Verified against Gemini in Gmail · 2026-08-03
Draft an email reply grounded in the actual Gmail thread, not a guess
A prompt that uses Gemini's @Gmail mention to pull the real thread into context before drafting a reply, so the draft reflects what was actually said instead of a generic template that could apply to any email.
The prompt
Ready to copy — highlighted parts are example details you can swap.
@Gmail the thread from Daniela Ruiz titled "contract renewal terms" Read the full thread above before drafting anything — don't draft from the subject line and the last message alone if there's earlier context in the thread that changes what an appropriate reply actually looks like. CONTEXT Who I'm replying as: the account manager on this client relationship What this reply needs to accomplish: confirm we can meet the price they asked for, but not the shortened payment terms Tone: warm but firm — this is a valued client, but we are not moving on payment terms INSTRUCTIONS 1. Summarize in one line what the other person is actually asking for or waiting on — if that's genuinely unclear from the thread, say so instead of guessing at a plausible interpretation and drafting around it. 2. Reference at least one specific thing they said in the thread. A reply that could apply to any email means you drafted from the reply_goal alone and ignored the actual grounded thread content. 3. Draft the reply. Mention the updated contract will go out by Friday. 4. Don't commit to anything not explicitly authorized here: 30-day payment terms, or any discount beyond what is already on the table — if the natural-sounding reply would require conceding one of these, flag that tension in your note to me rather than quietly softening the draft to avoid the conflict. 5. Keep it to under 120 words. Give me just the draft, plus a one-line note on anything you weren't sure about — I'll review before sending. Do not send anything yourself. IF THE THREAD HAS MULTIPLE OPEN THREADS OF DISCUSSION If the email thread has drifted to cover more than one topic (the original renewal question, plus a separate side conversation about a support ticket that got mixed into the same thread), address only confirm we can meet the price they asked for, but not the shortened payment terms in this draft and note that the other topic exists but is out of scope for this specific reply, rather than trying to resolve everything in the thread in one message. TONE MATCHING WITHOUT OVER-MIRRORING Match a professional register appropriate to warm but firm — this is a valued client, but we are not moving on payment terms, but don't mirror the other person's informality or urgency if it would be inappropriate for the account manager on this client relationship to respond in kind — an increasingly terse or frustrated tone in their messages doesn't mean the reply should escalate to match it. IF SOMETHING IN THE THREAD CONTRADICTS ITSELF If an earlier message in the thread states something that a later message from the same person seems to contradict, note the discrepancy to me rather than silently picking whichever version is more convenient for the reply you're drafting. CC'D PARTIES If the thread has other people cc'd who haven't spoken but whose presence might be relevant to how directly you can phrase something, mention that context in your note to me — a reply drafted as if only the sender will read it can land differently once someone else is also on the thread.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The @Gmail mention grounds the draft in the actual retrieved thread text rather than a description of it, which is why the prompt requires quoting or referencing something specific from the thread — a draft that could be sent to any email means the model drafted from the reply_goal alone and ignored the grounded thread content it was actually given access to, defeating the entire reason to invoke the mention in the first place. The do_not_commit_to boundary exists because a model with read access to a real negotiation thread will sometimes draft a reply that sounds reasonable but concedes something — a deadline, a discount, a payment term — that was never actually approved, since a fluent, agreeable-sounding reply is exactly what the model is otherwise optimizing for; naming the exact things it can't offer, and asking it to flag the tension rather than quietly avoid it, keeps the draft inside your actual authority instead of inside what merely reads well. Requiring the model to state plainly when what the other person wants is unclear, rather than guessing and drafting around that guess, matters because a wrong guess about intent produces a reply that's fluent and specific-sounding but answers a question the other person didn't actually ask, which reads as more careless than a generic reply would have. Ending with 'I'll review before sending' and an explicit instruction not to send anything itself is a deliberate human-in-the-loop step: this prompt produces a draft for a real Gmail send action with real consequences attached, which should never go out unreviewed regardless of how confident the draft sounds. Scoping the reply to only the stated goal when a thread has drifted across multiple topics prevents a specific failure where a model, having read the full thread, tries to be comprehensive and resolve every open item it noticed — which can produce a reply that commits to something on the side topic that was never actually part of what this specific message was supposed to accomplish. The instruction not to mirror an increasingly terse or urgent tone protects the actual relationship being managed: matching frustration with frustration, even unconsciously through shorter sentences and a curter register, tends to escalate a tense exchange rather than de-escalate it, which is the opposite of what a reply drafted on someone's behalf in a professional capacity should risk doing.
What you get back
Hi Daniela, Thanks for the note — happy to confirm we can hold the renewal at the $18,400 rate you mentioned. On the 30-day payment terms, I want to flag that we're not able to move off our standard 45-day terms this cycle, so let's keep that as-is for now. I'll get the updated contract over to you by Friday. Let me know if the terms above work and we'll get this locked in. Best, [name]
Verified against
Gemini in Gmail Gemini 3 Pro · 2026-08-03
Changelog
- 2026-08-03 — Initial publish, verified against Gemini in Gmail on Gemini 3 Pro.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
