Gave Dots only a name — it recalled the background, asked 3 follow-ups, signed and emailed the reference letter inside ChatGPT
kevingibbonX (Twitter) · 2026-10-27
Read originalI asked @OpenAI Dots to write a reference letter for someone. All I gave it was their name. It already knew their background, asked 3 follow-ups, then handled signing and emailing after my approval. All inside @ChatGPT. Bye-bye @DocuSign. And most SaaS apps.
Business model
Author @kevingibbon asked ChatGPT Dots to write a reference letter, supplying only the person’s name. Dots already knew the background, asked three clarifying questions, then after approval handled signing and emailing — all inside ChatGPT. Author presents it as a replacement for DocuSign-style tools. Useful for founders and managers who occasionally need formal reference or signed documents without leaving the chat.
Workflow
- Dots presents known background from existing context (how it obtained the background is not stated in the original post)
- Dots asks up to 3 follow-up questions
- Generate letter draft, prepare signature, draft outbound email
- Execute send
Adaptation plan
Results are as reported by the original post author. How Dots obtained the background is not stated in the original post; whether signing was a real digital signature or a typed name is unclear and is not presented as a legally binding signature. Workflow, prompts, and trial I/O are AgentFlow reconstructions and were not live-tested here. No income or time-saved figures; document workflow case only.
Prompt
AgentFlow draft · Untested.
You are my document and formal-comms assistant (ChatGPT Dots style).
- Person’s name: {{person_name}}
- Purpose / recipient: {{purpose_or_recipient}} (e.g. new job, school, visa; “unspecified” if unknown)
- My relationship and time worked together: {{relationship}}
- 2–3 specific strengths or examples to emphasize: {{key_points}} (leave blank if you should propose from known context)
- Tone: {{tone}} (formal / warm-formal / concise)
1. From the name alone, summarize what you already know about the person and list gaps.
2. If critical information is missing, ask at most 3 concrete follow-up questions and wait for my answers before writing the body.
3. Produce a complete reference-letter draft (header, body, closing, signature block).
4. Prepare the signing step and outbound email draft, but wait for my explicit confirmation before any irreversible action.
- Known-background summary
- Clarifying questions (max 3) or “information sufficient”
- Full letter draft
- Outbound email subject + body draft
- Confirmation checklist (signing method, recipient, send now?)
- If background details conflict, list the conflicts and ask which version to use.
- If there are not enough concrete examples to support the claims, say so and ask me to supply them — do not invent.
- Never auto-send email or apply a legally binding signature on my behalf.
- Execute send only after I reply “confirm send” or “sign and send”.
Untested reconstruction based on the original post.Running costs and pitfalls
Results are as reported by the original post author. How Dots obtained the background is not stated in the original post; whether signing was a real digital signature or a typed name is unclear and is not presented as a legally binding signature. Workflow, prompts, and trial I/O are AgentFlow reconstructions and were not live-tested here. No income or time-saved figures; document workflow case only. No independent cost benchmark is available. Check subscriptions, permissions and external actions before running a workflow.