ChatGPT for B2B Sales: The Prompts We Actually Use (DACH Edition)
Most ChatGPT-for-sales guides list 77 prompts nobody uses twice. Here are the six jobs where ChatGPT genuinely earns time in our outbound work, the exact prompt structure behind them, and the two rules that keep it from embarrassing you in German.
Key takeaways
- ChatGPT earns real time in six sales jobs: account research summaries, ICP hypotheses, first drafts, objection preparation, call preparation and follow-up structuring. Everything else is demo material.
- The prompt structure that works is always the same: role, goal, context, constraints, format. Vague prompts produce the generic filler your prospects already recognize.
- Rule one: nothing goes out unedited. Rule two: German output needs a native rewrite pass, because LLM German reads like translated English and Mittelstand recipients notice in one sentence.
- The gap between teams is not access to AI, it is the quality of the inputs: a sharp ICP and real research beat a clever prompt every time.
The ChatGPT-for-sales genre has settled into a formula: a listicle of 77 prompts, each used once, none surviving contact with a real pipeline. This is not that. We run outbound for B2B clients in the German-speaking market every day, ChatGPT and Claude are embedded in that work, and what follows is the short, honest list of jobs where they genuinely earn time, with the prompt patterns we actually reuse.
The structure behind every prompt that works
Generic in, generic out. The prompts that survive in daily use all share five parts:
- Role: who the model is supposed to be ("You are an SDR researching a logistics company...").
- Goal: the one deliverable ("...to find a concrete reason to contact them this week").
- Context: the raw material you paste in: website text, LinkedIn profile, job postings, your own offer in two sentences.
- Constraints: what to avoid ("no superlatives, no marketing language, maximum 40 words, formal Sie").
- Format: exactly what the output looks like ("three bullet points, each one sentence").
Miss one of the five and you get filler. The teams disappointed by ChatGPT are almost always feeding it prompts like "write a cold email", which is a request for the average of every bad cold email it has read.
Job 1: Account research summaries
The highest-value use, and the least glamorous. Paste in the website's about page, two job postings and a news item, then: "Summarize in five bullets what this company does, whom it sells to, and what changed in the last six months. Then name the single most likely reason they would care about [your offer in one sentence] this week." The output is not a message, it is the raw material a human turns into one. At list scale, this same pattern runs inside Clay columns, which is where enrichment and drafting meet.
Job 2: ICP hypotheses worth testing
"Here are the ten customers where we won fastest [paste short descriptions]. What do they have in common that our website does not mention? Suggest three sharper ICP hypotheses, each with a way to identify matching companies from public data." The model is genuinely good at spotting patterns humans are too close to see, and each hypothesis becomes a testable ICP segment.
Job 3: First drafts, never final drafts
The rule that keeps this useful: ChatGPT writes draft one, a human writes draft two, and only draft two exists as far as the prospect is concerned. Ask for three variants with different opening angles rather than one "perfect" email, because variety is where drafts help and polish is where they mislead. The frameworks the human applies in draft two are in our copywriting guide.
Job 4: Objection preparation
"You are a skeptical [CFO of a 200-person machine builder]. We will pitch you [offer]. Name the seven objections you would raise, ordered by how likely they kill the deal, and for each the evidence that would change your mind." Sales teams rehearse against this before real calls, and the ordering is usually uncomfortably accurate.
Job 5: Call preparation in five minutes
Paste the prospect's LinkedIn profile, company page and your CRM notes: "Build a one-page call brief: three things this person cares about professionally, two recent company events worth referencing, one risk in this conversation, and the opening question you would ask." This replaces twenty minutes of tab-hopping before every discovery call.
Job 6: Follow-up structuring
After a call, paste your rough notes: "Structure into: decisions made, open questions, agreed next steps with owners. Then draft a five-sentence follow-up email in formal German, no filler phrases." Note the human still reads it, because rule one never bends.
The two rules that keep this from backfiring
Rule one: nothing ships unedited. Not because the drafts are bad, but because accountability cannot be delegated. The prospect is talking to your company, not to a model.
Rule two: German output gets a native pass, always. LLM German has a recognizable accent: too enthusiastic, oddly direct, Sie/Du wobbles, idioms no German businessperson writes. In a market where recipients delete machine outreach on pattern recognition, this is not cosmetic, it is the difference between a reply and a spam report. The full argument about where AI helps and fails in this market is in our KI im Vertrieb guide, and the tool stack around it in AI sales tools.
The honest limit
ChatGPT compresses preparation, it does not create demand. A sharp ICP, real signals and a working outbound system decide whether there is a pipeline; the model decides how fast the humans inside that system move. Teams that get this order right save hours every day. Teams that get it backwards automate their way into spam folders faster than ever before.
Frequently asked questions
Can ChatGPT write cold emails in German?
It can draft them, and the drafts read like translated English: too enthusiastic, oddly direct, with formal-address wobbles. German-speaking recipients recognize the pattern in one sentence. Use it for draft one, have a native speaker write draft two, and never send unedited output.
What are the best ChatGPT use cases in B2B sales?
Six jobs hold up in daily use: account research summaries, ICP hypothesis generation, first drafts of outreach, objection preparation, call briefs and follow-up structuring. The common thread: the model compresses preparation work while a human keeps judgment and the final word.
How should a sales prompt be structured?
Five parts, every time: role (who the model is), goal (the one deliverable), context (paste real material: website text, profiles, notes), constraints (tone, length, what to avoid) and format (exact output shape). Prompts missing one of the five produce generic filler.
Is using ChatGPT for cold outreach legal in Germany?
The tool is not the issue, the outreach pattern is. German and Austrian law regulate cold contact regardless of who wrote the message, and autonomous volume sending is exactly the pattern they punish. Drafting with AI inside a compliant, low-volume, researched process is fine; automating the whole chain is where teams get burned.