On Business context in Settings, type your domain into the AI Assistant card and click Generate. Oneprofile researches your company and drafts six sections, among them Ideal customer profile and Ideal buyer personas, in up to about 3 minutes.
Read the draft before you apply it. Choose Keep current or Use new section by section, then click Apply.
The scorer in step 3 returns nothing at all without this, on purpose. A score against a guessed customer is worse than no score.
Describe your buyer and add the matches
On Find leads, type who you want into the box at the top and press Enter. Something like "finance directors at US accounting firms with 20 to 200 employees" is enough. The AI Assistant opens the search with the filters filled in, and you can read and change each one.
Check the count first. Counting the matches is free, and nothing is added until you ask. A few dozen means the description is too narrow, and tens of thousands means you described a market rather than a buyer.
When it looks right, click Continue, then Save to a new table. Find work email arrives ticked on a people search. Untick it, since step 4 finds emails for the strong fits only.
Tick Enrich company instead, which adds each company's industry and a description, and click Create table. The button shows the price before you commit.
Score each lead against your ICP
From the table, click Tools, open the Agent tab, and add the ICP scorer from Built-in agents. In its Inputs, point Company domain, Company description, Industry, and Job title at the matching fields. It reads each lead against the profile and personas in your business context and returns three things:
ICP fit: strong (70 or above), possible (40 to 69) or poor (under 40)
Fit score: 0 to 100, calibrated to those bands
Fit reasons: the two or three facts that drove the decision, each naming where it came from
For an invented firm, Marlow Advisory (marlowadvisory.example), a strong result might read: fit strong, score 82, because the industry is accounting, the description names audit work for owner-run businesses, and the contact's title matches one of your buyer personas.
A lead with too little known about it comes back with no band and a list of what was missing. It never gets a false poor. The scorer costs 3 credits for each lead it scores, and the deeper method, with criteria drawn from closed-won accounts, is in the ICP scoring guide.
Ranking is the part most automated pipelines skip, and it is where the money goes.
"Playing the numbers game in outreach is expensive and wasteful. A better tool lifts your reply rate far more than sending more ever will."
Utku Zihnioglu, CEO of Oneprofile
How to automate lead generation with AI: write and send
Find and verify emails for the strong fits
Now spend on email, but only where it counts. Add Find work email from Tools, expand Run conditions in its editor, check Only run some rows, and add the condition ICP fit is strong. Leads that do not qualify are marked Skipped and never charged. On a thin market, widen it to strong or possible.
Find work email checks its data sources in an order set by accuracy, stops at the first answer, and costs up to 10 credits a lead. A work email lookup that finds nothing costs nothing.
Add Verify email beside it, at up to 0.8 credits an address, so every address carries a deliverability verdict before the first send. An automated pipeline sends more mail than a person would, and Google's email sender guidelines are the bar every domain sending to Gmail is held to. The bulk email finder guide covers catch-all addresses and what to do with them.
Filter the table to ICP fit is strong, sort by Fit score with the highest first, and save it as a view called Strong fits. The sequence enrols from that view.
Hand the writing and sending to a sequence
On Sequences, under Ways to start, pick the Email and LinkedIn campaign. It opens with three email prompts and a LinkedIn connection request after the first email, sent with no note and withdrawn if nobody accepts within 7 days.
On the Audience tab, click Choose an audience, pick your table and the Strong fits view, and click Use these rows. Under Columns we need from this table, check that Email address, Email deliverability, and LinkedIn profile URL are set. Contacts are reached in the order the view shows them, so your highest scores go first.
Every message comes from the sequence's outbound agent. It reads the step's prompt, the contact's data, your business context, and the thread so far. A new sequence starts with Web research and Person research on, so before the first message it reads the web about the person and their company, and their LinkedIn profile. One line under those switches prices the first message and each one after it.
Open the AI writing tab and click Write the emails to read one contact's whole sequence before anything sends. Looks good keeps it. Needs work, with one line on what's wrong, rewrites the thread and proposes a learning. Once you save it, every later contact follows it.
Three settings decide how hands-off this gets:
Send mode, on each sender's Settings tab. Send when confident, where new senders start, sends what the agent rated high and had enough data for, and holds the rest in Tasks.
LinkedIn run mode, on the same tab. A new connection starts on Manual run, where each LinkedIn step waits for Approve. On Automatic run, steps go on their own within the daily limits and the sending window.
Schedule, in the sequence's Settings. Optimized for deliverability is one preset, and Each sender's timezone keeps a New York sender on New York hours.
Email sends from Google Workspace mailboxes, connected one at a time or imported in bulk. LinkedIn steps run through the Oneprofile extension in the sender's own browser, which has to be open for them to run. The Outreach page covers mailboxes and step types in more depth.
How to automate lead generation with AI on a schedule
So far this ran once. The last step makes it repeat.
Put the search on a weekly schedule
Click Table settings at the bottom of the table and open Lead search schedule.
Switch from Manually to On a schedule.
Pick Weekly under Frequency, a time, and a start date, then click Save schedule.
In Table settings, turn on Run automatically on new and changed rows, so the scorer and the email lookups run on every lead the search adds.
Turn on New search matches in notifications to hear when a batch is waiting.
On the sequence's Audience tab, leave Keep in sync on. New reachable contacts in the Strong fits view join automatically, and anyone who leaves the view stops.
Each weekly run counts the people your table does not hold yet, prices them, and waits. The Lead search panel shows New matches waiting for your approval, with the price on the Approve button. That click is the only weekly job left for you on the finding side, which is the judgment point from earlier, moved to the one place where it costs you a second instead of an afternoon.
What runs on its own, and what waits for you
Full automation sounds better than it works. The setup above lets most of the chain run unattended and keeps a person at the points where a mistake is expensive.
Stage | Runs on its own | Waits for a person |
|---|---|---|
New search matches | Counted and priced on the schedule | Approve, before any are added or charged |
Scoring and email lookups | Every lead the search adds, once automatic runs are on | Nothing |
Email messages | Rated high with enough data, on Send when confident | Everything else, in Tasks |
LinkedIn steps | On Automatic run, inside limits and the window | Every step on Manual run, the default |
Replies | A reply from the contact stops their follow-ups | Every reply the agent drafts |
Opt-outs | Added to do not contact, every sequence stopped | Nothing |
Opt-outs belong on the automatic side. The FTC's CAN-SPAM compliance guide makes no exception for business-to-business email, and every commercial message needs a working way to opt out. A sequence's Unsubscribe footer setting adds the opt-out line, and a reply asking not to be contacted again, in any words, adds the address to your do-not-contact list and stops every sequence that person is in.
A tangent, since it comes up whenever AI and outbound share a sentence. Plenty of lead generation dashboards lead with messages sent. Sent is the cheapest number in outbound to grow and the least useful one to read. Replies are what count, and they are probably the number most of those dashboards bury.
What it costs to automate lead generation with AI each week
Every price below is per lead, from the published catalog, and each run shows its price before it starts.
Step | What runs | Price per lead |
|---|---|---|
Count the matches | Find leads search | Free |
Add the matches | Enrich company, ticked when you add them | Shown on Create table first |
Score | ICP scorer | 3 credits for each lead it scores |
Find email, strong fits only | Find work email | Up to 10 credits, nothing when none is found |
Verify | Verify email | Up to 0.8 credits |
Possible and poor fits | Skipped by the run condition | 0 credits for the email lookups |
Say the weekly search turns up 50 new people and 15 of them score strong. Adding them costs the price on the Approve button, and scoring all 50 is 150 credits on top. Emails and verification for the 15 strong fits are up to 162 more, and usually less, since a lookup that finds nothing costs nothing. Finding and verifying emails for all 50 before scoring would have been up to 540 credits, most of it on people you were never going to write to.
The agent's writing is priced separately, on the line under its research switches, and each draft is charged when its own step falls due. Team starts at $20 a month for 1,000 credits, Pro is $200 a month for 5,000, and extra credits are $20 per 1,000 on every paid plan. Free has 200 credits a month, 1 seat, up to 3 integrations, 3 tables, and 100 rows per table, enough to run the chain once on a sample. The personalized cold email guide goes further into verification verdicts and how the research shapes each opening line.
Start with one week. Describe your buyer, score the first batch, and read the Fit reasons on twenty leads you already know something about. If the reasons match what you would have said, put the search on its schedule. Score your first batch of leads free and judge the scoring on your own market.

