How to Write Better AI Prompts for B2B Sales

1. Research a prospect without inventing a reason to buy

We sell [solution] to [ideal customer profile], helping them [outcome you can evidence]. I'm researching [Prospect] to decide whether to pursue the account.

Use the company website, recent filings or annual reports, earnings commentary, and credible news. Give me:

• The business model and current stated priorities, in five lines
• Three recent changes that could affect those priorities, with source links and dates
• Three problems we might be able to address, labeled as hypotheses
• For each hypothesis, the evidence that weakens it

Use current public sources. If you cannot access the web or the sources I've supplied, say so clearly and don't fill the gaps from memory.

Don't infer a pain point from the industry or job title alone. Distinguish facts, reasonable inferences, and unknowns.

Why it works: it turns research into qualification and stops the model from manufacturing a convenient buying trigger.

Follow-up: "Which two hypotheses are best supported, and what would I need to verify in a first conversation?"

2. Prepare for a first meeting

I have a [duration]-minute first meeting with [Name], [Role] at [Company], on [date]. We provide [solution and outcome]. The meeting was arranged because [known reason].

My objective is to find out whether there's a real problem worth exploring, not to pitch.

Here's the research and correspondence so far: [paste]

Give me:

• An opening that confirms the purpose of the meeting in two sentences
• Five prioritized discovery questions, each tied to a specific unknown
• Three assumptions I should avoid making
• One way this meeting creates value for the buyer, even if there's no opportunity

Keep the language conversational. Don't write a word-for-word script.

Why it works: it prepares the seller to listen and adapt, rather than perform an AI-written script.

Follow-up: "Role-play a skeptical buyer and challenge these questions before I use them."

3. Find the blind spots in an outreach message

Here's a prospecting message I'm about to send: [paste the email, LinkedIn message, or voicemail script].

The reader is [role] at [type of company]. My legitimate reason for contacting them is [reason and the evidence for it].

Review it from their point of view and identify:

• What's genuinely relevant to them, and what's really about us
• Any personalization that isn't supported by the evidence I've given you
• Phrases that sound generic, inflated, or AI-written

Then rewrite it in under [word count] words, keeping only claims the evidence supports.

Give me one version, not a list of alternatives, and explain the three most important changes.

Why it works: the critique comes before the rewrite, so the model exposes weak thinking instead of polishing it.

Follow-up: "On a busy Tuesday morning, does this get a reply? Why or why not?"

4. Pressure-test a cold call opener

I'm planning to use this opening on a cold call: [paste your opener].

Evaluate it against five criteria: does it lower resistance, is it easy to answer, does it sound conversational rather than scripted, is it about them rather than me, and does it earn the right to keep talking?

Where the answer is no, explain why, then rewrite it so it meets all five. Give me two versions with different tones.

Why it works: five explicit criteria produce specific feedback. "Improve this opener" produces a rewrite you can't evaluate.

Follow-up: "Give me three plausible responses I might get, labeled as hypotheses rather than predictions, and how I'd handle each."

5. Prepare discovery questions

Discovery call with a [role] at a [industry] company of around [size]. We're exploring [the problem you think exists].

Give me eight questions grouped by what each is designed to uncover: the business problem, its measurable impact, who else is involved, and how urgent it is.

For each question, tell me what an answer would need to contain for the point to count as established, and what kind of answer would leave it unestablished and warrant a follow-up.

Then flag the two most likely to make me sound like every other vendor, and replace them.

Why it works: asking what "established" looks like is what turns a question list into preparation. It also avoids treating every vague answer as deliberate evasion — buyers are often just busy or genuinely unsure.

Follow-up: "Which of these questions would a buyer find presumptuous this early in the relationship?"

6. Find what you missed on a call

Review this sales call transcript as an experienced B2B sales coach: [paste transcript].

Identify moments where I could have asked a stronger follow-up, clarified business impact or urgency, tested an assumption, or confirmed a commitment.

For each, quote only the minimum words needed to locate it, explain why it mattered, and give me the better question.

Then: the two coaching priorities that would most improve my next calls. Don't infer tone, intent, or emotion unless the transcript supports it.

Why it works: it anchors coaching to observable moments rather than producing a generic scorecard.

Follow-up: "What did the customer say that I should be worried about, and what did they say that's more encouraging than it sounds?"

7. Turn call notes into a CRM update

Convert the notes below into a CRM update, using only information explicitly stated in them.

Return: customer objective, current problem, business impact, stakeholders named and their stated roles, decision process and timing, competition mentioned, risks and open questions, our commitments, their commitments, next step with owner and date, and a recommended opportunity stage with a one-sentence rationale.

Use "Not established" for anything missing. Do not turn assumptions into CRM facts.

Our stage definitions: [paste your organization's actual definitions].

Call notes: [paste]

Why it works: "Not established" is more valuable than a confidently completed but inaccurate record. It also makes the gaps visible rather than burying them.

Follow-up: "Which three missing items create the greatest deal risk, and how would I resolve them?"

For Salesforce users, this can become a repeatable workflow rather than a prompt that every salesperson runs separately. Agentforce Pipeline Management can use connected notes, emails, and calls to suggest updates to opportunity fields such as Stage and Next Step, and the salesperson can see the source context and review, edit, or approve the proposed update before it changes the record.

The quality of the result still depends on the organization's stage definitions, source data, and business rules. Automating an ambiguous sales process only produces ambiguous CRM updates more efficiently.

If your stage definitions aren't clearly written enough to paste into a prompt, that's worth fixing first — see our guide to Salesforce opportunity stages.

8. Pressure-test a proposal through the buying committee

Below is a proposal we're about to send to [company] for [value].

[Paste.] Known stakeholders and their decision criteria: [list them; say so if unknown].

Review it four times, once from the perspective of: the CFO, the operational lead who'll use it, the technical stakeholder, and procurement. Use the stated decision criteria where I've given them. Treat anything inferred from a job title as a hypothesis, not a known priority, and label it as such.

For each reader, tell me:

• What they'll object to first
• What they'll skim past that we needed them to read
• What's missing that they'd need in order to say yes

Don't rewrite the whole proposal. Give me the five changes most likely to improve their decision quality, with revised wording for those sections only.

Why it works: it treats the proposal as a decision document rather than marketing copy, and the four-reader structure surfaces the objection that kills deals in committee.

Follow-up: "What would a skeptical CFO challenge first, and what evidence would answer it?"

9. Prepare for price pressure without reaching for a discount

A customer has asked for [discount or concession] on a [value] proposal.

Here's the proposal, the commercial context, and the exact wording of their request: [paste].

First, set out the plausible explanations: a genuine budget constraint, standard procurement practice, insufficient perceived value, or an unvoiced risk concern. For each, tell me what evidence supports it, what contradicts it, and what we don't know. Don't pick one unless the evidence clearly justifies it.

Then give me:

• Three questions that would distinguish between those explanations
• Three responses that hold the price, in language I'd actually say out loud
• Two concessions cheaper than the discount, and what we should ask for in return

Also, tell me the weaknesses in our value case that they're most likely to exploit.

Why it works: it forces diagnosis before prescription, without letting the model invent a buyer motive to be confident about. Most objection-handling prompts jump straight to responses, which is how sellers end up answering a question the buyer didn't ask.

Follow-up: "Role-play the procurement lead. Don't make this easy for me."

Related reading: 9 Expert Tips That Improve Discount Control in Salesforce

10. Challenge a forecast rather than summarize it

For sales leaders and RevOps.

Review the opportunity data below as a sales leader preparing the forecast. Use our stage definitions and forecast rules exactly as written.

For each material deal:

• The evidence supporting the current stage and forecast category
• Missing evidence or contradictions
• Close dates not supported by a customer event or decision plan
• Whether what's recorded is a customer commitment or a salesperson's activity
• The next question the manager should ask the rep

Then an executive summary: likely outcome, range of uncertainty, and the assumptions driving it.

Don't change stages, probabilities, amounts, or dates.

Don't treat a detailed CRM record as proof that a deal is healthy.

Stage definitions and forecast rules: [paste].

Opportunity data: [paste approved export].

Why it works: it looks for evidence and inconsistency rather than converting optimistic CRM data into polished commentary. The distinction between customer commitment and salesperson activity is where most forecast error lives.

Follow-up: "Which three assumptions, if wrong, would change the forecast most?"

Related reading: Why an Expected Revenue Forecast Stands Up to Scrutiny.