Evaluate an AI email writer on the message you can actually approve. The draft should connect a verified fact to a relevant offer, make any inference visible and ask for an appropriate next step. Fluent prose is only one part of that review. A polished sentence can still describe the wrong company or imply a problem nobody has confirmed.
This guide works through the review of one outreach email. It does not select your market, run a sending system or promise a reply rate. The original post described combining contact data and research with drafting. Here the practical question is how to turn those inputs into a defensible message.
in
“We know so much about every single target decision-maker that it's almost scary. One of the most powerful workflows we run at Growth Cab combines fresh contact data with Claude Code to automate highly personalized outreach.”
Prepare a Short Evidence Brief
Give the writer the intended recipient, the offer you can actually deliver and the reason you are considering the account. Include a source and observation date for any fact you want to use. Record missing information separately. The model should not fill an empty field with a plausible detail just to complete the draft.
Separate the brief into observed facts, possible implications and the proposed help. A published job opening is an observation. A prediction that the company is struggling to onboard sellers is an inference. Your offer may be relevant to that situation, but the draft should not turn the inference into a claim about the buyer.
Draft Around One Connection
Ask for one clear connection between the evidence and the offer. Avoid stacking unrelated facts about a prospect to create the appearance of personal attention. A useful detail helps explain why the conversation could matter. A personal detail with no business relevance usually adds distraction rather than a reason to respond.
Keep the output narrow: a subject, a short message and an internal explanation of the evidence used. Keep that explanation in the internal review record. Require the writer to identify unsupported assumptions before it proposes language. That makes review easier than reverse-engineering an attractive paragraph after the fact.
Compare a Weak Draft With a Reviewable One
Consider a fictional company whose public careers page lists openings for account executives. Suppose the sender offers help documenting sales qualification. These inputs describe a fictional example. They allow us to examine whether the writing preserves what is known.
A weak draft says: Your team is scaling fast and losing deals because onboarding is broken. Our proven system will fix it. The job listings do not establish growth speed, lost deals or broken onboarding. The claimed outcome also needs evidence. Replacing the company name with a token would not repair the argument.
A more reviewable draft says: Your careers page lists account executive openings. We help teams document qualification decisions for incoming sellers. Are you reviewing that process as these roles are filled? This keeps the observable fact separate from the question. It still needs verification of the listing, recipient and offer before use.
Apply the Swap Test Carefully
Replace the company reference with a different account. If the reasoning remains unchanged, the draft may be too generic. Ask which verified detail actually affects the proposed conversation. The test checks relevance. Accuracy and authorization require separate review.
A message can fail the factual review even when it passes the swap test. An invented acquisition makes an email specific and wrong. A legitimate offer can also apply to several similar businesses. Do not force uniqueness through increasingly personal details. Keep the connection useful and proportionate to the information available.
Review the Ask and Remove Pressure
The ask should follow from the evidence and the offer. A question about whether a process is being reviewed can be appropriate when the need is uncertain. A meeting request needs a clear purpose. Avoid adding urgency, familiarity or a previous relationship that the record does not support.
Read the draft aloud and remove language you would not use in a real conversation. Check the recipient, company, dates, product references and every quantitative promise. Keep one next step. If there is no credible reason to contact the person, reject the draft instead of spending more time making it sound personal.
Return a Useful Review Decision
Classify the draft as accepted for the next authorized step, needs revision, or rejected. Record why: incorrect fact, unsupported inference, unclear offer, irrelevant detail or unsuitable ask. A reason helps improve the brief or drafting instruction. A request to make it better gives the writer little information about what went wrong.
Compare versions against the same brief so a rewrite cannot introduce new facts unnoticed. Preserve the accepted text and the evidence used to approve it. Acceptance as a draft is distinct from sending, delivery and a positive reply. None of those outcomes follows automatically from good writing.
Know What the Writer Cannot Resolve
A writer cannot fix an unsuitable offer or establish a buyer's interest from public information alone. Thin evidence may justify a simpler message or no message. Tool quality also needs to be assessed on your own work; this guide provides review criteria rather than a ranking of current products.
Disclosure: Growth Cab uses and partners with Prospeo, which was mentioned in the original workflow. That relationship is retained here for transparency. The method does not depend on a particular provider's current coverage, verification claims or pricing.
Use the AI SDR responsibility guide to assign ownership around the writing task.
Use the batch quality workflow when applying the same review across many records.
