Generic output comes from generic input. The AI prompts for sales that actually work are scoping instructions, not message templates: you name the signal that matters, name the accounts in play, say what the output should look like, then read the shortlist and interrogate it in chat. You write the send yourself. By the end of this you will have a ranked list of accounts, a reason each one ranked where it did, and a draft you can edit before it goes anywhere.
TL;DR
- A sales prompt is a scoping instruction, not a copy generator. It tells the agent which signal to weight and which accounts to rank, and the output is a shortlist, not a paragraph.
- Generic output is almost always a thin prompt. A prompt with no named signal, no account set, and no output format gets a generic answer because nothing in it narrows anything.
- The step most reps skip is the follow-up question. Asking "why did this account rank above that one" turns a list into something you can defend on a call.
- Tealde surfaces, ranks, and drafts. You review, you edit, and decide when and what to send. Nothing leaves your hands without you reading it first.
What you need before you start
- A Tealde account with the chat interface open
- A clear ICP: industry, company size band, and the two or three titles you actually sell to
- One buying signal you trust, named specifically (hiring for a role you sell into, a funding round, a tool change, a prospect commenting on your problem space)
- Fifteen minutes with no calls booked
Time to complete: about 20 minutes Difficulty: Beginner
What should you decide before you begin?
Decide which signal you are willing to act on before you write anything. A prompt is only as narrow as the signal you put in it, and "companies that might need us" is not a signal. Pick one you could explain to your manager in a sentence: "they posted three SDR roles in the last month" or "a director commented on a post about manual CRM updates." If you cannot name it, the output will be as vague as the input.
Step 1: Name the signal, not the outcome
Start by telling the chat what evidence you care about rather than what you want to happen. "Find me good leads" has no signal in it. "Show me accounts hiring for revenue operations roles in the last 30 days" does.
- Write the signal as something observable with a date range.
- Add the one disqualifier that most often wastes your time, like headcount under 50 or a region you do not cover.
- Send that as your first prompt, with no mention of messaging yet.
Check: the response comes back as accounts with the signal stated next to each one, not as a paragraph of advice about prospecting.
Step 2: Scope the account set
Narrow to a universe you would actually dial. The live lead database covers far more accounts than you can work in a quarter, so the prompt's job is to cut, not to collect.
- Name the industry and the company size band.
- Name the two or three titles you sell to, by title, not by seniority level.
- State how many you want back. Thirty is a workable day. Three hundred is a spreadsheet you will never open.
Check: the shortlist length matches the number you asked for, and the titles on it are ones you recognize from your closed-won deals.
Step 3: Ask for the list ranked, and say what by
Ranking is where a list becomes a call order. Lead scoring runs in the background, but the prompt decides what gets weighted, so say it out loud.
Tell the chat which signal outranks which. "Rank by recent hiring activity first, then by whether anyone on the account has engaged with content about manual reporting." That ordering is your judgment, not the agent's, and stating it is the difference between a sorted list and an arbitrary one.
- Name the primary ranking signal.
- Name the tiebreaker.
- Ask for the score and the reason to be shown together.
Check: the top five accounts each show a visible reason, and you agree with at least three of them.
Step 4: Ask why a lead ranked where it did
This is the step most reps skip, and it is the one that makes the rest useful. Pick the account at the top and ask the chat directly: "why did this one rank above the rest?"
The answer gives you two things. It tells you whether the ranking is built on something real or on a weak proxy, and it hands you the opening line of your call. If the reason is "three open roles in revenue operations and a VP who changed jobs last quarter," you now know what to lead with. If the reason is thin, you know to reweight and ask again.
Check: you can state in one sentence why the number one account is the number one account.
Step 5: Ask for a prospect read, not a message
Before you draft anything, ask for a read of the contact. Three bullets on role, company context and the likeliest pain point, tied to the signal that put them on the list.
This is a briefing, not copy. You are asking what this buyer is probably dealing with, so you can decide whether it is worth the touch at all. Half the value is the accounts you cross off.
- Ask for three bullets: role and scope, company context, likely pain point.
- Ask it to cite the specific signal behind the pain point guess.
- Ask what would make this account a bad fit.
Check: each bullet references something specific about that account, and none of them would read identically for the next contact on the list.
Step 6: Ask for a draft, then rewrite the first line
Now ask for a draft. Give it constraints or it will give you filler: a length, a single point to make, one ask, and a ban on anything you would not say out loud.
Then rewrite the opening line yourself. The draft gets the facts in the right order. Your first line is what makes it sound like a person who read the account page. Tealde drafts and you send, so this edit is the whole job, not a formality.
- Specify the length in sentences.
- Specify one value point, not three.
- Specify the ask: a 10-minute call, a yes-or-no question, whatever you actually want.
- Replace "make it better" with the specific change: "cut to three sentences, drop the jargon, keep the hiring reference."
Check: you have a draft you would send after one edit rather than a rewrite.
Step 7: Save the prompt as a scheduled agent
Once a prompt gives you a shortlist you trust, set it to run on a schedule so the ranked list is waiting for you instead of being something you rebuild each morning. Scheduled agents re-run the scope as new signals land.
- Save the prompt wording that produced the list you liked.
- Set the cadence that matches how fast your signal moves. Hiring signals warrant weekly. Funding warrants less often.
- Leave the ranking weights as you tuned them in Step 3.
Check: the next time you open Tealde, a ranked list is already there with new accounts flagged against your signal.
Is a sales prompt the same as a message template?
No. A message template is output you fill in; a sales prompt is an instruction about what to look for. The template assumes you already know who you are writing to, which is the part that is usually wrong.
A template cannot tell you that twelve of your thirty target accounts just started hiring for the role you sell into. A scoping prompt can, and it reorders your day around that. Templates become useful after the prompt has done its work, at the drafting step, which is the last five minutes of the process rather than the first.
Why does AI give me generic sales output?
Because the prompt had nothing specific in it to work with. Thin inputs produce thin output, every single time, and no amount of rephrasing fixes a prompt that never named a signal, an account set or an output format.
Three things are usually missing:
- No named signal. "Companies that could use our product" describes your whole market. "Companies that posted two or more SDR roles this month" describes forty of them.
- No constraints. Without a length, a format or a single point to make, the model fills the space with the safest sentences it knows.
- No output shape. If you do not say "a ranked list with the reason next to each account," you get prose, and prose is not something you can work.
The other cause is asking for a message before asking for a read. A message request with no account context has nothing to be specific about, so it reaches for the generic.
What are prompt examples for sales prospecting?
Here are four that work because each one names a signal, a scope and a format. Adapt the nouns to your territory.
- Scoping a signal: "Show me accounts in logistics software, 100 to 500 employees, that have posted a revenue operations or sales ops role in the last 30 days. List the account, the role, and the posting date."
- Ranking a set: "Rank these 40 accounts by recent hiring activity first, then by engagement with content about manual reporting. Show the score and the reason for each."
- Reading a lead: "Give me three bullets on this contact: role and scope, company context, and the likeliest pain point based on the hiring signal. Name the signal you used."
- Drafting the touch: "Draft a three-sentence first touch referencing the two open ops roles. One value point, one ask for a 10-minute call, no buzzwords."
Follow-ups matter as much as the first prompt. Once you have a list, chaining is just asking the next question in the same conversation: "narrow to the ones in my territory," "drop anyone we have an open opportunity with," "why is this one third and not first." Each answer builds on the last, which is why you keep it in one thread rather than starting over.
Prompting beats filtering manually when the thing you care about is recent and not a filter field. Titles and headcount are filters, and you can set those yourself in seconds. "Started hiring for this role in the last three weeks and someone senior just engaged with our problem space" is not a filter, and that is exactly when signal intelligence earns the extra step. Salesforce found that reps spend around 60% of their time on non-selling tasks (opens in a new tab), and most of that is this: assembling context by hand that a scoped prompt returns in one pass.
What goes wrong, and how do you fix it?
The most common failure is a prompt that asks for a message when it should have asked for a list, which guarantees generic copy because nothing in the prompt is specific to anyone.
- The output reads like a blog post: You did not name an output format. Add "return a ranked list with the account, the signal, and the reason" to the end of the prompt.
- The list is full of accounts you cannot sell to: Your disqualifier is missing. State the hard requirement explicitly: region, headcount floor, or exclude anyone already in an open opportunity.
- Every lead read sounds the same: You asked for a summary without naming the signal behind it. Ask it to cite the specific evidence for the pain point, and cross off any where it cannot.
- The ranking does not match your instinct: Ask why the top account ranked there. If the reason is a weak proxy, change the weighting in the next message and ask again.
- You keep rebuilding the same prompt: Save it and schedule it. A prompt you retype every Monday is a prompt you will eventually stop using.
Common questions about AI prompts for sales
Should AI write my outreach message or should I?
You write it, and you use the draft as a starting point. Tealde drafts the structure and gets the facts in the right order, which saves you the blank-page problem. The first line and the ask should be yours, because those are the parts a buyer reads closely and the parts that reveal whether anyone actually looked at their account.
How do I stop AI copy sounding robotic?
Give it one point to make, a hard length, and a specific detail to reference. Robotic copy is usually overstuffed copy: three value props, two questions and a sign-off in four sentences. Ask for three sentences with one ask, then rewrite the opening line in your own words.
What do I do when the prompt returns the wrong leads?
Add the disqualifier rather than rewriting the whole prompt. Most wrong-lead problems are one missing constraint: a region you do not cover, a headcount band that is too wide, or accounts your team already owns. Say it in the next message in the same thread and the list narrows without you starting over.
How do I adapt prompts for a different industry or segment?
Keep the structure and swap the nouns. The signal, the size band, the titles and the pain point all change by segment, but "name the signal, scope the set, rank it, ask why" holds everywhere. Build one prompt per segment you sell into and save each as its own scheduled agent.
What if I only have ten minutes a day for prospecting?
Open the ranked list a scheduled agent already produced, and spend the ten minutes on the top five. Skip the drafting prompts entirely and ask only for the lead read, three bullets per account. That gives you five calls you can make with context, which beats thirty you make blind.
Which prompt should an SDR start with?
Start with the lead read. Asking for three bullets on a contact you were already going to call is the lowest-risk way to see whether the output is specific enough to trust. Once you believe the reads, move upstream to scoping and ranking, which is where the real time saving sits.
Do these prompts work in ChatGPT or only in a sales tool?
The writing rules transfer anywhere: name the signal, name the constraints, state the output format. What does not transfer is the data. A general-purpose chat tool has no live account signals to rank against, so it will answer confidently about accounts it knows nothing current about. Scoping prompts only return a usable shortlist when there is a live lead database behind them.
Which buying signals can I name in a prompt?
Name anything observable and recent: hiring for a role you sell into, a funding round, a leadership change, a tool or technology change, or a prospect engaging with content about the problem you solve. The test is whether you could point to the evidence on a call. If you cannot, it is a hunch, and a hunch in a prompt produces a list you will not trust.
