What actually moves pipeline (3 concrete uses)
- Pre-meeting research. LinkedIn + prospect site + meeting brief → 3 smart questions that open the conversation. Make it a ritual before every call: you build a reputation as a prepared rep in 4 weeks.
- First drafts of proposals. Structure, language and a first draft that saves 60-70% of the time. You handle the close — ChatGPT handles the first 80%.
- Call summaries with action-item extraction. Paste the transcript, ChatGPT returns: topics covered, objections, next steps. Saves 30-45 minutes per call.
What looks useful and isn't
- Generating fresh cold emails. Without prospect data enrichment, they're templates in disguise. Zero reply rate.
- Repeated "improve this email" loops. You burn more time iterating than you would writing it from scratch.
- Replacing human discovery. ChatGPT can suggest questions — it can't read between the lines in a live conversation.
- Closing over chat. Negotiation and closing need live commercial judgment.
How to connect ChatGPT to your CRM without code
No engineering required — two reasonable paths:
- Zapier + ChatGPT plugin. Workflows that fire ChatGPT when something changes in HubSpot/Pipedrive and save the response. Fine for semi-automated tasks (summarize, suggest).
- Make + OpenAI integration. Same idea, more flexible, slightly more technical. Similar cost.
- HubSpot Copilot or [Salesforce](/soluciones/ia-para-salesforce) Einstein. If your CRM has native AI that covers your use case, it's usually the easiest option — but watch the lock-in.
When ChatGPT isn't enough and you need an AI SDR
ChatGPT is the rep's tool; an AI SDR is the system. If you need:
- High-volume prospecting (>200 leads/week).
- Personalization at scale with automated data enrichment.
- Deliverability management (warm-up, rotation, monitoring).
- Bidirectional CRM integration with continuous sync.
- Reporting on pipeline generated vs. emails sent.
Then ChatGPT isn't enough. You need an AI SDR — a different project, with its own architecture and its own budget.
Examples of versioned prompts
Versioning prompts (keeping a shared library of the ones that work) is what separates a professional team from one that improvises every time:
- Prompt v1 (prospect research): "I'm sending you LinkedIn + the prospect's website. Return 3 smart questions to open a first meeting, based on concrete signals you see in that data. No generic questions."
- Prompt v2 (call summary): "I'm sending you the transcript. Return: 1) Topics covered (bullets), 2) Real objections (not the surface-level ones), 3) Next steps with owner and date."
- Prompt v3 (proposal draft): "Customer: [data]. Need: [summary]. Generate a proposal structure with: context, problem, proposed solution, success metrics, next steps. Tone: [formal/friendly depending on customer]."