How to Write a Sales Proposal Email Using AI That Wins Business

Why most proposal emails get ignored

A proposal email is not a pitch. It’s a receipt. Its entire job is to prove you were listening on the call, then make saying yes to the next step trivially easy. Everything else — your credentials, your methodology, the case studies — belongs in the attachment, if it belongs anywhere.

Watch what happens when the first sentence is about the wrong party.

The second version contains no selling at all, and it is the one that gets read to the end. Most sales directors will tell you the same thing in blunter language: nobody has ever bought anything because a vendor described itself well. Reflecting the problem back accurately is what buys you the next thirty seconds.

Length settles the rest. A sales proposal email using AI should land between 150 and 250 words — short enough that a decision-maker finishes it in under a minute, long enough to name their problem, two deliverables, the investment and one ask. Anything technical goes in the attached document, where it can be read properly or ignored safely.

What the Echo-Then-Offer Method is

The Echo-Then-Offer Method is a four-pass sequence that stops the AI from writing anything at all until it has proved it understood the client’s problem. You never ask for the email in one go. You ask for the understanding first, check it, and only then ask for the pitch.

1. Scrub — strip identifying detail out of your call notes before they go anywhere near a chat box.
2. Echo — make the AI restate the client’s problems back to you, with the line from your notes that supports each one.
3. Offer — draft the email around that problem, under a hard word limit.
4. Ground — audit every number and claim in the draft against your notes before you send it.

The whole thing hinges on pass two. A single-prompt request gives the AI a blank canvas and it fills the gaps with plausible corporate filler; the Echo pass forces it to work from evidence you can see. That’s the Echo-Then-Offer Method we teach at PromptPeakAI, and it’s the same discipline behind the Role, Goal, Context, Format structure we use everywhere else.

Order matters more than wording. Most guides to writing a sales proposal email using AI hand you one long master prompt and hope the model holds every instruction at once; splitting it into four short passes means you can see the failure and correct it at the point it happens, instead of staring at a finished draft wondering which of eleven instructions got dropped.

Pass one: scrub the client details

Before anything gets pasted, take thirty seconds and make your notes anonymous. Swap the company name for [Client], the contact for [Contact], and replace confidential figures with brackets. You keep the shape of the problem, which is all the AI needs, and you lose everything that could embarrass you later.

The test is simple: if this text appeared on a screen behind you in an open-plan office, would you mind? Here’s where the lines usually fall.

This matters more on consumer plans than people assume. OpenAI’s own documentation on how your data is used to improve model performance states that content from its consumer services may be used to train its models unless you opt out, while business tiers are excluded by default. Other vendors draw the line in similar places. If you want the fuller picture on what work data is safe to hand over, we’ve covered using AI safely with company information separately.

Pass two: make the AI echo back

This is the pass everyone skips, and it’s the one that does the work. You are not asking for an email yet. You are asking the AI to prove it read your notes, by naming the client’s problems and showing you where in the notes each one came from.

You are helping me prepare a client proposal email. Do NOT write the email yet.

Below are my notes from a discovery call. Using only what is in these notes:

1. List the two business problems this client actually described, in their own phrasing where possible.
2. Under each one, quote the line from my notes that shows it.
3. List anything I would need to confirm with them before proposing a solution.

If a problem is implied but never stated, say so instead of assuming it.

Notes:
[paste your scrubbed notes here]

Read what comes back before you go any further. If the two problems it names aren’t the two you’d have named, your notes were thin — that’s a signal about the call, not about the tool. Fixing it upstream is why designing better discovery questions and researching the prospect before you dial pay off twice over.

A good echo sounds oddly plain. It says things like “they lose most of Monday to manual reporting” rather than “they face operational efficiency challenges.” Plain is what you want — the client’s phrasing is the raw material for the opening line, and abstraction at this stage is the first sign the draft will read like a brochure.

Pass three: draft the offer email

Now you can ask for the email — in the same conversation, so the echoed problems are still in context. The structure below is deliberately rigid, because a sales proposal email using AI goes wrong in exactly the same two places every time: it grows, and it drifts back to talking about you.

Now write the proposal email.

Role: you are me, a [your job title] at [your company], writing to [their role] at [Client].
Goal: get a yes to a 20-minute sign-off call. Not a signature.

Structure, in this order:
1. Their problem, in the words you extracted above. Two sentences maximum.
2. What it is currently costing them, using only figures from my notes.
3. The two things we would do about it. No feature lists.
4. The investment: [amount, or leave this bracket in].
5. One ask, naming a specific day and time.

Constraints:
- Under 200 words in total.
- Plain business English. No buzzwords, no "excited to", no "thrilled to partner".
- Do not invent prices, dates, headcounts or timelines. If a figure is missing, leave the bracket in.
- Plain text only. No headings, no bold, no bullet symbols beyond simple dashes.

Those last four lines are doing most of the anti-robot work. Telling a model what to avoid is consistently more effective than telling it to “sound natural” — a technique worth understanding properly, since negative constraints fix output quality across every writing task, not just this one.

Pass four: ground every number

Language models fill missing parameters with whatever is statistically plausible, which is a polite way of saying they will invent a price if you leave a gap where one should be. That failure mode has a name and a fix, and it’s worth reading up on why AI confidently states things that aren’t true if you’ve never seen it explained.

Audit the email you just wrote against my notes.

For every number, date, name, deliverable and claim in the draft, tell me either:
- the exact line in my notes it came from, or
- NOT IN NOTES

Do not rewrite the email. Just give me the list.

Anything flagged as absent is either something you know and forgot to supply, or something that was never true. Fix the first, delete the second. This takes about sixty seconds and it is the single step that keeps a sales proposal email using AI from committing you to a delivery date nobody agreed to.

Two prompts for other situations

The four passes stay the same; only the Offer prompt changes shape. Two variations cover most of what lands in a working week.

The first is the quick estimate — a prospect asks for a ballpark by email and a ten-page deck would be absurd.

I am a [your role] pitching [type of work] to a prospect who asked for a quick estimate by email.

What they said they need: [one or two sentences, in their words]
What I would deliver: [item 1], [item 2], [item 3]
Timeline: [number] weeks
Investment: [amount]

Write a proposal email under 180 words that:
- opens by restating their need before mentioning anything I do
- lists the three deliverables as short dashed lines
- includes a "What this does not cover" line so scope is clear from day one
- ends by asking whether they want the full scope document

Warm and direct. No buzzwords. Do not add deliverables I did not list.

The second is the cover email attached to a formal proposal — the one that gets forwarded to two people who never met you.

Write a short cover email for a proposal document that will be forwarded internally.

Primary contact: [name and role] — was on the call.
Likely forwarded to: [role 1] and [role 2] — were not.
The problem we are solving: [one sentence]
Operational impact: [what changes day to day]
Financial impact: [only figures I have confirmed]

Requirements:
- Under 200 words.
- Written so someone who never met me grasps the business case in 30 seconds.
- Three short labelled lines: The problem / What we would do / What it costs.
- No company history, no credentials paragraph.
- Leave any figure I have not given you as a bracket.

Two neighbouring jobs deserve their own treatment rather than a tweak to these: chasing a proposal that went quiet needs a proper follow-up sequence, and writing to someone who has never spoken to you is cold outreach, a genuinely different discipline.

Save whichever variation you use most somewhere you can reach in two seconds — a pinned note, a saved project, a text expander. The reason this stops feeling like work by the third proposal isn’t that you get better at prompting; it’s that you stop writing the prompt at all and only fill the brackets.

Which AI tool to write it in

Any of the four major assistants will do this on a free plan. They differ in where the friction sits, not in whether the job is possible.

Drafting inside your inbox removes the copy-paste step, which is where most formatting damage happens. Google’s help page on drafting emails with Gemini in Gmail notes one catch worth knowing before you rely on it: on a personal Google Account the feature is currently restricted to the US, while eligible Workspace and Google AI plans get it more widely. We’ve written up the Gmail workflow in detail if that’s where you live.

Don’t pay for anything on account of this task. A sales proposal email using AI is a few hundred words built from notes you already have, which sits comfortably inside every free tier. The paid plans earn their money when you start uploading fifty-page transcripts or hitting daily message caps mid-afternoon — neither of which happens writing one email.

Fixing a sales proposal email using AI

Five failures account for nearly everything that goes wrong here, and each one traces back to a missing instruction rather than a bad model.

That last row about formatting deserves a note. Chat interfaces produce markdown, email clients render rich text, and the collision leaves you with stray asterisks and hash symbols. In Gmail, the reliable route is the remove-formatting button in the compose toolbar, or switching the compose window to plain text before you paste; Outlook offers the same thing through its paste options. Most browsers also support a paste-without-formatting shortcut, though it varies by app — the toolbar button always works.

One diagnostic habit is worth building. When a sales proposal email using AI comes out wrong, resist rewriting it by hand — identify which of the four passes let the error through and rerun that one. Brochure tone is an Echo failure. An invented figure is a Ground failure. Fixing the pass fixes it permanently; fixing the draft fixes it once.

Run this check before you send

Four questions, about fifteen seconds. If any answer is no, the fix is upstream in the passes, not in another rewrite.

Read it aloud once, too. Anything you’d never say out loud to this person’s face is a line the model wrote for itself, not for them.

What AI genuinely cannot do here

Every honest guide to this owes you the boundary, so here it is. The drafting is the easy part of a proposal, and it’s the only part that’s been handed over.

There’s also the question everyone asks quietly: will the client know? Not if you did the Echo pass. What gives AI-written sales email away isn’t vocabulary, it’s the absence of anything specific — no dates, no figures from the call, no reference to the thing they were annoyed about. An email built from their own words fails that test in the right direction.

Run the four passes on your next real proposal rather than a practice one. You’ll have the Echo and Offer prompts saved after a single use, and from then on the whole thing is about the length of a coffee break.