What sales teams use it for
Every job below starts from something real: the prospect’s site, your CRM notes, a call transcript, an approved answer library. Where the input is real, the output is worth sending. Each job links the tools and templates that fit.
Account research and discovery prep
Paste the prospect’s homepage, a recent press release or an earnings call excerpt, plus your CRM notes. The discovery prep template returns a one-page brief: what they do, up to three hypotheses about why they might need you, each tied to something in the sources, six to eight questions in a natural order, what to listen for, and the unknowns to confirm.
It works from what you paste, not the model’s memory of the company, which may be a year out of date or simply wrong. For battlecards, the competitor template compares products from their current pages in the same way.
First touches and follow-ups
A first email earns a reply when it is about them and short enough to read on a phone. The first-touch template builds the email around one real observation from your research, with no invented compliments and no claim about their business that your notes do not support.
The follow-up template adds something new in each touch (a relevant example, a short insight, a direct question) instead of bumping the thread. Score your version in the Grader before the team adopts it, and test it in Prompt Tester on accounts you already know.
First cold email Cold email follow-ups Prompt Grader Prompt Tester
Call recaps and CRM fields
Within an hour of the call, paste the transcript or your notes. The recap template drafts the email to the buyer: what you heard, in their words where possible, what was agreed and the next step with a date.
A second prompt, like the one further down this page, uses the same transcript to fill qualification fields such as MEDDICC or BANT, with a firm instruction to write “not discussed” rather than infer a budget or a decision process nobody mentioned. The rep reads both before anything is sent or saved, because a recap that overstates commitment is worse than no recap.
Quotes, and the business case your champion forwards
Your champion has to sell the deal internally, usually to a finance team you will never meet. The business case template was written for that reader: the ask, the problem today, the options side by side including doing nothing, the payback with every calculation shown, and the risks, so finance can check each number against the material.
Pair it with the quote email template, which states what is included, what is not and what happens next, so the price never arrives without its scope.
Business case for a purchase Quote or estimate email Decision memo
RFPs and security questionnaires
An RFP with well over a hundred questions is mostly questions you have answered before. The RFP template drafts answers only from your approved answer library, cites the source of each statement, rates its confidence, and marks NEEDS INPUT wherever the library has no answer, so a subject matter expert can take those questions.
For a sales engineer’s own search over that library, the cited-answer system prompt (a form of retrieval) answers from the documents or says plainly that they do not cover the question.
Proof: case studies and references
Ask a happy customer for a 30-minute interview and record it, with their permission. The case study template keeps their quotes verbatim, uses only numbers from the interview or their approval notes, and lists the questions to ask if the interview produced no measurable result. Sending the draft back for the customer’s approval is what keeps a reference willing to take the next call.
Pipeline reviews and forecast notes
Before the weekly forecast call, the status update template turns a rep’s scattered notes into an update that leads with what changed since last week and names each blocker, without upgrading “started” to “done” or a hoped-for date into a commitment. When win rate or cycle time moves, the metric change template helps the manager separate what the data shows from the explanation that is most comfortable.
The forecast number itself stays in the CRM, set by the rep and the manager. A model has no idea whether the economic buyer returned the call.
A worked example: one account, first touch to recap
A hypothetical: a sales team selling software to logistics companies wants every SDR’s first email built from real research, and every discovery call recapped the same day. The manager sets it up once, in six steps.
- 1Prepare from their own siteA one-page brief with hypotheses and questions, from pages you paste.
- 2Draft the first touchOne real observation, 60 to 110 words, about them rather than you.
- 3Score the promptThe free Grader estimate shows what the brief leaves to chance.
- 4Test it on 15 past accountsLength, banned phrases and an AI judge for specificity, on every draft.
- 5Compare models blindReps vote on drafts before they see which model wrote which.
- 6Recap the call the same dayWhat you heard, what was agreed and the next step, with nothing overstated.
Testing the first-touch prompt
The manager picks 15 accounts the team already worked, with the research notes each SDR had at the time, and loads them into Prompt Tester as test cases. Three checks run on every draft: a Length check of at most 120 words, subject line included; a Does not contain check for the phrases the team has banned (“I hope this finds you well”, “just checking in”, “circle back”); and an AI judge with one criterion: “The email refers to at least one specific fact from the account notes and makes no claim about the prospect that the notes do not support.”
The judge is the check that matters, and also the one to watch. Read its reasons on every case it fails for the first run or two, because a judge can be swayed by confident wording. When a version passes all 15 cases, it goes to the team.
Choosing the model
In Model Compare, the same account brief goes to three models. With Blind vote on, two reps and the manager pick the draft they would send before the model names appear. If a fast, inexpensive model wins or comes close, that is the one the team uses, and the test suite is rerun whenever someone changes the prompt.
The recap
After the discovery call, the AE pastes the transcript into the recap prompt below. The email goes to the buyer the same afternoon, after the AE has read it, and the qualification fields go into the CRM with a quote from the call beside each one, or “not discussed”.
What it costs
Assume an SDR team of six, each producing about 25 researched drafts a working day: 150 calls a day. Each call sends around 3,000 tokens (account notes, a page of the prospect’s site and the instructions) and gets about 350 back. About a third of each prompt is the standing instructions and examples, read from the provider’s cache. Twenty-two working days a month.
Monthly cost of 3,300 calls with about 3,000 tokens in and 350 out each, 35% of the input read from cache. List prices checked October 20, 2026; your bill may differ.
The monthly bill is modest on every tier. The real cost sits elsewhere: a generic first email to an account you only get one first impression with. If a cheaper model passes your 15 accounts, use it and spend the time saved on research; if it does not, the stronger model is still a small line next to one rep’s time.
A prompt to start from
A call recap that writes the buyer email and the CRM fields from one transcript, and refuses to fill a field the call did not cover.
You are drafting notes for {{rep_name}}, an account executive, after a
discovery call with {{prospect_company}}. The rep checks everything
before it is sent or saved.
Transcript:
"""
{{transcript}}
"""
Part 1. A recap email to {{buyer_name}}, at most 150 words:
- What we heard: their situation in two or three bullets, in their own
words where the transcript has them.
- What we agreed, and the next step with its date if one was set.
- No pricing, discounts, delivery dates or roadmap items unless the rep
said them on the call. If so, use the rep's exact words.
Part 2. CRM fields, one line each: Metrics, Economic buyer, Decision
criteria, Decision process, Identified pain, Champion, Competition.
For each field, quote the line of the transcript it is based on. If the
call did not cover it, write "not discussed". Do not infer.
Guardrails for sales
- Personalize only from what you know. The prompt may use your notes and the pages you pasted, nothing else. A model’s memory of a company is often stale, and a wrong “congratulations on the acquisition” ends the conversation. This is where hallucination costs a deal.
- No commitments the deal desk has not made. Discounts, delivery dates, contract terms and roadmap features appear in an email only if the rep supplied them. Unreleased features are never described as available.
- Follow the email rules. In the US, CAN-SPAM requires commercial email to carry a working opt-out and a valid postal address. The EU and UK set stricter rules for emailing individuals. AI changes none of that; it only makes it easier to send more email that breaks them.
- Get consent to record. Recording and transcribing a call needs the consent your jurisdiction requires, and some US states require everyone on the call to agree. Say you are recording at the start.
- Keep one customer out of another’s email. Details from a current customer’s account go into a prospect’s email only if they come from a published, approved case study.
Key takeaways
- Prepare from the prospect’s own pages and your notes, not the model’s memory of the company.
- Build each first email around one real observation, and test the prompt on accounts your team already worked.
- Recap calls the same day, and let “not discussed” fill any qualification field the call did not cover.
- Answer RFPs only from your approved answer library, with every gap sent to the right expert.
- No prices, terms or roadmap promises in a draft unless the rep supplied them.
Related templates, guides and posts
Questions
Will prospects notice the email was written with AI?
They notice generic email, whoever wrote it. An email built on a specific, accurate observation about their business reads as researched, because it was. The risk is volume: AI makes it easy to send many emails that say nothing, and buyers delete those on sight.
Can AI fill in our CRM from call transcripts?
It can draft the fields, and it saves reps real time. Tell it to quote the line each field is based on and to write “not discussed” for anything the call did not cover, then have the rep confirm before saving. Make sure you have the consent your jurisdiction requires to record and transcribe the call.
Is it safe to use AI for RFP answers?
From approved content, yes, and it can save days on a long questionnaire. Draft only from your answer library, quote the source answer for each draft, and send anything the library does not cover to the right expert. Security, legal and pricing commitments are reviewed by the people who own them before submission.