Build a Cold Email List With an AI Agent, Apollo and Hunter

The short answer: if someone on your team assembles cold email lists by hand and the monthly list stays under the 1,000 uploaded contacts of Instantly's Growth plan, build this: an n8n agent searches Apollo at no credit cost, a person approves its shortlist in a sheet, and only approved rows are enriched, checked by Hunter and loaded into a draft Instantly campaign. On monthly billing that is $65 for Apollo Basic, $49 for Hunter Starter, $47 for Instantly Growth and $24 for n8n Starter, $185 a month in all, on the Apollo, Hunter, Instantly and n8n pricing pages, checked on 26 October 2026. For example, at 4 minutes of hand research per prospect, 400 prospects cost 400 x 4 = 1,600 minutes a month; the agent turns most of that into reading one line per row.

Disclosure: I may earn a commission through some links on this site. It does not change what this page says.

Searching is the cheap part of a prospect list. Judging who belongs on it is the expensive part, and paying for contact details of people nobody would approve is where the money leaks. Apollo's People API Search costs 0 credits and returns no emails or phone numbers, so an agent can search as often as it needs. Credits are spent later, on rows a person has already said yes to.

Before you build:

1. The month's bill, worked out. For example, 400 approved prospects use at most 400 Apollo credits, because people enrichment charges 1 credit for an email or the person's details and 0 when nothing is found; that sits inside Basic's 2,500 a month. Hunter checks them at 0.5 credits each, 400 x 0.5 = 200 of Starter's 2,000, per its pricing page. And 400 uploads fit Growth's 1,000 contacts.
2. The accounts you already sell to. A Google Sheet with one row per target account (company domain and tier) and a short brief: titles you sell to, titles you never want, at most two people per company. Apollo's search takes up to 1,000 domains per request, per the search reference.
3. Keys with narrow permissions. Apollo can limit a key to one endpoint, so the agent gets a search-only key and enrichment runs on a second key in a fixed step (search and enrichment references). Give the Instantly API v2 key only lead scopes, chosen at creation (getting started), and add Hunter's key to the n8n Hunter node.
4. The numbers from your last list. Bounce rate and replies per 100 contacts from the last campaign you sent. That is the bar the new list has to clear.

The workflow, from account row to draft campaign

1. Trigger. A Google Sheets Trigger on Row added starts a run for each new account. A polling trigger counts an execution only when it finds new data, per n8n's executions page.
2. The agent searches and shortlists. An AI Agent node takes your brief as its System Message and gets one tool: an HTTP Request node pointed at Apollo's search with the search-only key. The Tools Agent page lists HTTP Request among its tools, and Max Iterations, 10 by default, caps the model's runs. The agent picks the titles and seniorities to try, reads the results and drops the misfits. With Require Specific Output Format on and a Structured Output Parser attached, it returns at most two people per account: Apollo ID, first name, title and one sentence on why the person fits.
3. The shortlist lands in a Review tab. A Google Sheets node appends each row with an empty Approved column.
4. A person approves. Someone who knows the accounts reads the reasons and types yes or no. At this point nothing has cost a credit and nothing has reached a prospect. To approve in Slack instead, see the review step in how to use n8n.
5. Enrichment for approved rows only. A second workflow starts on Row updated, and an IF node lets through rows marked yes. An HTTP Request node calls Apollo's people enrichment by ID with the enrichment key. Leave the phone and personal email options off; a mobile adds 8 credits.
6. Hunter checks every address. The n8n Hunter node's verify operation returns Hunter's status, and the API documentation defines each one. Valid goes on. Invalid is dropped. Accept_all, where the server accepts any address so the mailbox cannot be confirmed, and unknown go to a Hold tab for a person. If Apollo found no email, the same node can generate the likely address from a name and domain, at 1 credit per email found.
7. Instantly receives a draft list. An HTTP Request node posts the valid rows to Instantly's bulk endpoint, up to 1,000 leads per call into one campaign, with skip_if_in_workspace set to true so anyone already in another campaign or list is left out. Add a custom variable source set to agent, for the measurement below, and write the response's counts of uploaded, blocklisted, duplicated and invalid leads back to the sheet.
8. A person launches. The campaign stays a draft until someone reads a sample of rows in Instantly and clicks Launch. Mailboxes, warm-up and sending limits are covered in how to use Instantly.

One rule no verifier settles. When contact data did not come from the person, GDPR Article 14(3) puts the notice, including where the data came from, at the latest at the first communication (GDPR). Put that line in the first email, and check each country's ePrivacy rules for company recipients.

The free version, done by hand

Run it once by hand first. Thirty-five accounts are enough to test the brief, and free tiers cover them.

For example, 35 accounts x 2 people = 70 emails, inside the 75 credits a month of Apollo's Free plan on its pricing page. Verifying those 70 addresses at 0.5 credits is 35 credits, inside the 50 a month of Hunter's Free plan. Instantly's free trial takes 250 leads and 1,000 emails over 14 days, with no card.

Do the agent's part yourself: search in Apollo by domain and title, and write the one-line reason in the sheet before saving anyone. After 70 rows you will know whether the brief is clear enough to hand to a model. Import the CSV into Instantly by hand, because its bulk endpoint answers 402 to a workspace with no active paid plan, per the API reference.

What the paid tools and the agent add

Volume, first. Apollo Basic grants 2,500 credits a month against Free's 75, Hunter Starter 2,000 against 50, and Instantly Growth 1,000 uploaded contacts and 5,000 emails a month, on the Apollo, Hunter and Instantly pricing pages. The agent adds the search and the reasoning for every new account, and the sheet keeps every reason and every yes, so a list can be explained a month later.

All prices were checked on 26 October 2026. Yearly billing lowers three of them on the same pages: Apollo Basic to $49 per seat, Hunter Starter to $34 and n8n Starter to $20 a month. Executions stay small: for example, 200 new accounts and 400 approved rows make at most 200 + 400 = 600 runs, under a quarter of Starter's 2,500 on the n8n pricing page.

Instantly will also verify leads on import, at 0.25 credits a lead from its Credits subscription, which is sold separately from the Outreach plan, per its verification article and pricing file. A sending tool that also sells the checks earns more when every step happens inside it, so decide whether a second paid check on a row Hunter passed minutes ago is worth the credit before ticking Verify leads.

How to know it worked

The first measure is the bounce rate on the first 200 sends. Instantly checks a campaign once it has sent 200 emails and, by default, pauses it when bounces exceed 5%, per its auto-pause article. Your last list's rate is the baseline; the new list should come in under it.

The second is replies per 100 contacts, against the same baseline. The third is the reviewer's no rate. My rule of thumb: if the reviewer rejects more than one row in four, rewrite the brief before adding accounts, because the agent is following instructions that do not say what you mean.

Then a holdout. For the first month, build a quarter of the accounts by hand as before and load them with source set to hand, in the same campaign. After 200 sends, compare bounces and replies per 100 contacts by source. If the agent's rows do no better than the hand-built ones, the hours saved are the whole benefit. That can be enough. You should still know which it is.

FAQ

Does the agent spend Apollo credits?

Not with this design. Its only tool is the search endpoint, which Apollo's reference lists at 0 credits, and its key cannot call enrichment. Credits start in the fixed enrichment step, on rows a person approved.

Why check an Apollo email again with Hunter?

Because the second check sorts the risky addresses for you. Hunter's API documentation reports accept_all separately from valid: the server accepts any address, so a mailbox that does not exist can still look fine until it bounces. Those rows go to a person instead of the campaign.

Can a chat assistant do this without n8n?

For a small batch, yes. Apollo runs its actions in Claude, ChatGPT and other AI tools on your existing permissions and credit balance (Apollo MCP article), Hunter publishes an MCP server, and Instantly's MCP server exposes its whole V2 API, campaign pausing included, so give it a key with narrow scopes. What n8n adds is the record: the sheet keeps every reason and approval after the chat is closed. Each tool's own screens: how to use Apollo.io and how to use Hunter.