An AI chatbot can produce a cold email draft in seconds. It cannot tell a rep whether that draft is any good, and sending it unscored is how a sales team ends up with a thousand grammatically correct, strategically empty emails. A better prompt only gets a team a better starting point. What closes the gap between a draft and a sendable email is a repeatable scoring pass applied to whatever the AI produces, followed by specific instructions back to the AI about exactly what to fix.
A nine-category scoring rubric rates each part of an AI-drafted cold email from 1 to 5: Subject Line, Opener, Relevance, Bridge, Problem Framing, Intrigue, Offer, Ask, and Formatting & Readability. Any category that scores below a 5 gets a specific re-prompt instruction fed back to the AI, which rewrites just that element rather than the whole email, turning editing into a repeatable loop instead of a one-time guess.
Why AI Drafts Still Need a Human Scoring Pass
An AI model can mimic the structure of a good cold email without understanding why any part of it works. It has never felt the specific frustration a buyer feels when a process breaks down, so its attempts at empathy tend to read as generic. Spell-check didn’t eliminate the need to proofread, and a drafting tool doesn’t eliminate the need to apply judgment to what it produces. It’s a tool capable of making mistakes, particularly around the psychological and emotional weight of a message, which is exactly the part a scoring rubric is built to catch before the email goes out.
Saleshandy’s test of 12,000 cold emails found that AI-only campaigns produced a 4.1 percent reply rate, compared to 10.4 percent for human-written emails, and 14.7 percent for a hybrid approach combining AI output with human review before sending. That gap between AI-only and hybrid is the entire argument for a scoring step: the draft is a starting point, and the difference between those two reply rates is what a scoring pass is designed to close.
The Nine Scoring Categories
Each category gets scored 1 to 5 on the AI’s first draft. A score of 5 means the element is ready to send as written. Anything lower triggers a specific re-prompt rather than a general “make this better” instruction.
| Category | Low Score (1-2) Looks Like | Re-Prompt Cue |
|---|---|---|
| Subject Line | Generic, salesy, or blog-headline style | “Rewrite subject as 1-4 words, human, peer-style. Use a micro-scenario or curiosity gap, not a headline.” |
| Opener | A generic greeting (“Hope you’re well”) | “Start directly with the observable trigger or event. No greetings.” |
| Relevance | Could be sent to anyone in any role | “Anchor the message in their role and current trigger. Why them, why now?” |
| Bridge | An abrupt jump straight into the pitch | “Connect trigger to root cause to business impact before mentioning the solution.” |
| Problem Framing | Vague, abstract, or centered on the product | “Add a concrete micro-scenario the buyer can picture. Make it believable and uncomfortable, not melodramatic.” |
| Intrigue | Flat, no tension in the message | “Inject subtle discomfort, reputational risk, or doubt, kept credible, not exaggerated.” |
| Offer | Just a meeting or demo request | “Replace with a tangible, low-lift asset: a checklist, playbook, case study, or short video.” |
| Ask (CTA) | High-friction (“got 30 minutes for a call?”) | “Rewrite the CTA as a binary, low-friction yes or no. No meeting ask.” |
| Formatting & Readability | A wall of text, or robotic rhythm | “Add white space, vary sentence length, remove dashes, make it scannable on mobile.” |
A mid-range score (3-4) on any category gets a lighter touch than a full rewrite. A subject line scoring 3 or 4, for example, gets the cue “tighten subject to feel like an interruption, not marketing copy,” rather than the full rewrite instruction reserved for a 1 or 2. Problem Framing at a 3 or 4 gets “sharpen the example so it feels situational,” since the problem is already present, just underdeveloped.
How the Re-Prompting Loop Runs
The categories exist so a rep can isolate exactly what’s broken instead of regenerating an entire email and hoping the new version fixes the one part that failed. If a subject line scores a 2, the re-prompt sent back to the AI is specific and short: “Subject line scored 2. Rewrite subject line as 1-4 words, human, peer-style. Use a micro-scenario or curiosity gap, not a headline.” If the Problem Framing scores a 3, the re-prompt is just as narrow: “Problem scored 3. Add a concrete micro-scenario the buyer can imagine. Make it believable and uncomfortable, but not melodramatic.”
This narrow targeting matters because a full-email regeneration risks fixing the weak category while breaking a category that already scored a 5. Scoring each part separately and re-prompting only the parts that need it keeps everything that already works intact.
Scoring in Practice: What Changes Between a First Draft and a Rewrite
One worked example, built around a cold email to a CFO of a 300-store retail chain, shows how this plays out. The inputs given to the AI were a recent press announcement of store closures and cost-cutting, an offer (a 3-line audit checklist to spot hidden merchant fees), and proof (a peer CFO who found fees running 24 percent higher than expected and saved $410,000 in a year).
The AI’s first draft connected the store closures to a “tough but necessary call” framing that read as condescending, and it never drew the line between the store closures (a cost-cutting move) and merchant fees (also a cost-cutting lever). The re-prompt addressed both problems directly: “Be less judgmental in tone. Bridge needs to better connect the observation to the problem.” The rewrite that followed kept the same observation and the same proof point, but fixed the Bridge category specifically, connecting the two cost-cutting threads the first draft had left disconnected.
A second example, aimed at a CMO at a mid-market consumer brand whose company had just tripled its content output around a product launch, showed a different pattern: the AI’s first draft listed root causes but never named a business impact, leaving out the actual consequence (slipping campaigns handing market share to competitors while the CMO’s team scrambled). The re-prompt was a single line: “Clearly state the impact.” The category being fixed here was Problem Framing rather than Bridge, which is the point of scoring by category instead of reacting to a draft as a single pass-or-fail unit.
What a Rep Still Has to Bring
A score of 5 across all nine categories describes a structurally sound email. The rubric catches structural weakness: a buried ask, a generic opener, a flat subject line. Supplying the writer’s own voice, the specific phrasing that sounds like the rep rather than like anyone else who ran the same prompt, is work the rubric doesn’t do. The nine categories this rubric scores against are the same nine building blocks that define a complete cold email in the first place, and blending a rep’s own tone into a structurally sound draft stays manual work even after the scoring pass is finished.
This loop of drafting, scoring, and re-prompting is one piece of a broader approach to AI-assisted outreach covered in Gap Prospecting, where the scoring categories are paired with a master prompt built to generate the first draft the rubric then grades.
Frequently Asked Questions
What are the nine categories used to score an AI-generated cold email?
The nine categories are Subject Line, Opener, Relevance, Bridge, Problem Framing, Intrigue, Offer, Ask (CTA), and Formatting & Readability. Each one gets scored from 1 to 5 against the AI’s first draft, with a specific re-prompt instruction attached to any score below 5.
Why does an AI-drafted cold email need a scoring step instead of just being sent?
A test of 12,000 cold emails by Saleshandy found AI-only campaigns reply at 4.1 percent, compared to 10.4 percent for human-written emails and 14.7 percent for a hybrid of AI drafting with human review. The scoring step is what closes that gap: it catches the parts of an AI draft, tone, specificity, and emotional weight, that a model can produce in form but not in substance.
What does a re-prompt look like for a low-scoring subject line?
A subject line that scores a 1 or 2 for being generic or salesy gets the cue: “Rewrite subject line as 1-4 words, human, peer-style. Use a micro-scenario or curiosity gap, not a headline.” A subject line scoring a 3 or 4 gets a lighter note instead: “Tighten subject to feel like an interruption, not marketing copy.”
Should a low score on one category trigger a full email rewrite?
No. The categories are scored and re-prompted individually, which protects a category that already scored well, such as a strong Offer, from getting disturbed while a weak category, such as a flat Subject Line, gets fixed. Regenerating the entire email in response to one weak section tends to introduce new problems in sections that were already fine.
Can AI write a cold email’s tone and personality on its own?
Not reliably. An AI model hasn’t lived through the frustration, urgency, or ambition a buyer or seller feels, so its attempts to project empathy or humor tend to read as generic or derivative. The scoring rubric catches structural weaknesses, but a rep still has to apply their own voice and judgment to the draft the rubric helped sharpen.
What’s the difference between a score of 3-4 and a score of 1-2 on the same category?
A 1-2 means the element is fundamentally missing or wrong, such as an Offer that’s just a meeting request with nothing tangible attached, and gets a full rewrite instruction. A 3-4 means the element exists but is underdeveloped or generic, such as a problem that’s stated but not specific, and gets a lighter instruction aimed at sharpening what’s already there rather than replacing it.
0 Comments