# How AI agents find B2B leads, and what to check

> Where an AI agent’s list of businesses and decision makers comes from, why models alone get work emails wrong, the UK and US email rules, and what a good list includes.

- URL: https://ikag.dev/blog/how-ai-agents-find-b2b-leads
- Author: Ronit Ahuja, Founder, IKAG (https://ikag.dev/authors/ronit-ahuja)
- Published: 2026-10-17
- Updated: 2026-10-17
- Reading time: 5 min
- Publisher: IKAG

## Overview

An AI agent finds B2B leads by looking them up, not by remembering them: company registries say who a business legally is, its own website says who does what, and B2B contact databases hold work email addresses. Without those tools a model guesses, and guesses badly: in a September 2026 test by Hunter, 76% of the checkable email addresses that models produced were invalid. A good list names the source and date of every detail, lists what it couldn’t find, and follows the email rules where you send.

## Key points

- An AI model doesn’t hold a directory of work email addresses. To build a lead list it needs tools that reach company registries, business websites and B2B contact databases.
- In a test Hunter published in September 2026, 76% of the email addresses that language models produced on their own, and that its verifier could check, were invalid. Hunter sells email-finding tools.
- Company registries confirm a business’s legal name, status and officers, but not its inbox: Companies House doesn’t publish the email address each limited company gives it.
- In the UK, PECR lets you email limited companies without consent if you say who you are and offer an opt-out, but a named person’s work email is personal data under UK GDPR. In the US, CAN-SPAM makes no exception for business email.
- A good lead list gives the source and date of every detail, says which email addresses were verified, and lists what it could not find.

An AI agent can build a list of businesses and decision makers, with their work contact details, but not from memory. A language model doesn’t hold a directory of who works where, and when it is pushed for an email address it tends to make one up. The agent has to look each detail up in a source that holds it, and a good list tells you which source that was.

## Where does B2B lead data come from?

Most lead lists combine three kinds of source.

**Company registries.** A registry records who a business legally is. In the UK, Companies House gives anyone a company’s [registered address, date of incorporation and current and resigned officers](https://www.gov.uk/get-information-about-a-company) at no charge, and its [register](https://find-and-update.company-information.service.gov.uk/) shows whether the company is active, with the note that Companies House doesn’t check the accuracy of what is filed. What a registry doesn’t give you is an inbox. A limited company must give Companies House an email address when it is set up, but that address is [not published on the public register](https://www.gov.uk/limited-company-formation/company-address).

**The businesses’ own websites.** Team pages, contact pages and press releases say who does what, and often list a work email or a general inbox. Reading them is ordinary web data work, with the usual rules about each site’s terms and robots.txt (see [how AI agents get live data](https://ikag.dev/blog/how-ai-agents-get-live-data)).

**B2B contact databases.** Commercial databases hold named contacts and work email addresses at scale. ZoomInfo, for example, [says it builds its database](https://www.zoominfo.com/data-sources) from public information on more than 28 million sites and domains a day, data partners, an in-house research team, and more than 200,000 professionals who share business contacts in exchange for access to its tool. A database like this is broad, but you are relying on its sourcing, and people change jobs.

| Source                 | What it gives                            | What it lacks                 | Watch for                       |
| ---------------------- | ---------------------------------------- | ----------------------------- | ------------------------------- |
| Company registry       | Legal name, number, status, officers     | Email addresses               | Trading and legal names differ  |
| The business’s website | Who does what, published contact details | A date on most pages          | Site terms; pages that go stale |
| B2B contact database   | Named contacts and work emails at scale  | How each record was collected | Job changes; your lawful basis  |
| The model on its own   | Plausible guesses                        | Any check at all              | Invented addresses              |

## Why can’t the AI just know the email address?

Because it is guessing. In a [test Hunter published in September 2026](https://hunter.io/blog/llms-get-email-data-wrong/), language models were asked for the work email addresses of people in roles such as head of marketing at 125 real companies, and every address they returned went through Hunter’s email verifier. Of the 1,604 addresses that got a clear verdict, 76% were invalid, 10% could not be confirmed either way and 14% were valid. When prompts pushed for a best guess, models filled the gaps with addresses they had made up.

Hunter sells email-finding and verification tools, so read its numbers with that in mind. Its own comparison makes the point, though: the strongest model it tested ended with a valid address for 19.5% of requests on its own, and for 78.1% when it could call Hunter’s lookup tool. A model is good at deciding who to contact and why. The address has to come from a source that can check it.

## What are the rules for emailing businesses?

Who you may email, and how, depends on where you and they are. In outline:

- **UK, PECR.** The [ICO’s guidance on business-to-business marketing](https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/business-to-business-marketing/) says the PECR rule on marketing by email doesn’t apply to “corporate subscribers” such as limited companies and limited liability partnerships, so you don’t need their consent. You must say who you are and give a valid address for opting out. Sole traders and some partnerships count as individuals, so you need their consent or the “soft opt-in”. If you can’t tell which a contact is, the ICO says to treat them as an individual.
- **UK, UK GDPR.** A work email that identifies a person, such as a first name and surname at a company, is personal data. You need a lawful basis, often legitimate interests after a three-part test, and if you got the data from somewhere other than the person, you must usually tell them within a month. People have an absolute right to object to direct marketing. The ICO says this guidance is under review after the [Data (Use and Access) Act 2025](https://www.legislation.gov.uk/ukpga/2025/18/contents).
- **US, CAN-SPAM.** The [FTC’s compliance guide](https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business) says the law “makes no exception for business-to-business email”. Commercial messages need honest headers and subject lines, a valid postal address and a clear way to opt out, and opt-outs must be honoured within 10 business days. Hiring someone else to send them doesn’t move the responsibility.
- **Platform terms.** Sites have rules of their own. [LinkedIn’s User Agreement](https://www.linkedin.com/legal/user-agreement), for example, forbids software that scrapes or copies its services, including profiles. The ICO also notes that messaging someone on a professional networking site is often not business-to-business marketing at all, because people use those sites in a personal, professional capacity.

This is a summary, not legal advice. Check the rules for your own campaign, and ask a lawyer when the stakes are high.

## What should a good lead list include?

Here is the kind of request you might give your AI. Illustrative:

> “Find 40 dental practices in Bristol and the practice manager at each, with their work contact details. Say where each detail came from and when, and flag any practice that’s a sole trader or partnership.”

A list you can act on has, for each row:

- **The business:** its legal name and registry number, because names aren’t unique, and whether the registry shows it as active.
- **Its legal form:** a company, or a sole trader or partnership, because PECR treats them differently.
- **The decision maker:** their role, and the page that names them.
- **The work email address:** with its verification result (valid, can’t be confirmed, or invalid) and the date it was checked.
- **Sources and dates:** where every value came from and when, as described in [what the source record behind an AI answer contains](https://ikag.dev/blog/what-the-source-record-behind-an-ai-answer-contains).
- **Gaps, listed:** “no named manager on the practice’s website” is more useful than a guessed name.
- **Opt-outs:** checked against your own suppression list before anything is sent.

Fewer rows with sources beat a full sheet of guesses, and the [five-step check for AI answers](https://ikag.dev/blog/how-to-check-ai-answers) works on lead lists too.

## Where IKAG fits

Lead data is one of the [next kinds of tools](https://ikag.dev/#what) IKAG is building: your AI would ask for businesses and decision makers, with their work contact details, and IKAG is designed to return each detail with where it came from, where the source supports it. You would pay per call instead of a subscription to a lead database, and no prices have been set. IKAG’s first tool, company records from GLEIF, SEC EDGAR and Wikidata, already runs inside it. IKAG is opening to early users and data partners first.

## Questions

### Can an AI assistant find someone’s work email address?

Only if it can reach a source that holds it, such as the company’s own website or a B2B contact database. On their own, models tend to guess: in Hunter’s September 2026 test, 76% of the checkable addresses that models produced were invalid.

### Do I need consent to email businesses in the UK?

Not for limited companies and other corporate bodies under PECR, as long as you say who you are and give a valid address to opt out. Sole traders and some partnerships need consent or the soft opt-in, and a named person’s work email is still personal data under UK GDPR. This is not legal advice.

### Does CAN-SPAM apply to business-to-business email?

Yes. The FTC says the law makes no exception for business-to-business email. Commercial messages need honest headers and subject lines, a valid postal address and a clear way to opt out, and opt-outs must be honoured within 10 business days.

### Can an AI agent collect contacts from professional networking sites?

Check the site’s terms first: LinkedIn’s User Agreement, for example, forbids software that scrapes or copies its services, including profiles. The ICO adds that people on such sites are often there in a personal, professional capacity, so UK GDPR and PECR still apply.

### What should a B2B lead list include?

For each business: its legal name and registry number, whether it is active, the decision maker’s role, a work email address with its verification result, where each detail came from and when, and a note of anything that could not be found.

## Sources

- [Hunter: LLMs get 76% of emails wrong, but Hunter can easily fix it (1 September 2026)](https://hunter.io/blog/llms-get-email-data-wrong/)
- [ICO: Business-to-business marketing (guidance under review after the Data (Use and Access) Act; checked 17 October 2026)](https://ico.org.uk/for-organisations/direct-marketing-and-privacy-and-electronic-communications/business-to-business-marketing/)
- [legislation.gov.uk: Data (Use and Access) Act 2025](https://www.legislation.gov.uk/ukpga/2025/18/contents)
- [FTC: CAN-SPAM Act, a compliance guide for business (August 2023)](https://www.ftc.gov/business-guidance/resources/can-spam-act-compliance-guide-business)
- [LinkedIn: User Agreement (effective 3 November 2025)](https://www.linkedin.com/legal/user-agreement)
- [GOV.UK: Get information about a company (checked 17 October 2026)](https://www.gov.uk/get-information-about-a-company)
- [Companies House: Find and update company information (checked 17 October 2026)](https://find-and-update.company-information.service.gov.uk/)
- [GOV.UK: Check the rules for registered office addresses and email addresses (checked 17 October 2026)](https://www.gov.uk/limited-company-formation/company-address)
- [ZoomInfo: Data sources, where we get our data (checked 17 October 2026)](https://www.zoominfo.com/data-sources)