Explore AI Summary

AI sales prospecting: how to use AI, step by step

AI sales prospecting guide thumbnail: a one-sentence ICP for US IT services owners and directors, the first of seven steps

Contents

AI sales prospecting is prospecting with the research and assembly work handed to AI: turning your ideal customer into a list, finding verified emails and mobile numbers, researching each account and drafting the first touch. You keep the judgment and the conversation.

Below is the 7-step workflow, a worked example with copy-paste prompts, and the mistakes that make AI prospecting backfire.

What is AI sales prospecting?

AI sales prospecting is the use of AI tools to find, research and prepare outreach to potential buyers. You describe who you sell to, AI builds and enriches the list, summarizes each account from sources you can check and drafts a first message. The rep reviews, edits, calls and follows up.

It is the same activity as prospecting, minus most of the manual assembly. Salesforce’s 2026 State of Sales survey found sellers expect AI agents to cut prospect research time by 34% and email drafting by 36% once fully implemented. Those are expectations, so treat AI prospecting as a way to work more accounts well, not to send more messages.

How to use AI for sales prospecting in 7 steps

The seven steps below take you from a blank page to a first call. Steps 1 to 3 build the list, steps 4 to 6 prepare each touch, and step 7 is the conversation.

AI sales prospecting in 7 steps: one-sentence ICP, free count, list with verified emails and mobiles, rank on signals, sourced research, draft, call first
AI does most of the work in steps 1 to 6; step 7 is still a person on the phone.

1. Write your ideal customer as one sentence

Put the title, seniority, industry, location and company size in one plain-English sentence. “Owners and directors at US IT services companies with 11 to 200 employees” is usable. “SMB tech companies” is not.

If you do not have a clear buyer yet, start from an ideal customer profile template. For phrasing that tools read correctly, use our guide to describing your ICP in one sentence.

2. Size the market before you spend a credit

Run the sentence and check two things: how the tool understood each part, and how many people match. A wrong readback, like “IT services” read as “software”, is cheap to fix now and expensive to discover after a campaign.

In Scalelist’s Lead Finder, the count is free. If the number is far too big to work, add a detail it can read, such as company size or time in role.

3. Build the list with verified emails and mobiles

Pull the people and their contact details in the same step. Every email should come back with a status: Valid (the mailbox was verified), Risky (the domain accepts every address, so it cannot be confirmed) or Not found.

Do not let a language model produce addresses on its own. A guessed email looks right and bounces, and bounces hurt the domain everything else runs on. ZeroBounce’s 2026 report found at least 23% of email lists degrade within a year, so verify close to the send.

4. Rank the list on signals, not alphabet

Put accounts where something changed at the top: a job post, a new location, a new leader, a funding round. A rep working an alphabetical export spends the same effort on a company that has not changed in two years as on one that just doubled its help desk.

Our list of B2B buying signals groups the ones worth checking. An assistant with web access can check each account’s careers and news pages for you each morning.

5. Research each account, with a source for every claim

Ask the assistant for a short list of facts about the account, each with a date and a link. Research at the account level, not per contact: the account has the problem, and one research pass can support two or three contacts in different roles.

This is where AI saves the most time and causes the most damage. Fluent, wrong research is the most common failure; the section on mistakes below shows which claims are safe to send.

6. Draft the first touch with AI, then edit it

Give the assistant your offer, one fact from step 5 and a word limit, and ask for a draft. Then rewrite it in your voice. One specific, accurate detail beats three generic ones, and anything the prospect gets in every other email is worse than a plain message.

Edit because it is the step that decides replies, not because drafts are bad. 6sense’s 2026 BDR study found content production and automated outreach “are not reliably associated with increases in quota attainment.”

7. Call the mobiles first, email the rest, log everything

Start with the people you have a mobile number for. Nooks analyzed nearly 3 million dials and found mobile numbers had a 45% higher connect rate than other numbers. Email the rest from the same research, and let the assistant log outcomes and draft follow-ups for routine replies.

AI prospecting decision tree: mobile found means call first, Valid email means sequence, Risky email means secondary domain, nothing found means next contact
Route each contact by what was found, so the best channel for that person goes first.

AI prospecting example: an SDR selling to US IT services firms

Here is the workflow on one real market. Maya Chen is an SDR at Patchwise, a fictional Austin, Texas company that sells patch management software to managed IT service providers. The people and companies below are fictional; the count is real.

Maya types one sentence into Lead Finder. Scalelist matches over 120,000 decision-maker profiles (director level and above, including owners) at US IT services companies with 11 to 200 employees, as of September 2026.

Illustration of an AI sales prospecting search in Scalelist Lead Finder: US IT services owners and directors, over 120,000 profiles match, with email status and mobiles
Illustration of a Lead Finder search: the count is free, and a missing mobile costs nothing.

She works 50 new accounts a week, one decision maker each. At most that costs 1,100 credits a week: 50 for the leads, 50 for the emails and 1,000 for the mobiles. Misses cost nothing, so the real number is lower. About 4,400 credits a month fits the Scaler 5K plan at 199 USD a month.

Next, she researches the top-ranked account with this prompt:

Account research prompt (Claude or ChatGPT with web access)
Research [company] ([domain]) for a first cold call about [my offer].
Return a list of facts, not a paragraph. For each fact: the fact, its date, and the URL where you found it.
Check: services they sell, tools named on their site, open job posts, new locations, leadership changes in the last 6 months.
If you cannot find a source for something, leave it out. Do not guess.
End with the one fact most relevant to [my offer], and one sentence on why.

For Brightpath IT in Round Rock, Texas, the assistant returns three facts with links: two open help desk technician roles on the careers page, “managed patching” listed on the services page, and a new office announced on the news page. It finds no source for funding or headcount, so neither appears. Maya picks the hiring fact.

First-touch email, edited by the rep
Subject: help desk hiring at Brightpath

Hi Tom,

Saw Brightpath is hiring two help desk technicians in Round Rock. When new techs start, who owns patching across your clients?

Patchwise runs patch windows for every client from one screen, so a new tech can take it over in week one.

Worth 15 minutes next Tuesday to see it on one of your clients?

Maya
Call opener for the same account
Hi Tom, it's Maya from Patchwise. I saw you're hiring two help desk techs, so I'll be quick.
When new techs start, who owns patching across your clients: them, or you?

The email has one fact, one question and one ask. The call opens on the same fact, so the two touches reinforce each other. More openers that earn the first 30 seconds are in our list of cold call opening lines.

Type your own one-sentence ICP and see how many people match before you spend a credit.

Describe your buyer in one sentence

Common AI prospecting mistakes (and the fix)

Most AI prospecting problems come from trusting output that reads well. Each of these has a simple fix.

Research that is fluent and wrong

Automated research fails by producing plausible statements that are false. A prospect congratulated on a funding round that never happened learns something lasting about your company. Fix: require a link for every claim and use only the claim types below.

Table of AI research claims safe to send in prospecting emails: website facts and job posts yes, funding only with a link, inferred problems no
If the research cannot link it, it does not go in the email.

Personalization everyone else also sends

A first line about the prospect’s latest post or podcast is what most AI drafts reach for, so it reads as automated. Fix: use one fact about the account that changed recently and ties to your offer. Our guide to personalization at scale covers where to find those facts.

Unverified or Risky emails in the main sequence

A guessed or unchecked address bounces and drags down the domain. Fix: only Valid emails go into your main sequence. Scalelist’s help center reports that about 20 to 30% of Risky emails are actually deliverable, so send those from a secondary domain at low volume. More on handling catch-all emails.

More volume from the same list

When list building gets cheap, the tempting move is to send more. Gartner found 73% of B2B buyers actively avoid suppliers who send irrelevant outreach. Fix: hold volume steady, spend the saved time on fit and research, and cut volume if reply rate drops.

No number to call

Many AI prospecting tools stop at the email. The phone is where the conversations are.

Why mobile numbers matter in AI prospecting: 45% higher connect rate on mobiles, 4.6 vs 3.4 quality conversations a day for phone-centric teams, 87% phones found
Better research gets you nowhere if no one picks up: the mobile number is what turns preparation into a conversation.

Fix: pull mobiles for the accounts you will actually call, and call those first. Combining both channels well is covered in multichannel prospecting.

Best practices for AI sales prospecting

The teams that get meetings from AI prospecting treat AI output as a first draft and the rep’s time as the scarce resource.

  • Multi-thread from one research pass. Pick two or three roles who feel the problem differently, and change what the problem costs each of them, not the greeting.
  • Verify contact data close to the send. A list built last quarter needs a fresh check before it enters a sequence.
  • Read 20 drafts end to end before a launch. Skimming a template misses the invented detail.
  • Keep a list of banned phrases. Remove the lines every AI draft uses, so yours do not sound like the rest.
  • Measure meetings per hour of rep time. Not emails sent, not leads found.

What to look for in an AI prospecting tool

The best AI prospecting tool is the one whose data holds up on your own accounts. For a tested shortlist, see our comparison of AI prospecting tools for B2B teams. Before you buy, check for:

  • A readback of what it understood from your sentence, and a free count before you spend.
  • Mobile numbers as well as emails, with coverage measured in your region.
  • A status on every email (Valid, Risky, Not found) and no charge for a miss.
  • Export to CSV, your CRM or your sequencer, so a person approves every send.
  • A connector for Claude or ChatGPT if your team researches and drafts there.

If you want the assistant to do more of the loop, read how to automate SDR workflows with AI and where fully automated AI agents for lead generation break. The wider picture is in our hub on AI lead generation.

How Scalelist helps with AI prospecting

Scalelist is an AI Lead Finder: you describe your buyer in one sentence and get matching people with verified work emails and mobile numbers. It covers steps 1 to 3 of this workflow, and step 7’s mobile numbers.

  • Lead Finder shows what it understood and how many people match, for free. Then 1 credit per lead found, 1 per email found (Valid or Risky), 20 per mobile found, 0 for a miss.
  • The MCP connector runs the same search inside Claude, ChatGPT or Cursor, so research, drafts and contact data sit in one chat. Setup and prompts are on the Scalelist MCP server page. If you work in ChatGPT, see what it can and cannot do alone in ChatGPT for lead generation.
  • Coverage: in our own 8-provider benchmark, Scalelist found an email for 92% and a phone number for 87% of the North America list.

Plans start at 29 USD a month for 500 credits, with no per-seat pricing; see Scalelist pricing. New accounts get 100 free credits for 7 days.

Describe your buyer, check the free count, then reveal verified emails and mobiles only for the accounts you will work.

Get 100 credits to test it

FAQ

What is AI sales prospecting?

AI sales prospecting is using AI to find, research and prepare outreach to potential buyers. You describe your ideal customer, AI builds and enriches the list with verified emails and mobiles, summarizes each account from sources you can check and drafts a first message. The rep reviews, edits, calls and follows up. It removes assembly work, not the conversation.

Can AI do sales prospecting on its own?

Not well. AI can build the list, research accounts and draft messages, but it cannot judge which accounts matter this week or hold a real conversation. Fully automated sending also scales mistakes: an invented fact or a weak offer goes to hundreds of people at once. The reliable setup is AI for preparation and a person for every send and every call.

What is the best AI for prospecting?

There is no single best tool, because prospecting needs two kinds: a data source that finds verified emails and mobiles, and an assistant for research and drafting. A common setup is an AI lead finder such as Scalelist plus Claude or ChatGPT, connected over MCP so the assistant can search and enrich inside the chat.

How do I stop AI from making up facts in prospecting emails?

Ask for research as a list of facts, each with a date and a link, and tell the assistant to leave out anything it cannot source. Then use only claim types you can check, such as the company’s own website and published job posts. Never send a funding round, a number or a quote without the link.

How much does AI sales prospecting cost?

It depends on volume and whether you need mobile numbers. In Scalelist, a lead found costs 1 credit, an email 1 and a mobile 20, and nothing is charged for a miss. Working 50 new accounts a week with a mobile for each costs at most about 4,400 credits a month, which fits the 199 USD plan.

Is AI prospecting just for email?

No. AI does most of its work before the channel: the list, the research and the draft. The same research should feed a call, and calling people on their mobile is where conversations happen. Nooks found mobile numbers connect 45% more often than other numbers, so a list with mobiles makes AI prospecting pay off.

Sources

Arnaud Renoux

Co-Founder at Scalelist

Scalelist - Agent Storefront, powered by ZeroClick