A training company can turn discovery-call notes into a proposal by loading its template, price list and past outcomes into a Claude or ChatGPT Project, pasting anonymised notes, and asking for scope, an options table, outcomes and terms. Then you check every figure and claim against your own records before the client sees it.
What saves the time is the first draft: once the Project is set up, a draft that might take two hours from a blank page can take twenty minutes, though your own timings will differ. What protects you is the check, because a model will happily write a price, a duration or a promised result that you never agreed. A courier firm's illustrative brief, priced into three options, shows where those slips appear.
Set up a Project that holds your standard proposal material
A Project is a workspace in Claude or ChatGPT where instructions and reference files persist between chats, so you do not repeat yourself. Both products offer them on every plan, with different limits. In ChatGPT, OpenAI's help centre lists project sharing on Free, Go, Plus, Pro, Business, Enterprise and Edu. In Claude, every account can create Projects (a Free account up to five), but sharing a Project with colleagues needs a Team or Enterprise plan. For client material, the plan matters more than the feature: OpenAI says it does not train on ChatGPT Business data by default, and Anthropic says the same of commercial plans such as Claude Team. Check each vendor's current plan page before you rely on any limit, as these change.
Put four things in the Project. First, your proposal template: the headings you always use and the wording of any standard paragraphs. Second, your price list: day rates, delegate rates, programme prices, travel and materials policy, with the date it applies from. Third, three to five short descriptions of past engagements, written in general terms with no client names, including the real outcomes measured, if you have them. Fourth, your standing terms, such as payment, cancellation and intellectual property, or a link to them.
Then write the instruction that governs every chat in that Project. It is the most important paragraph you will write.
You help a training company write client proposals. Use ONLY the price list, past engagement summaries and terms in this Project, plus the discovery notes I paste.
Rules: never invent a price, duration, number of delegates, result, statistic or client name. If something is missing write [NEEDED: what is missing]. Use placeholders for client and contact names. Plain English, British spelling, reading age about 12. Keep paragraphs short. When I ask for options, present them in a table. After any draft, list every number and every outcome claim in a checklist with its source (price list, notes, or "not sourced").The last sentence turns the model into its own first auditor. It will not catch everything, but a checklist with "not sourced" next to a figure is exactly where you look first.
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowPrepare discovery notes the AI can use
Rough notes written in a hurry produce vague proposals. Spend ten minutes after each discovery call turning them into a short brief with the same headings every time, and remove names. A structure that works: the client's business in a sentence; the problem in their words; who needs training and how many; what good looks like in three months; constraints (dates, budget range, location type, systems); decision-makers and process; and anything they said they do not want.
An illustrative brief: "[Client] is a courier company with 60 drivers and 12 dispatchers. Problem: customer complaints about delivery updates, and dispatchers handling rebooking differently. Wants: consistent rebooking process and calmer calls. Constraint: drivers can leave the depot only on Tuesday or Thursday mornings; dispatchers anytime. Budget mentioned: 'around the five-figure mark', not exact. Decision: operations director plus finance sign-off. Does not want: e-learning or anything long." The last line is as important as the first, because proposals fail when they sell what the client said they do not want.
If you recorded the call, only use a transcript if the client knew it was being recorded and agreed. A transcript is raw material with names and offhand comments in it. Convert it into a brief by hand, or with the model after you have stripped names, and then discard the transcript according to your own retention rule.
The prompt that drafts scope and outcomes
With the Project set up and the brief ready, the drafting prompt is short, because the instructions already live in the Project.
Draft a training proposal from this brief. Sections: 1) our understanding of the problem in the client's words; 2) objectives, each a measurable outcome; 3) three options (half-day, full day, programme) in a table with audience, content, delivery format, group size, what is included, and price from the price list; 4) what we need from the client; 5) how we will measure results; 6) timeline; 7) terms.
Mark any missing information [NEEDED]. Recommend one option and say why in two sentences, based only on the brief.
[paste brief]Expect a good structure and some overreach. The commonest overreach is in objectives: models write "reduce complaints by 40 per cent" because it sounds persuasive. You never agreed that. Replace it with an outcome you can stand behind, tied to something you will measure together, for example "dispatchers use one agreed rebooking script on 90 per cent of observed calls by the follow-up visit".
Outcomes: write what you can measure and influence
Outcomes are the part of a training proposal that clients remember, and that you are held to. Good ones are specific, observable and within the training's reach. "Improved communication" fails the first two tests. "Dispatchers can explain a delay and offer a new slot within one call, observed in a sample of ten calls at four weeks" passes.
Use three levels so you promise honestly. Learning outcomes are what delegates can do at the end of the session, which you control. Behaviour outcomes are what they do back at work, which you influence and can check with a follow-up. Business outcomes, such as complaint rates, depend on much more than training, so word them as indicators you will review together, not guarantees. A short table in the proposal makes this clear and protects you.
| Level | Example for the courier firm | How measured | Your commitment |
|---|---|---|---|
| Learning | Each dispatcher can run the five-step rebooking call | Role-play check at the end of the session | We deliver and assess |
| Behaviour | The five steps used on observed live calls | Listening to ten calls at four weeks | We design the check; client provides calls |
| Business | Fewer repeat contacts about the same delivery | Client's own contact data, reviewed together | We review with you; no guarantee |
A worked example: three options with prices checked
An illustration for a training company with this price list (all invented): half-day workshop, up to 12 delegates, $1,400; full day, up to 12 delegates, $2,400; four-session programme of half-days with a follow-up call, up to 12 delegates, $5,200; materials $18 per delegate; trainer travel at cost. The courier client needs 12 dispatchers trained and a first group of 12 drivers.
| Option A: Half-day | Option B: Full day (recommended) | Option C: Programme | |
|---|---|---|---|
| Audience | 12 dispatchers | 12 dispatchers | 12 dispatchers, then 12 drivers |
| Content | Rebooking steps and call script | A plus handling upset callers, practice with real scenarios | B plus four weeks of practice tasks and a call review |
| Delivery | On site, 09:00 to 12:30 | On site, 09:00 to 16:30 | Four half-days over six weeks, plus follow-up call |
| Fee | $1,400 | $2,400 | $5,200 (dispatchers); drivers priced separately |
| Materials (12 x $18) | $216 | $216 | $216 |
| Total before travel | $1,616 | $2,616 | $5,416 |
Check each sum yourself, because this is the bit models get wrong. $1,400 plus $216 is $1,616. $2,400 plus $216 is $2,616. $5,200 plus $216 is $5,416. If the draft shows a different total, or applies a discount you never offered, correct it before anything else. Also check that "recommended" is reasoned from the brief: the client said no long programmes, and dispatchers can attend anytime, so the full day fits; drivers, who can only leave the depot on two mornings, suggest a separate half-day on a Tuesday, a point worth adding to option B as an extension.
Price that extension the same way, so the client sees the full picture rather than a teaser. A driver half-day on a Tuesday morning is $1,400 plus 12 delegates at $18 for materials, $1,616, so option B with the driver session comes to $2,616 plus $1,616, or $4,232 before travel. That is less than option C's $5,416 for the dispatchers alone, and it respects the "nothing long" line in the brief. Put that comparison in one sentence under the table; it argues for your recommendation better than any adjective the model might offer.
This sort of cross-check, constraints against options, is where you add the judgement a model cannot. The brief's "does not want long e-learning" line should shape the whole table, and your edit should show it did.
What the first draft looks like, and what you fix
To see the editing in practice, here is an illustrative extract from a first draft for the courier brief, followed by the changes a careful owner would make.
Our understanding
[Client] has told us that customers are frustrated by delivery updates and that dispatchers rebook missed deliveries in different ways. You would like a consistent process and calmer calls, delivered in a way that suits drivers' limited availability. You have said you do not want e-learning or anything long.
Objective 1: Reduce customer complaints by 40% within three months.
Objective 2: Ensure all dispatchers are fully confident in handling difficult calls.
Recommended: Option B. Our award-winning workshops have transformed customer service teams.The first paragraph is good, because it plays the client's own words back and includes the "does not want" line. The rest needs work. Objective 1 invents a figure; replace it with an indicator you will review together. Objective 2 uses "fully confident", which cannot be observed; replace it with a behaviour you can check on a sample of calls. The last sentence is an unsourced claim, "award-winning", that appears nowhere in your Project; delete it. After the edit, the section is shorter, plainer and safe to sign.
Edits like these are why the "list every claim with its source" instruction earns its place. In this extract it would have marked "40%" and "award-winning" as "not sourced", and you would have gone straight to them.
Outcome wording: a before and after
| Draft wording | Problem | Better wording |
|---|---|---|
| Improve team communication | Cannot be observed | Each dispatcher gives the delay reason, the new slot and a contact route in one call |
| Boost customer satisfaction | Depends on much more than training | We review repeat-contact data with you at six weeks |
| Delegates will master the system | "Master" is unmeasurable | Delegates complete five set tasks in the system without help |
When the client asks for a discount
A proposal often comes back with "can you do it for less?". The worst answer is to shave the price and leave the scope unchanged, because you have then agreed to deliver the same work for less. Reprice the scope instead. Ask the AI to produce a version of the options table with one lever changed at a time, such as a smaller group, one fewer session, client-supplied venue, or a train-the-trainer element where your own staff run follow-up sessions.
A quick sum shows the effect. Say option B costs $2,400 for 12 delegates and the client has $2,000. Offering the same day for 10 delegates at the per-head rate gives $2,000, but your trainer's time is the same, so your margin falls. A better response may be option A at $1,400 plus a one-hour follow-up call at $200, total $1,600, which fits the budget and keeps the follow-up that makes behaviour change likelier. The AI is good at generating such alternatives quickly; deciding which ones you can afford is yours.
The cover note and the follow-up
A proposal attached to an empty email is easy to ignore. Ask the AI for a three-sentence cover note that names the problem from the call, says which option you recommend and why, and proposes a date for a short call to answer questions. Use a placeholder for the name. Send the follow-up three working days later, from a template that asks one question rather than "just checking in": "Does the recommended option fit what you need, or would a different scope help?"
Keep a record of the questions clients ask after proposals. Add the common ones, such as "can you do this online?" or "do you provide certificates?", to the Project as short standard answers. Over a few months, the proposal's first draft will answer them before they are asked, and your win rate is likelier to move for that reason than for any change of wording.
Pricing wording that avoids surprises
State what the fee includes, what it excludes and what is estimated. Say whether prices are per group or per delegate, the maximum group size, what happens if more attend, and when travel and expenses are charged. State the currency and that sales tax is added where applicable, without naming any regime. Say how long the proposal's prices are valid, for example 30 days.
Ask the AI to read the pricing section as a sceptical finance manager and list every way a number could be misread. A typical output: "does 'up to 12' mean a charge for 13?", "are materials per delegate or per course?", "is travel capped?". Fix each answer in the text. A proposal that leaves finance with questions is a proposal that stalls.
Terms: link rather than paste, and keep them consistent
The terms section should be short and point to your full terms. Cover payment timing, cancellation and rescheduling, intellectual property in materials, confidentiality, and what the client must provide, such as room, equipment and delegate list. Do not ask the AI to write new legal terms, because it will invent plausible clauses. Paste your own terms into the Project and ask it to summarise them in five lines for the proposal, then compare the summary with the original.
Cancellation is the clause clients ask about most. The tutorial on training course cancellation terms drafted with AI covers tiers, substitutes and postponement, and your proposal should repeat the tiers exactly as your terms state them. If you also run needs analysis for clients, link the finding to the proposal: a proposal built on a report, as in the needs analysis tutorial, can cite the gap table instead of a general promise.
The figure check before it leaves your desk
Every proposal gets the same checks, in the same order, however confident the draft sounds. The model's own "source" checklist speeds this up but does not replace it.
- Every price against the price list, and every total recalculated by hand or in a spreadsheet.
- Every number of delegates, days, sessions and hours against the brief.
- Every outcome claim: is it measurable, and are you comfortable being held to it?
- Every named tool, standard or accreditation: do you actually offer it?
- Dates: the proposal date, the validity period and any course dates.
- Names: the client's name, contact and company spelled correctly everywhere, since placeholders are added back by hand.
- Consistency: the options table, summary and terms say the same things.
Run a second model if you can. Paste the final proposal and the price list into a fresh chat with another assistant and ask it to find inconsistencies. A second model will not catch everything, but it often trips over different lines from the first, and any disagreement points you to a figure worth recalculating by hand.
Mistakes that show up after sending
The first realistic mistake is the invented credential. A draft says "our programmes are accredited" in a sentence you skimmed, when only one course is. A client asks for the certificate and you have to explain. The second is the recycled client: a draft reuses a detail from a previous brief that the model still held in the same chat. Start a fresh chat for every proposal, even inside a Project.
The third is tone mismatch. Models write warm, assured, slightly salesy prose that does not sound like you. Ask for a rewrite in your own voice by pasting two paragraphs of your own writing as a sample, then read the proposal aloud. The fourth is a missing "no": the proposal says what you will do and forgets what you will not, such as developing new content or travelling more than once. Add a short "not included" list, which prevents scope creep in the first week of delivery. For the courier proposal it might read: no new e-learning content; one trainer visit per session booked; no recording of sessions; driver training only where the driver extension is accepted; venue, room and equipment supplied by the client.
Measuring whether the process works
Track three numbers for the next ten proposals: time from discovery call to sending, the number of edits you make after the AI draft, and win rate. If time drops from two hours to forty minutes, and you still make about a dozen edits, the process works. If you make dozens of edits, improve the Project instructions and brief template. If win rate falls, the problem is probably in the options or the price, not the writing.
After a client accepts, save the proposal as a past-engagement summary, with names removed, so the next one starts from a better library. Over time, your Project holds your real outcomes and your real wording, and the drafts get closer to what you would write yourself. For a broader look at which other training admin jobs suit the same approach, see AI for corporate training companies, from proposals to certificates.
Questions about writing training proposals with AI
Should I use Claude Projects or ChatGPT Projects for proposals?
Either works. Pick the one your team already uses and trusts with client material. Both let you store your proposal template, price list and past outcomes, so each new proposal starts from the same instructions. Check the plan's data terms before uploading client notes.
Can I paste raw discovery-call notes into the AI?
Strip names, sensitive details and anything the client asked you to keep confidential, and use a plan that does not train on your content, such as ChatGPT Business or Claude Team. Replace the client's name with a placeholder, and add it back yourself in the final document.
How many options should a training proposal offer?
Three works well: a short, focused option, a recommended middle option, and a fuller programme. It gives the client a choice without a long menu. Make the difference between them clear in scope, not just price.
How do I stop the AI inventing results or prices?
Give it your price list and past results as source material, instruct it to use only those, and make it mark gaps as [NEEDED]. Then check every figure in the output against your own documents before the proposal leaves your hands.
Should the proposal mention that I used AI?
Not as a selling point. Clients buy outcomes and your judgement. If asked, say you use AI for first drafts and that you check the content and numbers yourself. Follow any client policy on AI and confidentiality.
Further reads
- How Consultants Use AI to Write Proposals in Under an Hour — Proposal drafting method from consulting, with timings.
- How to Evaluate an AI Implementation Proposal or Quote — How buyers judge a proposal, from the client side.
- Joining Instructions Template: Details and Reminders With AI — What you send once the proposal is accepted.
- How to Start a Training Company: Accreditation, Insurance and AI — Accreditation and insurance claims to get right in a proposal.
- How to Write Event Proposals With AI in Half the Time — Proposal writing for events, with similar structure.
- How Much Time and Money Does AI Staff Training Take? — What clients budget for staff training time and cost.
- Can AI Summarise Training Feedback Forms for a Client Report? — Score ratings in a spreadsheet, theme the comments with AI, and produce a one-page client summary that protects delegates' anonymity.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: OpenAI's Projects help article (project sharing on every plan), Anthropic's Projects help articles (Free up to five projects; sharing on Team and Enterprise), OpenAI's business data page and Anthropic's privacy centre, all read 4 November 2026.