A one-line prompt leaves an AI writing tool to guess the buyer's problem, the proof and the ask. The rep types a name and a company, and the tool fills the rest with generic pain points and proof nobody approved. A short brief replaces those guesses, and five common tools can hold it with the right permissions.
This applies to B2B outbound where reps draft first-touch or follow-up emails with an AI tool and send from their own mailbox. A brief narrows what the AI writes but does not replace a rep's review, and that review covers legal rules such as CAN-SPAM.
What Is An AI Sales Email Brief
An AI sales email brief is a one-page input that tells an AI writing tool who the buyer is, which problem the email addresses, which proof it may cite and what single next step to ask for, so the draft stays inside facts the team has approved.
A one-line prompt such as "write a cold email to a VP of Sales at a logistics company" gives the tool a role and an industry and nothing else. The tool then guesses the problem, guesses the proof and closes with a generic offer to chat.
A brief replaces those guesses with decisions the team has already made. The rep still adds what they found about the prospect. The brief decides what the email argues, what it may claim and what it asks for.
The brief is the core input of any sales personalization workflow that uses AI drafting. It is also the part most teams skip, because a prompt feels faster.
It matters more now because buyers check AI output with people. Gartner surveyed 645 B2B buyers in 2025 and found 69% prefer to validate AI-generated insights with sales reps. Those buyers used an average of seven information sources in a recent purchase, and 45% used generative AI, mainly to gather vendor and product information.
The rep is where buyers go to check what AI told them. A rep's email that carries an invented claim spends the trust the buyer was about to extend.
What Should An AI Sales Email Brief Include
A working AI sales email brief has seven fields, namely the audience, the reason to write now, the buyer problem, the approved proof, the clear ask, the tone and length limits, and the claims the AI must never make. Each field closes a gap the AI would otherwise fill with a guess.
The seven fields of an AI sales email brief
| Field | What to write | Weak entry | Strong entry |
|---|---|---|---|
| Audience | One segment, one job title | Sales leaders | First-year VPs of Sales at freight brokers |
| Reason now | The trigger for writing | Saw your company online | Opened a second warehouse |
| Buyer problem | The problem in the buyer's words | Wants more growth | Forecasts built from spreadsheets reps skip |
| Approved proof | Two or three library items | Our customers love us | A signed-off case study line, quoted exactly |
| Clear ask | One next step, word for word | Let me know your thoughts | Should I send the forecast checklist? |
| Tone and length | Register and a word cap | Professional | Plain, under 120 words |
| Banned claims | Claims and words never used | None | No ROI figures, no unlisted customer names |
Why the AI needs each field:
- Audience: Without one role, the AI writes for everyone and persuades no one.
- Reason now: A trigger gives the first line a real reason to exist.
- Buyer problem: This is the argument of the email. Without it, the AI picks generic pain points.
- Approved proof: The AI may only support the argument with items listed here.
- Clear ask: A fixed ask stops the AI ending on a vague offer to chat.
- Tone and length: Limits keep drafts short enough to read on a phone and consistent across reps.
- Banned claims: An explicit list of what never to say catches the claims the AI is most likely to invent.
The banned claims field is the one prompt templates usually leave out. The same Gartner survey found 51% of B2B buyers say they are more likely to encounter misleading information from generative AI, against 49% who say a sales rep. Buyers are split almost evenly on which source misleads more, so a rep's email has no room for an invented figure.
Approved proof also has a legal side. The FTC expects an advertiser to have a reasonable basis for any objective claim it makes, which is why the proof field points to a library with an owner rather than to whatever the AI can find.
How To Define The Buyer Problem
The buyer problem in a sales email brief is one sentence, in the buyer's own words, that names their role, the situation they are in and what it is costing them, because an AI tool told only the prospect's name and company will fill the gap with generic pain points.
Start from the segment you already target. If your ICP still lives on a slide, first turn your ICP into prospecting filters so the brief covers one segment a list can actually pull.
Then build the sentence from four parts:
- Role. One job title in one segment.
- Situation. What is true for them right now, often the trigger.
- Cost. What the problem costs in time, money or risk.
- Their words. How the buyer would describe it on a call.
A usable shape is "{{buyer-role}} who {{situation}}, which means {{cost}}". For example, "First-year VPs of Sales at freight brokers who inherited a forecast built in spreadsheets, which means every board meeting starts with numbers nobody trusts."
The buyer's words come from discovery call notes, lost-deal reasons, support tickets and what customers write in reviews. Copy their phrasing rather than your positioning language.
Surface personalization is not a buyer problem. Lavender's Personalization Assistant, for example, shows a recipient's recent news, LinkedIn and X posts, podcasts and work history in the inbox. That context helps a rep choose an opening line. It does not tell the AI why the buyer should care, so use a recent post to support the problem, never to replace it.
- Names one role in one segment
- Describes a situation the buyer would recognise
- States what the problem costs in time, money or risk
- Uses words a buyer has actually used
- Still makes sense without the prospect's name or a recent post
- Fits in one sentence
How To Build An Approved Proof Library
Approved proof is the closed list of customer results, figures and quotes a team has signed off for outbound use, and under the FTC's advertising substantiation policy a claim such as "studies show" must have at least the support it states, so every item needs a source behind it.
Proof includes case study results, customer quotes used with permission, product facts and any third-party figure the team wants reps to repeat. Each item in the library carries the same five fields:
- Claim: The exact wording reps and the AI may use.
- Source: The case study, customer approval or report behind it.
- Owner: The person who signed it off, usually in marketing, legal or customer success.
- Permitted use: First touch or follow-up, named customer or anonymised.
- Expiry: The date it must be re-approved or retired.
The policy behind this is old and plain. Objective claims imply the advertiser has a reasonable basis for them. A claim that names its support, such as "tests prove", needs at least that level of support. An AI tool has no way to know which of your numbers meet that bar unless the brief tells it.
The library is closed on purpose. AI tools fill gaps. Ask for a customer result when the brief holds none, and the draft may contain a plausible figure or customer name that nobody can source. Buyers already expect misleading information from generative AI, as the Gartner figures above show, so that one figure costs more than the email gains.
Store the library where the AI can read it but reps cannot edit it. The tool sections below show how each tool sets that up. In practice, a Notion database holds the master list and the AI tool holds a read-only copy of the items each brief uses.
How To Write The Clear Ask
The clear ask in an AI sales email brief is the one next step the email requests, written out word for word, so the AI does not close with a vague offer to chat or stack several questions in the last line. One email gets one ask, matched to how warm the buyer is.
For a cold first touch, ask for something a buyer can answer in one word, such as a yes to a resource. After a reply or a meeting, ask for a short call with a named day. Put the exact sentence in the brief and tell the AI to end every email with it.
Weak and strong asks
| Weak ask | Strong ask |
|---|---|
| Let me know your thoughts | Should I send the two-page forecast checklist? |
| Open to a quick chat sometime, or should I send more info? | Is fixing the forecast on your list this quarter? |
| Happy to connect whenever works | Worth 20 minutes on Tuesday to walk through it? |
A single specific ask also helps the rep. Buyers go to reps to check what they have read elsewhere, so a reply to a clear question starts the conversation the rep wants.
How To Write An AI Sales Email Brief Step By Step
To write an AI sales email brief, choose one segment and trigger, state the buyer problem in one sentence, attach two or three items of approved proof, write the clear ask word for word, set length and banned claims, and save it as standing instructions in the AI tool.
- Pick one segment and one trigger. A brief that covers two segments writes a vague email for both. Start with the segment that has the clearest trigger.
- Write the buyer problem. Use the four-part sentence from the buyer problem section and check it against the checklist.
- Pull two or three proof items. Take them from the approved proof library with their exact wording. Three is plenty for one short email.
- Write the clear ask. One sentence, copied into the brief exactly as the email should end.
- Set tone, length and banned claims. Give a word cap, a register and a short list of claims and words the AI must never use.
- Load the brief as standing instructions. In Claude, a project holds the brief as project instructions and the proof as project knowledge. In ChatGPT, project instructions apply only inside the project and override each user's own custom instructions. In Copy.ai, Infobase holds the brief for the workspace.
- Generate, review and send a small batch. Draft a handful of emails, review each against the brief, send to a small part of the segment and record the brief version on every send.
Copy the template below into your tool. Every slot uses double curly braces, so replace each one before you save.
AI SALES EMAIL BRIEF
Version {{brief-version}} | Owner {{brief-owner}}
AUDIENCE
{{buyer-role}} at {{segment}}
REASON TO WRITE NOW
{{trigger}}
BUYER PROBLEM
{{buyer-problem}}
APPROVED PROOF (use only these, word for word)
1. {{approved-proof-1}}
2. {{approved-proof-2}}
3. {{approved-proof-3}}
CLEAR ASK (end every email with exactly this sentence)
{{clear-ask}}
TONE AND LENGTH
{{tone}}, under {{word-limit}} words, short paragraphs
BANNED CLAIMS AND WORDS
{{banned-claims}}
RULES FOR THE AI
- Use only the proof listed above. If a point needs proof that is not listed, leave the point out.
- Never invent a customer name, figure, result or quote.
- Mention the prospect's posts or news only when they support the buyer problem.
- Make one ask only, the clear ask above.
- Write a subject line that describes what the email is about.The rules block matters as much as the fields. It tells the AI what to do when the brief runs out, which is to leave the point out rather than guess.
Where The Brief Fits In A Sales Personalization Workflow
In a sales personalization workflow the brief sits between research and drafting, so the facts the rep gathered and the proof marketing approved reach the AI tool in one place before any email is written. Every later step, from review to measurement, is checked against that same brief.
The brief in a sales personalization workflow
| Stage | Owner | Tool | Output |
|---|---|---|---|
| Research and trigger | Rep | CRM, personalization assistant | Trigger and context notes |
| Proof library | Marketing or legal | Notion database | Signed-off proof items |
| Brief | RevOps or marketing | Project instructions or Infobase | Versioned brief |
| Draft | Rep | Claude, ChatGPT or Copy.ai | First draft |
| Review | Rep | Review checklist, Lavender | Approved email |
| Send | Rep | Own mailbox or sequencer | Sent email |
| Measure | RevOps | CRM reports | Results by brief version |
Ownership is the point of the table. Marketing or legal owns the proof. RevOps or marketing owns the brief and its versions. The rep owns the buyer problem line for their accounts, the review and the send.
The rep stays in the loop because buyers come to the rep to check what AI told them. A rep who sends a draft they have not read gives up the one role buyers still trust them with.
Responsibility does not move with the work either. The FTC's CAN-SPAM guide says a business that hires another company to handle its email marketing cannot contract away its legal responsibility to comply. The team that sends the email answers for it, whoever or whatever drafted it.
If you are still choosing a drafting tool, compare the AI sales email writing tools first. The brief works with any of them, because it is the input rather than the tool.
Which Tools Run Each Step And What Permissions They Need
Claude and ChatGPT hold an AI sales email brief as project instructions, Copy.ai shares it across a workspace through Infobase, Notion stores the approved proof library with page-level permissions, and Lavender checks the finished draft inside the inbox. Each tool needs its edit rights set before reps start.
Where the brief lives in each tool
| Tool | Workflow step | Where the brief lives | Who can change it |
|---|---|---|---|
| Claude | Draft | Project instructions and knowledge | Can edit members |
| ChatGPT | Draft | Project instructions and files | Owner and Edit members |
| Copy.ai | Team drafting | Infobase and Workflows | Set by Teamspace role |
| Notion | Proof library | Database, one row per item | Full access holders |
| Lavender | Review | Rep's copy of the brief | Each rep |
A shared brief that every rep can edit drifts. One rep softens the ask, another adds a result from memory, and within weeks the approved wording is gone. Give edit rights to the brief owner and the proof owner only, and read-only access to everyone else.
Claude Projects For The Brief And Proof Library
Claude holds an AI sales email brief as project instructions and the approved proof as project knowledge, and on Team and Enterprise plans the project can be shared so reps get Can view access while only the proof owner has Can edit.
Role in the workflow: A drafting assistant that reads the brief at the start of every chat in the project.
Where the brief lives: Project instructions tailor Claude's responses inside that project, and the knowledge base holds the uploaded proof documents. Projects are available on every Claude plan.
Sharing and permissions: Can view shows the contents, knowledge and instructions. Can edit changes instructions and knowledge and adds or removes members. Admins can turn project sharing off, and existing shares stay in place.
Data use: Anthropic says that for its commercial products, Claude for Work and the API, it will not use chats to train models unless the customer joins its Development Partner Program or explicitly reports materials such as thumbs up or down feedback. Consumer plans use a separate setting, so company briefs belong on a work plan.
Key limitation: Project sharing works only on Team and Enterprise, so reps on individual plans each keep their own copy of the brief.
Real cost driver: Enhanced project knowledge with RAG, which expands capacity by up to 10 times, comes only with the paid Pro, Max, Team and Enterprise plans. A large proof library needs one of them.
Best for
- Teams on Claude Team or Enterprise that want one read-only brief per segment
Not for
- Reps on individual plans who need to share one brief
ChatGPT Projects For Shared Brief Instructions
ChatGPT projects keep an AI sales email brief as project instructions that override each user's own custom instructions, and giving reps Chat access rather than Edit access stops them changing the brief or the proof files.
Role in the workflow: A drafting assistant that starts every chat in the project from the same instructions and files.
Where the brief lives: Project instructions apply only inside the project, and project files hold the proof. Project-only memory draws context only from chats in that project, so one segment's brief does not leak into another.
Sharing and permissions: The owner updates instructions, adds files and invites or removes members. Edit access lets members update instructions, upload or remove files and invite others. Chat access lets members use the chats, files and instructions but not invite anyone.
Key limitation: OpenAI states that connectors are not supported in projects, custom GPTs cannot be used inside them, and chats in shared projects can be branched but not worked on together at the same time.
Real cost driver: The file cap per project rises with the plan, from 25 files on Go and Plus to 40 on Edu, Pro, Business and Enterprise, with 10 uploaded at a time. That cap bounds how large a proof library one project can hold.
Best for
- Teams that want a shared brief reps can use but not change
Not for
- Workflows that need connectors or custom GPTs inside the project
Copy.ai Infobase And Workflows For Team Briefs
Copy.ai, now part of Fullcast, stores an AI sales email brief and its approved proof in Infobase, which can be shared across a workspace, and runs it through Workflows that connect to Salesforce and HubSpot.
Role in the workflow: A team-wide home for the brief and a repeatable workflow that applies it to many records.
Where the brief lives: Infobase is a persistent knowledge store behind Copy.ai Chat and holds long-form guidelines. Brand voices are unlimited and chosen per chat, so each segment's tone can sit beside its brief.
Sharing and permissions: Teamspace roles are Admin, Editor and Collaborator. A personal space is visible only to the user and workspace admins, while Team Space is visible to colleagues, so the shared brief belongs in Team Space.
Data use: Copy.ai says Infobase and Chat data is siloed within the workspace and not used to train any LLM.
Key limitation: Workflow output still needs the review step. Because Workflows sync with the CRM, a weak brief scales across every record a workflow touches before anyone reads a draft.
Real cost driver: The entry Chat plan includes unlimited words in Chat but no workflow credits. The Growth, Expansion and Scale plans carry 20K, 45K and 75K monthly workflow credits, so running briefs through Workflows at volume depends on those credits.
Best for
- Teams that want one brief applied to CRM records through Workflows
Not for
- Small teams that only need a shared brief for chat drafting
Notion As The Approved Proof Database
Notion works as the approved proof database for AI sales emails because Can view access lets reps read every proof item while only the owner with Full access can change the wording, and Notion AI uses only pages the user is allowed to see.
Role in the workflow: The system of record for proof, with one database row per item and properties for owner, wording, source and expiry.
Where the brief lives: The brief can sit as a page beside the database. Notion AI can revise content to match a referenced style guide page and can search connected Slack channels and Google Workspace documents.
Sharing and permissions: Full access edits and shares, Can edit edits without sharing, Can edit content lets contributors add and edit database pages without changing structure or views, and Can comment and Can view are read-only levels. Page-level access by a person property comes with Business and Enterprise. Notion applies the broadest access a user is given, so check that no workspace-wide setting gives reps edit rights.
Data use: Notion AI follows workspace access rules and only uses information it has permission to view.
Key limitation: Notion AI comes with the Business and Enterprise plans, and lower plans get only a limited number of trial responses.
Real cost driver: High-credit models use Notion credits and stay off until an owner or admin turns them on.
Best for
- Teams that need one owned, read-only list of approved proof
Not for
- Teams that want the proof library and the drafting in one tool
Lavender For Checking The Draft In The Inbox
Lavender checks an AI-drafted sales email inside Gmail, Outlook, Outreach, Salesloft or HubSpot before it is sent, scoring the copy and showing recipient context, but it does not know which proof your team has approved.
Role in the workflow: Email Coach scores an email or template to find and fix what hurts reply rate. It also works with Apollo, Groove and Gong.
Where the brief lives: Lavender does not hold the brief. The rep keeps the brief open and checks the scored draft against it.
Data use: Lavender says its generative features use OpenAI models, self-hosted fine-tuned models and models trained on custom data. Security-conscious organizations may need an IT review before it can authenticate with their email provider.
Key limitation: The score measures how the copy reads, not whether its proof is approved. A well-scored draft with an invented figure still fails review.
Real cost driver: The Team Plan is priced per seat and adds the Coaching Dashboard and team analytics. The Basic tier is limited to 5 analyzed and 5 personalized emails a month.
Best for
- Reps who want a readability check inside the inbox before sending
Not for
- Teams looking for a place to store or enforce the brief
How To Review AI Sales Emails Before They Send
Before an AI-drafted sales email goes out, check every claim against the approved proof and the message against the CAN-SPAM rules, because the FTC's CAN-SPAM guide makes no exception for business-to-business email. The rep who sends the email owns both checks, and they take a minute when the brief is tight.
Checking AI output is a habit worth building. A Gartner survey of 846 US consumers in 2025 found 54% had to double-check the accuracy of all information generative AI tools gave them. That was a consumer study, not a B2B one, but it is a fair signal for any draft a rep is about to sign.
Run this checklist on every draft:
- Every claim, figure and customer name appears in the approved proof library
- No result, quote or statistic appears that the brief did not supply
- The subject line accurately reflects the content of the email
- The header and From information are accurate and identify the sender
- The message discloses that it is an advertisement
- A valid physical postal address is included
- A clear opt-out is included, and opt-outs are honoured within 10 business days
- The ask matches the clear ask in the brief
- Any prospect detail mentioned is accurate and supports the buyer problem
The stakes are concrete. The FTC guide puts the penalty at up to $53,088 for each separate email in violation, a figure it updated for inflation in 2024.
Handing the writing to someone else does not hand over the duty. The same guide says a business that hires another company to handle its email marketing cannot contract away its legal responsibility to comply. Whoever drafted the email, the sending business answers for it.
Deliverability is a separate job from review. To set up your first cold email campaign, including SPF, DKIM and DMARC, work through the campaign setup playbook before the first batch goes out.
This section is general information, not legal advice.
How To Measure Whether The Brief Works
An AI sales email brief works when its emails earn positive replies that turn into meetings and when reps send its drafts with few edits and no rejected claims, so track each brief version on those measures rather than on reply rate alone. Each measure points to a different field to fix.
- Positive reply rate: Replies that show interest, not all replies. A drop points at the buyer problem.
- Reply-to-meeting rate: How many positive replies become meetings. A low rate points at the clear ask.
- Drafts sent without edits: A rising share means the brief is doing the rep's work. A falling share means it is drifting.
- Review rejections by reason: Count rejections as unapproved claim, wrong problem, wrong ask or compliance. Each reason maps to one field.
- Unsubscribe rate: A rise after a brief change means the message stopped fitting the segment.
Version every brief and change one field at a time. Send each version to a small batch from the same segment, record the version on every send and compare versions only on the same segment and trigger.
Reply rate on its own can mislead, which is why high cold email reply rates can hide poor pipeline. Meetings are the better test of a brief, because buyers who validate with reps decide in that conversation whether the email told the truth.
Common Mistakes In AI Sales Email Briefs
The first mistake to fix in an AI sales email brief is leaving out approved proof, which pushes the AI to fill the gap with results, figures or customer names nobody has signed off.
- No proof list. The AI invents results to support the argument. Fix it with a closed library and a rule to leave out any point the library cannot support.
- Trivia in place of a problem. A recent post or podcast becomes the whole email. Fix it by writing the buyer problem first and using the post only to support it.
- Several asks. The email ends with a call, a resource and a referral request. Fix it with one ask, written word for word.
- A brief every rep edits. The approved wording drifts within weeks. Fix it with Can view or Chat access for reps and edit rights for the owner only.
- Consumer accounts for company data. Anthropic's commercial products and its consumer plans follow different data-use settings. Fix it by keeping briefs and proof on work plans.
- Skipping review because the score is high. A coach score rates the copy, not the claims. Fix it by running the review checklist on every draft.
FAQ
Can ChatGPT write sales emails from a brief?
Yes. ChatGPT can write sales emails from a brief when the brief is saved as project instructions and the approved proof is uploaded as project files, so every chat in that project starts from the same rules. Those instructions override each user's own custom instructions, and a rep still reviews every draft before it is sent.
How long should an AI sales email brief be?
One page is enough for most segments. The brief needs the audience, the reason to write now, a one-sentence buyer problem, two or three approved proof items, the clear ask, tone and length limits and banned claims. A brief that runs longer usually means two segments are sharing it, so split it in two.
Does CAN-SPAM apply to B2B sales emails?
Yes. The FTC's CAN-SPAM guide makes no exception for business-to-business email. A commercial message needs accurate header and From information, a subject line that reflects the content, a valid physical postal address and an opt-out honoured within 10 business days. This answer is general information, not legal advice.
Sources
Sources
- Gartner Survey Finds 69% of B2B Buyers Turn to Sales Reps to Validate AI-Generated Insights · businesswire.com
- CAN-SPAM Act: A Compliance Guide for Business · ftc.gov
- Gartner Survey Finds Consumers Want AI Shopping Help, But Not AI Purchase Decisions · businesswire.com
- FTC Policy Statement Regarding Advertising Substantiation · ftc.gov
- notion.com
- support.claude.com


