Where AI fits in prospecting
You prospect in five steps, whether you think of them that way or not. You decide who to go after, build a list, find a reason to get in touch, reach out, and sort the replies. AI is good at the reading and first-draft parts of each step. It's bad at judgement, and it's bad at knowing when a fact is wrong.
So the useful question is which step to hand it. The answer for most reps and small teams: let AI do the research and the first drafts, and keep the choosing, the sending and the conversation for yourself.
If you want the wider picture of AI across the sales job, AI in sales covers it. This post stays on prospecting.
Five jobs AI does well
1. Research an account before you reach out
Give an AI assistant a company's website, a recent news item or a job post, and ask what's changed and who's likely to care. You get a summary in a minute that would take you ten to put together. Treat every line of it as something to check before you repeat it.
Here is [company]'s website text and a recent news item: [paste].
In five bullet points: what they sell, who to, what seems to have changed in the last six months, and which of these might make [the problem you solve] more pressing.
Mark anything you're guessing as a guess.
2. Sort a list by fit
Paste in a list of 50 companies with a line or two each, plus a description of your best customers, and ask the AI to sort them into strong, maybe and no. The sort will have mistakes in it, but it's a faster first pass than reading 50 websites, and you check the strong ones yourself. If you haven't written down who your best customers are, start with the ideal customer profile page.
Our best customers are [type, size, situation]. They buy when [trigger].
Sort the companies below into Strong fit, Maybe and No, with one line of reason each. Don't invent details that aren't in the text.
[list]
3. Write a first draft you'll rewrite
AI writes a decent first draft of an opening message or email, and it rarely sounds like you. Use the draft for its structure, then rewrite the first line in your own words from one fact you've checked.
Write a first message to [role] at [company], under 70 words.
The one fact to mention: [the fact you checked].
The one question to ask: [question].
No compliments, no "I hope this finds you well", no claims about results.
4. Prep a call
Before a first call, ask for the three questions most worth asking this person and the two objections they're likely to raise, then use the answer to warm up rather than read it out. The cold calling guide covers what to say in the first 30 seconds.
5. Triage replies
When replies pile up, AI can sort them into interested, not now, wrong person and unsubscribe, and draft a short answer to each. You read the draft and send it yourself. The replies worth the most, the "not now, try me in March" ones, still need a dated follow-up somewhere you'll see it.
Three things to keep for yourself
Spam at scale. Tools can now write and send hundreds of messages a day. If the list is loose and the messages are generic, you're only sending bad outreach faster, and buyers notice. Keep the volume to what you can follow up on properly.
Facts you haven't checked. AI sounds just as sure about an out-of-date detail as a true one. A first line that names the wrong job title or a product the company dropped last year ends the conversation. Check the one fact your message rests on.
The conversation itself. A buyer who replies wants a person. Answer the questions yourself, book the meeting yourself, and make the call yourself. AI can prep you, but it shouldn't be the one talking once someone's interested.
A weekly prospecting routine with AI in it
| Day | What you do | What AI does |
|---|---|---|
| Monday | Choose this week's 20 to 30 accounts | Sorts a longer list by fit |
| Tuesday | Pick the one fact per account worth mentioning | Researches each account |
| Wednesday | Rewrite and send the first messages | Drafts them |
| Thursday | Calls and follow-ups | Preps questions before each call |
| Friday | Review replies, date every follow-up, check your numbers | Sorts the replies and drafts answers |
Keep the numbers every week: accounts reached, replies, conversations and meetings booked. If replies are low, fix the list or the first line. If replies are fine and meetings are low, fix what you say once they answer. The prospecting system guide shows how to work back from your target to the week's numbers.
AI can write the first message. Keeping the follow-up is still on you. ActivityTracker keeps every callback on today's list, and shows by Wednesday whether your week is enough for your target.
A made-up week
Made-up example: Tom sells scheduling software to physio and dental clinics. He sets aside four hours a week for new prospecting.
Monday: he pastes 80 clinics from a directory into an AI assistant with a description of his best customers. It sorts 24 as strong fit. Tom checks those and keeps 20.
Tuesday: AI summarises each clinic's website. For 14 it finds something useful, like a second location opening or a job ad for a receptionist. For 6 it finds nothing, and Tom drops them to next week rather than send a generic message.
Wednesday: AI drafts 14 messages. Tom rewrites every first line himself and cuts two drafts that got a detail wrong. He sends 14.
Thursday and Friday: 4 replies. One wants a call next week, two say "not now, try us in the new year" and one is the wrong person, who names the right one. Tom books the call, dates the two follow-ups for January, and adds the new name to Monday's list.
Result: 14 messages, 4 replies, 1 meeting and 3 future leads, in about four hours. The numbers are invented. The research took him 40 minutes instead of the usual three hours.
Tools, sorted by the job
The AI sales tools post covers more tools across the whole sales job. For prospecting, these are the main types, each described from its own site:
- Research and lists: Apollo lists data on 240M contacts and 30M companies, and agents that research accounts, prioritise leads, write outreach and book meetings. Clay lists agents that research companies and people with AI, a data marketplace of 200+ providers, and signals such as job changes and promotions.
- Outreach at volume: Instantly lists a lead database, email verification, an AI reply agent and a copilot that finds leads, writes copy and builds campaigns.
- Research and drafts without a new tool: a general AI chat assistant you already pay for does the prompts above well enough for a rep sending 20 to 30 messages a week.
- Answering the replies that come in on WhatsApp: ActivityTracker's AI agent answers new WhatsApp enquiries on your own number in your words, books the meeting and passes you the chat when a person should take it. It doesn't find prospects or build lists.
What to do first
Pick the one step that eats most of your prospecting time. For most people it's research. Use the first prompt above on next week's accounts, time it against how long you usually take, and check every fact it gives you. If it saves time and the facts hold up, add the second job the week after.
Keep reading
Questions people ask
Sources
- Apollo, homepage: data on 240M contacts and 30M companies; agents that research accounts, prioritise leads, write outreach and book meetings, read 28 September 2026
- Clay, homepage: Claygents research target companies and people with AI; data marketplace with 200+ providers; signals such as job changes and promotions, read 28 September 2026
- Instantly, homepage: lead database, verification, AI reply agent, copilot that finds leads, writes copy and builds campaigns, read 28 September 2026