Why Better Prompts Won’t Fix Generic Outreach

If ChatGPT or Claude keeps producing generic outreach for you, the answer probably isn't writing a better prompt every time.

The bigger problem is that you're starting from scratch every time.

How do you use AI for personalised outreach?

A good AI outreach system needs more than a prompt.

It needs to understand:

• who you are trying to reach
• what makes someone worth approaching
• why you are contacting them now
• what problems they are likely to care about
• why your business is relevant
• how you want your outreach to sound
• what good and bad outreach looks like to you
• how much personalisation is appropriate
• what the AI should never say

Once that knowledge is captured, AI can combine it with information that changes for each prospect, such as their company, role, hiring activity or another relevant trigger.

That is very different from opening ChatGPT and asking:

"Write me a personalised cold email."

I saw this problem first-hand recently while working through an outbound sequence with a recruitment founder.

Why does AI-written outreach become generic?

The founder was trying to build a five-email sequence for his recruitment agency.

He had already spent a fair amount of time explaining what he wanted.

Things like:

• Don't spend half the email talking about the agency.
• Don't fill it with generic recruitment claims.
• Don't keep throwing in testimonials.
• Focus on the client.
• Show that you understand their problem.
• Give them a genuine reason to care.
• Keep the message relatively concise.
• Don't make it sound as though an AI wrote it.

Every time he corrected the AI, the output improved.

But he kept having to correct the same types of problems.

Eventually, creating five emails had taken far longer than it should have.

And I think this is where a lot of businesses are currently using AI the hard way.

Every time they need some outreach:

New chat → new prompt → explain the business → explain the audience → explain the tone → remove the generic language → rewrite → repeat.

If you have to teach the AI the same thing every time, you haven't really built AI into your outreach process.

You're just using it as a writing assistant.

Why a better prompt only gets you so far

Imagine a recruiter wants to approach a company that has just advertised several software engineering vacancies.

They give ChatGPT the company name and ask:

"Write a personalised email offering our recruitment services."

The result will often look something like this:

"I noticed you're growing your engineering team and wanted to reach out. At XYZ Recruitment, we specialise in connecting innovative organisations with top technology talent…"

Technically, it is personalised.

The company is mentioned.

The vacancies are mentioned.

But there is very little reason for the recipient to care.

So the recruiter starts correcting it.

Make it shorter.

Don't spend so long talking about us.

Stop saying "top talent".

Mention the specific vacancy.

Make it less salesy.

Use the candidate we already represent as the reason for contacting them.

Don't force a testimonial into the message.

Eventually, the email improves.

But those corrections contain something valuable.

They represent how that recruitment business thinks good outreach should work.

That knowledge shouldn't disappear when the conversation closes.

How do you make outreach more personalised with AI?

One of the biggest misconceptions around personalised outreach is that personalisation simply means mentioning something specific about the person.

It doesn't.

An email isn't necessarily good because it begins:

"I saw you went to Manchester University…"

or:

"Congratulations on your recent promotion…"

Useful personalisation is usually about commercial relevance.

A good outreach system should be able to answer four questions:

Why this company?

What makes this business particularly relevant to you?

Perhaps they are:

• hiring heavily in your specialist market
• expanding into a new geography
• building a new team
• launching a new service
• recruiting for several related positions
• going through another change that creates a genuine reason to speak

Why now?

Why is today a better time to approach them than six months ago?

There should ideally be some form of signal or change.

Without that, you are often just contacting a company because it exists.

Why this person?

Is this individual actually connected to the problem you can help solve?

This sounds obvious, but no amount of clever AI copy will save outreach aimed at the wrong person.

Why are you relevant?

What can you genuinely bring to the conversation?

For a recruitment agency, this could be as simple as already representing someone who appears particularly relevant to a vacancy.

That creates a substantially stronger reason to contact the hiring manager than:

"We are a leading recruitment agency with 15 years of experience."

That distinction matters.

Personalisation should change the substance of the message, not just the first sentence.

Your business already knows what good outreach looks like

This is where AI starts becoming much more useful.

Most businesses already have an idea of what good outreach looks like.

The knowledge just isn't documented.

A recruitment founder might know instinctively that:

• three related vacancies are more interesting than one isolated role
• mentioning an available candidate can be more credible than pitching recruitment services
• long agency introductions reduce the strength of the message
• certain phrases immediately make outreach sound automated
• some testimonials strengthen a message while others feel forced
• certain types of companies are unlikely to be worth approaching
• some job adverts indicate a real hiring challenge while others don't

A good salesperson will develop similar judgement in almost any industry.

The problem is that much of this sits inside someone's head.

If you want AI to consistently improve your outreach, you need to start capturing that judgement.

What should an AI outreach system know?

Instead of thinking purely about prompts, think about the information and rules your outreach system should retain.

This is much more valuable than repeatedly giving an AI tool a giant prompt.

Once you've worked out what good looks like, that knowledge should become reusable.

Turn your outreach preferences into a reusable AI skill

I think of this as creating a skill for the AI.

The terminology may vary depending on the tools you use.

The principle is the important part.

Instead of explaining your business every time, you give the AI a reusable set of instructions, examples and context that describes how you want a particular task performed.

For the recruitment founder I mentioned earlier, that skill could contain information such as:

• the sectors his agency recruits for
• the types of businesses he wants to approach
• the problems those companies tend to face
• what constitutes a credible reason for getting in touch
• how candidate information should be incorporated
• his preferred tone
• how long emails should be
• phrases he dislikes
• how much the agency should talk about itself
• examples of messages he likes
• examples he rejected
• lessons from previous revisions

Now imagine he wants to approach another prospect.

Instead of spending twenty minutes explaining everything again, he provides the changing context:

• Company: ABC Logistics
• Contact: Operations Director
• Signal: Recruiting three transport managers
• Additional context: Recently opened a second distribution centre
• Relevant asset: An experienced transport manager already represented by the agency

The skill already knows how the agency wants that information turned into outreach.

That is the important change.

You teach the system once.

You improve it as you learn.

Then you reuse it.

Example: Building a reusable AI outreach skill

Here is an example of what that approach looks like in practice.

In this case, I took the outreach preferences we developed while working with a recruitment founder and turned them into a reusable skill.

Instead of repeatedly explaining the same rules to the AI, those preferences, corrections and examples can be applied to future outreach.

The important part isn't the particular AI tool or interface being used.

It is that the founder's judgement has been captured.

If he later decides:

"I don't want testimonials in the first email anymore."

You update the skill.

If he realises candidate-led outreach consistently produces stronger conversations, that learning can be reflected in the skill too.

If he sees another phrase he never wants his team using, it gets added.

The system becomes better because the business is learning.

How do you automate outreach with AI without making it generic?

This is where I think businesses often start in the wrong place.

They jump immediately to:

"How can I automatically send hundreds of AI-generated messages?"

But sending is probably the easiest part of the workflow.

The difficult part is deciding:

• Who should we contact?
• Why should we contact them?
• What should we say?
• Why would they care?

A better AI outreach workflow looks something like this:

1. Identify the right prospects

Start with companies and people who actually fit your target market.

AI cannot rescue a poor prospect list.

2. Find a meaningful reason to contact them

Look for a signal.

That might be:

• new vacancies
• expansion
• funding
• a leadership change
• a product launch
• a new office
• increased hiring activity
• another event genuinely connected to what you sell

3. Research the company and individual

Gather enough information to understand the situation.

The objective isn't to collect every possible fact about the company.

It is to identify the few pieces of information that materially affect the conversation.

4. Connect the signal to what you can genuinely help with

This is arguably the most important step.

A hiring signal by itself isn't enough.

You need to understand whether you have a credible reason to approach the prospect.

For a recruiter, this might mean asking:

Do we already have someone in our database who could be relevant?

For another business it might mean:

Have we solved this exact problem before?

5. Generate the outreach using your reusable rules

Now AI becomes useful.

It combines:

Prospect-specific context

with:

Your existing outreach knowledge

The AI shouldn't need to rediscover your tone, positioning and preferences every time.

6. Review before sending

Particularly while a system is new, keep a human in control.

Someone should still be asking:

• Is this accurate?
• Is the reason for approaching them strong enough?
• Does this actually sound like us?
• Would I respond to this?
• Has the AI invented anything?

7. Send and manage follow-up

Once the message is approved, the sending and follow-up process can be managed systematically.

8. Learn from what happens

This is the part that often gets missed.

If certain approaches generate conversations and others consistently fail, feed that learning back into the system.

Your outreach process should improve over time.

What should you automate first?

If you are starting with AI outreach, I wouldn't begin by trying to automate the entire process.

Start with the repetitive work surrounding the salesperson.

Research

Instead of manually visiting ten different pages to understand a company, AI can help consolidate relevant information.

Qualification

Rather than treating every prospect equally, you can use defined rules to identify which opportunities deserve attention.

Signal identification

AI can help surface the events or changes that create a genuine reason to contact someone.

Drafting

Once the context exists, AI can turn it into a message using your established outreach rules.

Follow-up preparation

AI can help draft different follow-ups depending on the previous message and available context.

The salesperson still controls the relationship.

The system reduces the repetitive work required to get them to the conversation.

A recruitment example

Recruitment makes this particularly interesting.

Imagine an engineering recruitment agency has thousands of candidates sitting in its ATS.

At the same time, hundreds of companies are advertising engineering roles.

A basic outreach process might be:

Find a company advertising a job → email them offering recruitment services.

A more useful AI-assisted process could be:

Identify a company hiring several relevant engineers → understand what they are hiring for → check whether the agency already represents relevant candidates → identify the appropriate decision-maker → prepare an outreach message built around the actual opportunity → recruiter reviews → send.

The difference is significant.

The first approach uses AI to produce more messages.

The second uses AI to help create better reasons for having conversations.

That is the distinction I think matters.

AI outreach shouldn't mean sending more rubbish

There is an obvious danger with all of this.

AI makes it incredibly easy to generate enormous amounts of outreach.

That doesn't necessarily make outreach better.

In fact, it can make the problem worse.

If every business can generate 5,000 superficially personalised emails, then phrases like:

"I noticed you recently…"

become less meaningful.

The advantage is unlikely to come from who can generate the most text.

It will come from businesses that are better at deciding:

• who deserves attention
• what matters about them
• when to approach
• what creates a credible reason to talk
• what information their own business already has that makes the conversation useful

AI should amplify judgement.

It shouldn't replace judgement with volume.

Stop correcting the same mistakes

There is a simple way to identify where you should start.

Open some of your recent conversations with ChatGPT or Claude.

Look at the instructions you keep repeating.

Maybe you regularly write things like:

"Make it less generic."

"Shorten it."

"Stop talking about us."

"Don't use that phrase."

"Make it sound less salesy."

"Focus on their actual problem."

"Use the job vacancy as the reason for reaching out."

"Don't include a testimonial unless it is relevant."

Those corrections contain your business knowledge.

If you keep making the same correction, it probably shouldn't remain a correction.

It should become a rule.

If you keep giving the AI the same piece of context, it probably shouldn't remain a prompt.

It should become part of the system.

Start with one outreach workflow

You don't need to build an enormous AI platform.

Start with one repetitive workflow.

Take ten outreach messages you think are genuinely strong.

Then take five or ten you would never want someone from your company sending.

Ask:

• What makes the good examples good?
• What makes the bad examples bad?
• What information changes the quality of a message?
• What signals make a prospect worth approaching?
• What should the AI always know about our business?
• Which instructions do we repeatedly give it?
• Which decisions still require human judgement?

Document the answers.

Turn them into reusable instructions.

Test them on new prospects.

Improve the system when you learn something.

That gets you much closer to using AI as part of your business development process rather than simply using it to write emails.

Frequently asked questions

Can ChatGPT write personalised cold emails?

Yes, but the quality depends on the context you give it.

Providing only a company name, contact and job title will usually result in superficial personalisation. Better outreach combines prospect-specific information with a clear reason for contacting them and reusable knowledge about how your business communicates.

How do you personalise outreach using AI?

Start with commercially relevant context.

Ask why this company is worth approaching, why now is an appropriate time, why this individual is the right person and why your business is relevant.

AI can then use that context to write the message. Personalisation should affect the substance of the outreach, not simply add a personalised opening sentence.

How can I automate cold outreach with AI?

Break the workflow into stages rather than automating message sending immediately.

A useful system might:

1. identify relevant prospects
2. find meaningful triggers
3. research the company
4. identify the right contact
5. connect the opportunity with your offer
6. draft outreach using reusable rules
7. send it for human review
8. manage follow-ups
9. learn from responses

Automate the repetitive preparation first. Then decide how much of the sending process should also be automated.

How do I stop ChatGPT from writing generic outreach?

Don't rely entirely on one-off prompts.

Give the AI reusable information about your customers, positioning, tone, personalisation rules, examples of good outreach, examples of poor outreach and phrases you want avoided.

Most importantly, provide a meaningful reason for contacting each prospect.

Is a better AI prompt enough for outbound sales?

For occasional messages, a good prompt can help.

If outreach is a recurring business process, reusable context is much more valuable. Your AI should already understand the parts of your approach that don't change from prospect to prospect.

What is an AI outreach system?

An AI outreach system is a repeatable workflow that combines prospect information, buying or activity signals, business context, outreach rules and AI to help identify, research and approach potential customers.

It may automate some parts of the process while leaving important decisions and sending under human control.

From AI writing to AI workflow

The biggest shift is relatively simple.

Don't think:

"How can I get ChatGPT to write this email?"

Think:

"What does our business already know about doing this well, and how can we make that knowledge reusable?"

Once that happens, AI stops being something you visit every time you need some copy.

It starts becoming part of the process.

You teach the system once.

You improve it as you learn.

You reuse what works.

That is the difference between using AI as a writing assistant and actually building it into your outreach workflow.

At Underpin Works, I'm exploring practical ways recruitment businesses can apply AI to business development, research and repetitive recruiter workflows without simply generating more generic outreach.

If you're spending more time correcting ChatGPT or Claude than you're saving, the problem might not be the prompt.

It might be that the workflow itself needs designing.