Most rep-facing prompt libraries in 2026 are junk. Generic prompts, generic outputs. I've watched hundreds of reps try to work an LLM into their day and give up because the outputs were too shallow to trust.
The reps who make it work do two things. They write prompts that carry deal context, not category context. And they know where the LLM will lie. This playbook is those two things — 15+ real prompts, organized by workflow, with the failure modes called out.
Account research prompts
1. Company one-pager
``` You are a senior enterprise sales rep researching a target account.
Company: [company name] Website: [URL] My product: [one-sentence description] My ICP: [one-sentence description]
Produce a one-page brief with:
- Company overview (2 sentences, no adjectives)
- Recent material events in the last 6 months (funding, exec hires, product launches, layoffs) — cite source URLs
- Likely business priorities inferred from those events (bullet list, 3-5)
- Top 2 hypotheses for why they'd care about my product
- Top 2 reasons they might not care right now
Do not speculate on private financials. If you can't find something, say so. ```
The instruction to say "I don't know" is the load-bearing line. Without it, LLMs invent details.
2. Recent news scan
`` List every material news event about [company] in the last 90 days. For each, provide: date, one-sentence summary, source URL. Skip PR fluff. Skip anything without a source. ``
Use this before a first call. Catching something the buyer didn't expect you to know earns credibility.
3. Product / roadmap intelligence
`` What products does [company] currently offer, and what has their public roadmap communicated in the last 12 months? Prioritize press releases, product launch pages, earnings-call mentions, and conference keynotes. Provide source URLs for every claim. Do not infer roadmap items from vague marketing copy. ``
Buying-committee mapping prompts
4. Committee inference
``` Given the following:
- Company: [company]
- Product I sell: [product]
- Typical buying committee for a [category] deal at a company this size
List the 6-10 stakeholders who would typically be involved, with:
- Role title
- Their likely primary concern about the purchase
- Their likely veto power (high/medium/low)
- One question I could ask a champion to identify who fits each role
Reference: Gartner's B2B buying committee sizing (6-10, up to 11 stakeholders). ```
This gives you a straw-model committee before discovery. Then discovery is about confirming or correcting it, not building from scratch.
5. Champion identification hypothesis
``` Given [company]'s org chart / LinkedIn presence, who is most likely to be an internal champion for [my category]? Criteria:
- They've publicly written or spoken about this problem
- They've changed roles into this space in the last 24 months
- They've hired for adjacent skills recently
- They follow / interact with competitors on LinkedIn
Rank 5 candidates with brief reasoning per each. ```
Competitive positioning prompts
6. Head-to-head positioning
``` I sell [my product]. My prospect is comparing me to [competitor]. Given both products' public positioning:
- What are the 3 dimensions where my product is stronger?
- What are the 3 dimensions where the competitor is stronger?
- What are the 2 dimensions where the buyer likely can't tell the difference?
Only use public information. Do not invent feature comparisons. ```
Without the last instruction, LLMs confidently invent feature parity or feature gaps.
7. Objection map
``` List the top 5 objections a buyer might raise when comparing my product [product] against [competitor], sorted by how likely each is to come up. For each objection:
- What the buyer is really worried about
- Two possible responses (one honest, one defensive — label which)
- What data or reference customer would help resolve it
```
The "label which" instruction forces the LLM to separate real answers from spin. Reps who use spin lose deals.
Discovery question generation prompts
8. SPIN-shaped discovery generator
``` I'm running a discovery call with a [title] at [company] about [problem area].
Generate:
- 2 situation questions I couldn't have answered from their website
- 3 problem questions to surface their current friction
- 4 implication questions that expand consequences into board-level or strategic terms
- 2 need-payoff questions the buyer can answer in their own words
Keep questions short. No jargon. No leading language. ```
The LLM won't get the questions perfectly right, but it gets you 80% there in 90 seconds.
9. Multithreading question generator
``` I've been talking to a [title] as my main contact at [company]. Given the buying committee for [category] deals at similar-sized orgs:
- What are 3 questions I should ask my current contact to surface who else needs to be in the deal?
- What are 2 questions I should ask the CFO or their equivalent if I get access?
- What are 2 questions I should ask the end-user team?
```
Multithreading is where deals die. Gartner data puts buying groups at 6-10 stakeholders. Deals with 40-55% more stakeholders engaged in stages 2-3 close at meaningfully higher rates.
Executive research prompts
10. Exec one-pager
``` Executive: [name] Company: [company] Role: [title] LinkedIn: [URL]
Produce a one-pager:
- Career history in 3 bullet points (companies, roles, dates)
- Publicly stated priorities in the last 12 months (podcast quotes, blog posts, keynote talks)
- Signals of what they care about (bullets)
- Two people from their past companies who I might have a mutual connection with
- One angle for a warm intro request
Cite URLs for anything specific. Say "not found" if you can't verify. ```
The last two bullets are where LLMs are weakest — they don't know your network graph. LLM + relationship graph beats LLM alone. More on that below.
11. Podcast / talk mining
`` Find every podcast episode, conference keynote, or webinar featuring [executive] in the last 24 months. For each: date, host/venue, one-sentence summary of what they said about [category or topic]. Include source URLs. ``
Referencing specific things an exec has said publicly earns credibility fast.
Objection response prompts
12. Draft objection response
``` Objection I heard: "[objection verbatim]" Buyer context: [1-2 sentences about who and why] My product: [one-liner]
Draft:
- What the buyer is likely really worried about (2 hypotheses)
- A concise response (3-4 sentences max)
- One clarifying question I could ask before responding
Do not defend. Do not sell. Acknowledge, clarify, then respond. ```
The "do not defend, do not sell" instruction is the whole game. Left to their defaults, LLMs write objection responses that sound like a 2015 sales script.
13. Follow-up email after a tough call
``` I just had a discovery call with [title] at [company]. They raised [objection]. My prior context: [1-2 sentences]. Draft a follow-up email that:
- Acknowledges the objection directly (no dodging)
- Adds one piece of information the buyer wouldn't have gotten from a public source
- Suggests one concrete next step (not "let me know if you have questions")
- Is under 120 words
Tone: direct, honest, no jargon. Do not use the words "circle back," "touch base," or "just checking in." ```
Banned-phrase lists are the most underused prompt technique. Naming what you don't want fixes the default cliché output.
Call recap prompts
14. Call notes to CRM update
``` Below are my rough notes from a call with [contact] at [company]. Produce:
- CRM-ready summary (5 bullets max)
- Explicit next steps with owner and date
- Any risk signals I should flag
- Any competitor mentions
Notes: [paste raw notes] ```
15. Follow-up email draft
``` Below are my raw notes from a call. Draft a follow-up email:
- Recap what we agreed to
- Attach the 2 specific artifacts I promised
- Confirm next step with date and owner
- Under 150 words
- Direct, no fluff
Notes: [paste raw notes] ```
16. Coaching prompt for the manager
``` Below is a call transcript with a rep and a buyer. Analyze:
- Where did the rep miss a chance to go deeper?
- Where did they retreat to product when they should have stayed on the problem?
- What implication or need-payoff question should they have asked?
- One concrete coaching note (single sentence).
Transcript: [paste] ```
What LLMs are good at vs where they hallucinate
The prompts above only work if you know where to trust the output. Here's the honest map.
What LLMs are genuinely good at
- Synthesis. Give an LLM 20 sources and ask for a one-pager, and you get a usable one-pager.
- Structure. Turning messy input into clean output. Notes to CRM entries. Transcripts to bullets. This is where the time savings live.
- Framing. Rewriting the same content for different audiences. Same case study, three different exec framings.
- Generating options. Give me five subject lines. Give me three angles for a follow-up. The LLM is good at breadth here.
Where LLMs still hallucinate consistently in 2026
- Recent news. Anything under 60 days old is unreliable. LLMs invent quotes, misattribute deals, mix up companies.
- Private company data. Revenue, headcount, funding rounds unless heavily reported. LLMs will fill in confident numbers that are wrong.
- Executive reporting lines. LLMs invent org charts. "Who reports to whom" is a common failure mode.
- Specific quotes. LLMs will attribute quotes to execs who never said them. If a quote matters, verify the source.
- URLs. LLMs fabricate URLs constantly. Any URL an LLM produces has to be checked. Every one.
- Competitor feature comparisons. LLMs make up features that don't exist or claim feature parity that isn't there.
The rule that saves reps time: if the fact is load-bearing, verify. If the fact is texture, use it.
Where warm graph plus LLM equals 10x research quality
Here's the piece most sellers miss. LLMs synthesize public data. They have no view into who at your company knows who at the prospect. They don't know that your VP of Engineering used to work with the prospect's CTO at a prior company, or that your customer champion at Company A is a college friend of the CFO at Company B.
That data lives in your organization's relationship graph — calendars, inboxes, LinkedIn, message threads, deal history. Combine LLM synthesis with structured relationship intelligence and research quality jumps by an order of magnitude. The LLM tells you what to say. The graph tells you who can actually get the meeting.
At Armis, combining LLM synthesis with warm-path discovery eliminated 1,400+ hours of manual research. Those hours got reinvested into deeper champion work and better multithreading, both of which show up in win rate.
The rep using ChatGPT for account research and the rep using ChatGPT + warm-graph are not doing the same work. One is prepping a cold call. The other is prepping an intro-led conversation with someone who already trusts a mutual contact.
Frequently asked questions
What's the best LLM for sales research in 2026? Depends on the workflow. Claude and GPT-4-class models both perform well on synthesis. Perplexity is stronger for recent-news scanning because it grounds against live search. The right answer is usually "test all three on your five most common workflows and pick per workflow."
How do I stop ChatGPT from hallucinating account data? Two techniques. First, prompt for source URLs on every claim and reject unsourced facts. Second, verify anything load-bearing before using it in a customer conversation. The LLM will not check itself. You have to.
Should I let ChatGPT draft entire emails to buyers? Draft, yes. Send without editing, no. LLM-drafted outbound sent without edits reads as generic to buyers and tanks response rates. Draft, edit, personalize, send.
Are my prompts safe to use? Does OpenAI train on them? Depends on the plan. Enterprise ChatGPT and Claude Team explicitly do not train on your data. Free tiers vary. If you're pasting confidential account context or customer names, use an enterprise-tier plan.
How much time does LLM-assisted research actually save? Realistically, 3-6 hours per rep per week on account research, meeting prep, and post-call notes. Gartner's data shows the average saving is about 5 hours weekly. The catch: 72% of orgs fail to reinvest that time into activities that compound.
What's the single highest-ROI prompt for a seller to use? The account one-pager combined with committee inference. Together they get a rep from cold to prepared in ten minutes and set up every subsequent conversation.
