How to Use AI for Sales Prospecting Without Losing Relevance

When you built the product and it finally works, you need people who will buy it.

So you open an AI tool and ask it to find prospects and write to them. An hour later you have fifty companies and fifty messages. They all sound fine. They also all sound the same. Nobody replies.

This guide is about that gap. It explains how to use AI for sales prospecting so that it saves you time without making you sound like everyone else. It is written for people who build B2B software and now have to sell it. You don’t need sales experience to follow it.

What does AI sales prospecting mean?

Use AI to read. Don’t use it to decide.

AI is very good at finding companies, collecting public facts about them and putting those facts in one place. It is weak at judging which company is worth your week and what one person actually cares about. This is task for human.

AI sales prospecting is the use of AI to find companies worth contacting, learn about them and prepare the first message. The job is the same as it always was. The reading and the first draft are faster.

It is not the same as automated outreach. Some AI sales tools send thousands of messages for you. This guide is about using AI prospecting tools to do better research, so each message you send has a real reason behind it.

That difference matters more every month. Your prospects use AI too. They get more AI-written messages than ever. They have learned to spot them.

Why do generic AI messages fail?

A generic message is one that could go to any company in the same industry with only the name changed. AI produces these very easily, because a language model is trained to write text that sounds normal. Normal is exactly the problem.

Two pieces of research explain why this costs you replies.

People expect you to know who they are. McKinsey’s Next in Personalization 2021 research found that 71% of consumers expect companies to deliver personalized interactions, and 76% get frustrated when that does not happen. That study looked at consumers, not B2B buyers. But the person reading your message is the same person who shops online at night. The expectation travels with them.

Buyers don’t trust AI on its own. In a Gartner survey of 645 B2B buyers, run in August and September 2025, 69% said they prefer to validate AI-generated insights with a sales rep. In the same survey, 51% said they are more likely to meet misleading information from generative AI than from other sources. Buyers already use AI to research vendors. What they want from a person is something AI cannot give them, it’s a view they can trust.

Put those two findings together. Your prospect expects you to know their situation. They also distrust anything that reads like a machine produced it. A generic AI message fails both tests at once.

The good news is that this is your opening. If most messages in their inbox are generic, a specific one stands out more than ever.

When to use AI and when to do it yourself?

If you use AI at work, there’s one study you really need to know about. In 2023, Harvard Business School and Boston Consulting Group did an experiment with over 750 consultants. Some of them used GPT-4. When the AI was given tasks it was good at, people using it got 12% more done, worked 25% faster, and the quality of their work was over 40% better.

But when the task was outside the AI’s comfort zone, things went completely the other way. Consultants who used AI were actually 19% less likely to get the right answer compared to those who just did the work themselves.

Researchers call this the “jagged technological frontier.” It basically means that AI doesn’t just do “easy” things well and fail at “hard” things. The line is really blurry. Two tasks might look very similar, but AI will be great at one and terrible at the other. The tricky part is that people using AI often didn’t even notice they were getting bad results because the work still looked fine.

If you work in prospecting, this means your first step isn’t just picking the right AI tool. It’s figuring out which parts of the job AI can actually handle, and which ones it can’t.

Here’s our guide to help you out. Keep in mind it’s based on our own judgment, not exact science, so make sure to test it out in your own market.

Prospecting task	Give it to AI?	Why
Finding companies that match a written profile	Yes	Matching public text against a clear description
Collecting public facts about a company	Yes	Reading and summarizing what already exists
Summarizing a long report, changelog or job post	Yes	Compressing text
Writing a first draft you will rewrite	Yes, with care	Fluent text is what models do best
Picking the one detail worth mentioning	Partly	AI can suggest. You judge if it is flattering or awkward
Deciding which account is worth your week	No	Needs your capacity, pricing and priorities
Guessing what a specific person cares about	No	The model predicts. It does not know
Choosing what not to say	No	Needs a sense of risk the model does not have
Pressing send	No, never	The only step where a mistake is still fixable

A 6-step guide for SaaS builders

Here’s the process we recommend. It only takes a few hours a week, which is usually all the free time a founder actually has.

Step 1: Base your ICP on real customers

AI makes your search wider, not smarter.

If your Ideal Customer Profile (ICP) is too vague, AI will just quickly generate a massive list of the wrong companies. The list might look impressive, but no one will ever reply to your emails.

That’s why you should create your profile first, based on the customers or users you already have. If you don’t have any yet, focus on the people who gave you the best feedback.

A solid ICP describes details you can actually see from the outside:

• what the company sells (in plain English)
• its size (as a range you can easily verify)
• the specific situation that makes them need you right now, not next year
• the role of the person who actually deals with the problem
• the role of the person who holds the budget

A useless ICP relies on buzzwords. Words like “innovative,” “fast-growing,” or “tech-forward” don’t help because you can’t really search for them.

Next, write down the second half: who isn’t a good fit, and why. Most people skip this step, but it’s exactly what stops an AI search from just giving you every single software company in the country.

Here’s a quick test. Show your ICP to a friend who doesn’t work with you. Ask them to name three companies that fit. If they can’t, your profile is not a filter yet.

Step 2. Let AI find leads, but you make the final call

Now, feed your profile to the AI and ask it to find matching companies and contacts. Make sure to ask why it picked each result. An answer like, “This company fits because they launched a self-serve plan in July,” is actually helpful. Just saying, “This company is a good fit,” isn’t. If the AI doesn’t give you a reason, it’s just guessing.

Keep your list short. Twenty solid accounts will always beat two hundred mediocre ones. Your real limit isn’t how many companies the AI can dig up—it’s how many meaningful conversations you can actually handle in a week.

This is also the honest answer to a question a lot of founders ask: “Is an AI lead gen tool enough to find customers for my SaaS?” The truth is, it helps you find potential prospects, but it won’t tell you who’s actually worth your time. That decision is still entirely up to you.

Step 3. Look for hard facts, not opinions

Putting together a proper account brief usually takes a human 30 to 60 minutes. You have to check their website, recent news, career pages, product updates, and maybe a few LinkedIn profiles. Most of that time isn’t even spent searching—it’s just reading and taking notes. And that’s exactly what AI can do for you.

But remember: how you phrase your prompt determines how good the answer will be.

• Ask for facts, not opinions. Asking, “What has this company announced in the last 12 months, and where?” is a great prompt. Asking, “Is this company a good prospect?” is a terrible one, because it basically invites the AI to make things up.
• Demand a source for every claim. And I don’t mean a random list of links at the very end. You want a source attached to every single sentence.
• Ask what it couldn’t find. If a brief doesn’t admit to having any missing info, it means the AI probably just made something up to sound smart.
• Set a timeframe. Otherwise, it might present a two-year-old funding round as breaking news.

A good brief has a source for every claim, clear dates, and blank sections when nothing was found. A bad brief is just smoothly written fluff with phrases like “the company appears to be focused on…” If you can’t tell the difference between hard facts and AI guesses, that brief is honestly worse than having no brief at all.

Step 4. Only fact-check what you’re going to use

Everyone says you should check AI research. Almost nobody says how much checking is enough. You can work it out.

Say a research brief contains n separate facts. Say each one has a small chance p of being wrong. If you check nothing, the chance that at least one wrong fact gets into your message is:

P = 1 − (1 − p)ⁿ

kōdlo ran this formula for the range that matters in account research.

Facts in the brief	1% error rate	2% error rate	5% error rate	10% error rate
5 facts	4.9%	9.6%	22.6%	41.0%
10 facts	9.6%	18.3%	40.1%	65.1%
20 facts	18.2%	33.2%	64.2%	87.8%
40 facts	33.1%	55.4%	87.1%	98.5%

Just look at the 20-fact row. Even with an error rate of just 2%—which sounds incredibly low—there’s a one-in-three chance that your brief contains a made-up detail. Is 2% realistic? Honestly, nobody officially tracks AI error rates for account research, so it’s hard to say. Treat these numbers as a way of thinking about the problem, not hard science. But here’s the best part: you’re never going to send the entire brief to a prospect anyway. Your first email will probably only use four or five key facts. So, just verify those five, and ignore the rest.

If we assume a 2% error rate, fact-checking only those five details drops your risk of making a mistake from 33.2% down to 9.6%. At a 5% error rate, it drops from 64.2% to 22.6%. It’s all about checking smart, not checking everything.

The rule: check what you will say out loud. Skim the rest.

With sources attached, checking one fact takes about a minute. Without sources, you first have to find where the fact came from. That can take five. This is the real, practical reason cited research matters. It makes checking cheap enough that you will actually do it.

Step 5. Build your message around a single, verified fact

Now that you have a shortlist of fact-checked details, it’s time to start writing. You can definitely use AI to write the first draft if you want, but you’ll almost certainly need to rewrite it. Keep these three rules in mind:

• Start with one highly specific fact about them. Mention a recent launch, a new job posting, a tweak to their pricing page, or a talk their team gave. Your opening sentence should be so tailored that it could only possibly apply to this one specific company.
• Explain what you do in one simple sentence. Drop the buzzwords. Just clearly state exactly what your product does and who it’s for.
• Keep your “ask” small and low-friction. Ask for a quick reply, a brief chat, or some honest feedback. Never push for a full demo in the very first email.

Keep the whole thing brief. Your prospect will decide in a matter of seconds whether you actually did your homework or just copy-pasted from a massive list.

Finally, always run the “swap test” before hitting send.

Step 6. Base your follow-ups on the same context

The reality is that most replies come from follow-ups, not the initial outreach. But this is exactly where most founders drop the ball. You get heads-down in coding for a week, and suddenly three weeks go by without you reaching out again.

The easiest fix is to keep everything centralized: why you targeted them in the first place, your research, what you originally sent, and their replies. When you follow up, work from those notes instead of relying on your memory. Reaching out with a new, relevant detail about their company works infinitely better than the classic, lazy “just checking in” email.

AI can actually be a huge help here, too. You can ask it to monitor your target accounts weekly and flag any changes—like a new key hire, a product launch, or a pricing update. This gives you a genuine, natural excuse to reach out again.

A 10-second trick to spot generic emails

This is the easiest check we know, and it instantly catches lazy, generic messages. Take your final draft. Swap out the prospect’s company name for their biggest competitor, and read it again. If the email still makes perfect sense, it’s too generic. Time to rewrite that opening line. But if the message suddenly sounds ridiculous because it relies on a detail that’s only true for your prospect? You’re good to hit send. The swap test works because it measures the one thing that actually separates a great outreach email from spam: a verified fact that belongs only to the person reading it. It takes ten seconds, and you don’t need any special software to do it.

Want to make it even stricter? Try swapping the person’s name and job title, too. If your email could easily be sent to anyHead of Engineering at any SaaS company, it’s simply not specific enough.

How many prospects can a founder actually handle a week?

A huge mistake people make is letting AI dictate their volume. Sure, an algorithm can find you thousands of companies, but you can’t realistically talk to thousands of people.

Instead, work backwards from the time you actually have. Here’s a quick breakdown—just plug in your own numbers:

Let’s say you have 5 hours a week dedicated to sales (that’s 300 minutes).

• Doing it the old-school way: 45 minutes of research + 10 minutes of writing = 55 minutes per account. 300 ÷ 55 gives you about 5 accounts a week.
• Using AI + the “check what you use” rule: 5 minutes to read the brief, 7 minutes to fact-check your chosen details, and 10 minutes to write. Let’s add a few minutes for setup, bringing you to roughly 27 minutes per account. 300 ÷ 27 means you can handle about 10 accounts a week.

You’re basically doubling the number of highly researched, relevant emails you can send in the exact same amount of time, without sacrificing an ounce of quality. That is what AI should be doing for you. It’s not about blasting out fifty emails—it’s about sending ten great ones instead of five.

If an AI tool promises to pump out a hundred emails a week in those same five hours, you really have to ask yourself: what happened to the actual thinking and fact-checking?

Why this is a big deal if you’re new to sales

If you’re a founder, there’s some great news here.

A 2025 study from Stanford and MIT looked at what happened when over 5,000 customer support agents started using AI. Overall, productivity jumped by 14%. But for beginners and less experienced staff? It skyrocketed by 34%. Meanwhile, the veterans barely saw any change at all. Sure, customer support isn’t sales, but the trend is pretty clear. AI shines brightest when you’re inexperienced because it essentially hands you the playbook of seasoned pros.

This perfectly describes most SaaS founders who suddenly have to do sales. If you’re an engineer who just started selling last quarter, AI can completely change your workflow. But if you’ve been selling in your industry for 15 years, the boost will be smaller—mostly just saving you reading time, rather than doing the heavy lifting for you.

Either way, the six steps we covered will protect you from the biggest trap of all: blasting a fast, confident, but completely generic email to the one company that actually wanted to buy.

The 5 most expensive mistakes

Researching everyone. Doing background research is practically free now, so people tend to research every single company on their list. But your bottleneck was never research—it’s your calendar.

Trusting smooth writing. Just because a paragraph reads beautifully doesn’t mean it’s true. Remember the “jagged frontier” from earlier? People getting bad results from AI rarely realized it because the output looked fine.

Sending the first draft. AI writes for the average reader. Your prospect is a specific person who has already seen that exact AI-generated format a hundred times.

Letting a tool hit send for you. Putting your outreach on autopilot removes your absolute last chance to catch mistakes. It risks your reputation and burns perfectly good leads.

Tracking emails instead of conversations. “Messages sent” is the easiest metric to inflate, but it has almost zero connection to actual revenue. Instead, track the replies you get from people who actually noticed the specific details you included.

How to put this into practice

Since we built kōdlo (an end-to-end sales workspace), it’s only fair we explain how our tool actually supports this whole process.

kōdlo is built exactly around the method we just covered. Our AI agent interviews you to nail down your ICP, then scours the open web for companies and contacts that actually match it. We call this feature Discovery. Every prospect it finds comes with a clear reason why they’re a good fit, plus a link backing it up. Our deep search creates research briefs where every single claim is sourced, meaning you can fact-check things in minutes.

The agent will even draft a first message in your tone of voice. But from there, you take the wheel: you edit it, pick the channel, and hit send. kōdlo never sends anything on autopilot.

Everything lives in one workspace: the reason you targeted the account, your research, the message, and the follow-ups. That’s what makes Step 6 actually doable, without pulling your hair out over a massive spreadsheet.

Want to see how this plays out for SaaS founders? Check out our page on AI prospecting. Selling services and need deeper research? Take a look at account research!

Free plan · no credit card required

Frequently asked questions

The bottom line

AI doesn’t automatically make you relevant. It just makes being relevant a lot cheaper.

The actual work of winning a customer hasn’t changed at all. You still need to know exactly who you’re selling to, uncover a real fact about them, state your value clearly, and follow up with fresh information. AI can shrink an hour of reading down to just a few minutes, but it can’t tell you what actually matters to the human on the other end of the email.

So, let AI do the heavy reading, but keep the final judgment for yourself. Fact-check the details you plan to use, and always run the swap test before you hit send.

If you’re looking for a workspace built specifically around this exact workflow, check out how kōdlo handles AI prospecting.

Sources

• Dell’Acqua, F., McFowland III, E., Mollick, E., Lakhani, K. R. and colleagues. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Working Paper 24-013, 2023. hbs.edu
• Brynjolfsson, E., Li, D. and Raymond, L. Generative AI at Work. The Quarterly Journal of Economics 140(2), 2025. academic.oup.com
• McKinsey & Company. The value of getting personalization right, or wrong, is multiplying. Next in Personalization 2021 Report, November 2021. mckinsey.com
• Gartner. Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights. Press release, 20 May 2026. gartner.com

Note: The two calculations in this article (error risk and weekly capacity) were run by the kōdlo team using the formulas and assumptions shown above. We highly encourage you to plug in your own numbers to see what works for your week.