AI Lead Generation for B2B: Build a Weekly Prospect List

AI lead generation for B2B means using AI to find companies that fit your best customers, find the right person, and write down a real reason to contact them. For a small team, run it as a weekly list of about 25 checked names.
Most pages on this topic are tool roundups. This guide covers the process: where names come from, a seven-step setup and the rules that can get you in trouble. For help building it with your team, see my AI sales enablement page.
What is AI lead generation for B2B?
IBM describes AI lead generation as using AI tools to help businesses “find high-quality prospects fast and efficiently” (IBM). In practice it is four jobs that software now does faster than a person: find companies, find the contact, rank the fit, and draft the message. The pages that rank for this term mostly come from software vendors, and many of them sell a platform that bundles these jobs. You do not have to buy that bundle to get the benefit.
| Job | What AI does | What a person still checks |
|---|---|---|
| Find companies | Reads directories, job posts, news and company sites and lists the ones that match your description | That the company really fits, and the page the AI used says what it claims |
| Find the contact | Suggests a role and a name, and sometimes an email | That the person exists, holds that role, and the email is real |
| Rank the fit | Scores each company against your rubric, with reasons | That the score matches how your past deals went |
| Draft the message | Writes a first email from the facts it found | That every fact is true and the tone sounds like your rep |
Ranking is its own topic. I cover it in AI lead scoring, and the message in AI sales emails. This post is about the first two jobs and about joining all four into one weekly habit.
What is a weekly prospect list?
A weekly prospect list is a short, checked sheet of new companies to contact this week. Each row has a company, a person, one reason to reach out now, the link that proves the reason, a fit score and a status. It replaces the huge exported list that nobody finishes.
Here is the layout I would start with:
| Column | Example of what goes in it |
|---|---|
| Company and website | Name, link |
| Person and role | The buyer, plus one backup role |
| Why now | One fact: a new hire, a product launch, a job post, a complaint about a competitor |
| Proof link | The page that shows the “why now” fact |
| Fit score | 0 to 100 from your rubric, with the reasons |
| Email check | Verified, risky or not found |
| Status | New, approved, sent, replied, meeting booked |
The “proof link” column is the one that keeps AI honest. If the AI cannot give you a page address for a fact, the fact does not go on the sheet.
I cold called for over two years, and it taught me to have no fear of picking up the phone. I think the same holds for outreach: a call or an email goes better when you have one real reason for it. The “why now” column is that reason, written down before anyone dials or types.
How do you build a weekly prospect list in seven steps?
Run these in order. The counts and the time limits are my suggestions, not an industry standard.
- Write your ideal customer in six lines. Company size, industry, location, the buyer’s role, one problem you solve and one disqualifier. Base it on your last 20 closed deals, not on who you wish you sold to.
- Pick two or three sources. The next section lists them. Start with fewer, because you have to check each one by hand.
- Collect candidates. Paste the text of each company’s own pages into an AI assistant, or have it search, and ask it to fill one row per company. Tell it to write “not found” when a fact is missing.
- Score the fit. Apply the rubric from AI lead scoring. Keep the reasons next to each number.
- Check the contacts. Confirm the person and role on the company site or their public profile. Run any email address through an email verification service before you use it.
- Review by hand in 15 minutes. Read each row. Cross out anything with no proof link or a weak reason. Approve the rest.
- Hand the approved rows to outreach, then log what happens. Send with the process in AI sales emails. Mark each row replied, meeting or nothing, so next week’s list gets better.
A prompt for step 3 can look like this:
Here is my ideal customer in six lines. Below is the text from 10 company websites. For each company, say whether it fits, give the one fact that makes it a fit, and paste the page address where you found that fact. Name the most likely buyer role. If a fact is not on the page, write “not found”. Do not guess.
The last two sentences do the real work. An AI that is allowed to fill gaps will invent a funding round, a hire or a quote. A page address next to every fact lets you check it in ten seconds.
If someone does the same task over and over every day, that task can probably be automated. Building this list is that kind of task. Run steps 3 and 4 by hand for the first two or three weeks. When you can predict what the AI will do, you can move them into a saved prompt or a scheduled routine. Keep step 6 for a person.
I run my own routines this way. In my video on Claude agents, I schedule routines that check my email, draft replies and look for new leads, and each one only writes an output file instead of messaging me, so I read one summary of what needs my attention.
Where should B2B leads come from?
Use sources that show a reason to buy right now, and that you can check yourself. Your own past leads are usually the cheapest and the most accurate.
| Source | What it tells you | Watch out for |
|---|---|---|
| Your closed-won customers | Which companies look like your best buyers, so you can find lookalikes | Only finds more of the same kind of buyer |
| Your closed-lost and quiet leads | People who already know you and may be ready now | Needs honest notes on why the deal stalled |
| Job posts | A company hiring for the exact role or problem you solve | A post does not prove budget |
| News and funding announcements | A change, such as a launch, a new executive or a new office | Often public to all your competitors too |
| Competitor reviews | Customers who complain about a rival’s weak spots | Reply with help, not an attack |
| Directories and trade lists | A wide, cheap list of companies in one niche | Out-of-date entries and thin detail |
| Your own website visitors | People who looked at your pages | Only works if you have enough traffic and a legal way to identify them |
I read all my competitors’ one-star reviews. Every complaint was ammunition: it showed me exactly how we could do things better. For lead generation, the same pages are a list of people who already want an alternative. That is a better “why now” than “I saw your company on a list.”
Be careful with LinkedIn. LinkedIn’s help page says it does not allow third-party software such as bots or browser extensions that scrape or automate activity on the site, and that accounts using them risk being restricted or shut down. Read profiles and posts yourself, or use a data provider that licenses its data, and do not point a scraper at your own account.
Which AI tools do this work?
The tool matters less than the sources and the review step, so choose by the job you need done. These are categories, not a ranking.
| Option | Best for | Limit |
|---|---|---|
| A general AI assistant (Claude, ChatGPT) with your rubric | Reading company pages, filling the sheet, writing the reasons | Does not give you verified emails, and can invent details if you let it |
| A contact database (Apollo and ZoomInfo are examples that a Gumloop roundup lists) | Names, roles and emails in bulk | Data can be stale, and you pay for volume |
| A workflow or enrichment tool (Clay is one example) | Joining several data sources into one table with AI steps | A bigger setup, more to learn and more to check |
| Your CRM’s built-in AI | Keeping the list next to your deal history | Usually limited to the contacts you already have |
Start with the first row and a spreadsheet. Add a database or a workflow tool only after you know which columns you actually use. Before pasting real customer details into any AI tool, check your company’s data policy and the tool’s privacy settings.
What rules limit AI outreach?
Three sets of rules matter most for a small US team, and AI does not change any of them. This is not legal advice.
- CAN-SPAM covers business email. The FTC says the law makes no exception for business-to-business email. Your headers and subject lines must not mislead, you need a valid physical address, and you must tell people how to opt out and honor opt-outs within 10 business days. Each separate email in violation can carry a penalty of up to $53,088, and hiring another company to send does not remove your responsibility. Source: FTC, read November 18, 2026.
- Inbox providers set their own limits. Google’s sender guidelines say to keep your spam rate in Postmaster Tools below 0.10% and to avoid ever reaching 0.30%. Senders of 5,000 or more messages a day to Gmail accounts must also authenticate with SPF, DKIM and DMARC and offer one-click unsubscribe for marketing mail. A weekly list of 25 sits far below that volume, but the spam-rate lesson applies to everyone. Source: Google Workspace Admin Help, read November 18, 2026.
- Platforms have their own terms. The LinkedIn rule above is one. Read the terms of any site you pull data from.
If you sell into Canada, the UK or the EU, rules differ and are often stricter. Check them before you send there.
What goes wrong with AI lead generation?
Five problems come up again and again.
- Invented facts. An AI that fills gaps writes a believable lie. The proof-link column and the “write not found” rule exist for this.
- Stale data. People change jobs. Confirm the role and run emails through a verifier the week you send, not the week you collected them.
- Volume over fit. At my last company we did a million things, but only two or three of them brought in most of the results. Lead generation is the same. Two sources that book meetings beat ten that fill a sheet.
- Nobody reviews. If step 6 gets skipped, the AI’s mistakes reach your prospects. Put the review on the calendar, not on a to-do list.
- No feedback. If you never log who replied, the list cannot improve. A lead source is a guess until you count what it produced.
How do you know it is working?
Measure meetings, not names. Track a few numbers each week, then judge each source after four to six weeks (my suggested trial length).
| Metric | What it tells you |
|---|---|
| Replies per source | Which sources reach people who care |
| Meetings booked per source | Which sources produce real pipeline |
| Rows rejected at review, and why | Whether the AI’s research needs tighter rules |
| Minutes spent per approved row | Whether the process saves time compared to doing it by hand |
Drop the sources that produce nothing and give their time to the ones that do. To decide whether the time saved is worth it, see how to measure AI ROI, and to run the first month as a small trial with a stop rule, see how to run an AI pilot project.
What should you do next?
This week, do three things. Write your ideal customer in six lines from your last 20 deals. Pick two sources from the table. Build the first sheet of 25 rows, review it by hand, and send only the rows with a proof link and a real reason.
If you are still deciding where else AI belongs in sales, start with the guide to AI use cases. Once the list is working, AI call auditing shows how to check what happens after a prospect says yes to a call.
If you want a person to design the process, build the sheet with your data and train your team to run it, start on my AI sales enablement page or my AI consulting page.
Related guides
- AI lead scoring: how a small sales team can set it up
- AI sales emails: prospecting and follow-up in the rep’s voice
- AI call auditing: score every sales call against your playbook
Sources
- Federal Trade Commission, CAN-SPAM Act: A Compliance Guide for Business, read November 18, 2026. Source of the B2B coverage, the eight requirements and the $53,088 per-email penalty.
- Google Workspace Admin Help, Email sender guidelines, read November 18, 2026. Source of the 0.10% and 0.30% spam-rate figures and the bulk-sender requirements.
- IBM, AI for lead generation, read November 18, 2026. Source of the definition quoted above.
- LinkedIn Help, Prohibited software and extensions, read November 18, 2026. Source of the rule on bots and scrapers.


