Recommended resources
The freely available resources this lesson will direct you to.
Hello again. In the previous lesson, you improved a weak sales update by identifying one prompt defect at a time. That helps produce a better draft, but it does not answer the final question: Is the draft fit to use? An AI response can follow your requested format and still misstate a prospect’s position or sound wrong for the recipient.
This lesson gives you a four-part review checklist—accuracy, relevance, specificity, and tone—and shows how to use it to make a clear use, revise, or hold decision. We’ll continue with the fictional Harborline opportunity so you can compare an AI-generated message with the notes it was supposed to use.
A draft is not an approved message
AI is useful for producing a first version quickly. Approval is a different task: it requires someone to judge whether the message is true, useful to this recipient, concrete enough to act on, and appropriate to send. A fluent sentence deserves no special trust merely because it sounds confident.
Read the human-review portion of HubSpot’s Personalization in AI Prospecting: What Actually Works. It identifies common mistakes in AI-assisted outreach and explains why editing is part of the workflow.
Personalization in AI prospecting: What actually works
Read HubSpot’s guidance on human editing and approval to see how factual accuracy, relevance, and tone become practical review responsibilities.
In “4. Implement human editing and approval,” start with the review principle. Continue through the checklist and the paragraph beginning “Build editing checklists”; stop after the common mistakes. Notice that a message can fail by being generic even when it contains no obvious falsehood.
For this lesson, keep the four checks distinct. Accuracy asks whether claims are supported. Relevance asks whether the content serves the recipient and the sales task. Specificity asks whether it uses concrete, warranted detail rather than interchangeable or invented detail. Tone asks whether the wording suits the relationship and your organisation’s voice. A message may pass one check and fail another.
The four-part quality checklist
Before reviewing, place the AI output beside its task, intended reader, and permitted source notes. Otherwise, “looks good” can quietly become your standard. Use the same four questions each time:
| Check | What would earn a pass? | What should you flag? |
|---|---|---|
| Accuracy | Material claims match the supplied evidence; proposals and possibilities are not presented as agreed facts. | Invented results, unsupported figures, or a request described as an approval. |
| Relevance | The message addresses this reader’s situation and the purpose of this communication. | True but distracting details, or a sales ask that ignores what the prospect requested. |
| Specificity | It uses useful, supported details and gives a clear, appropriate next step. | Generic praise, vague value claims, or precise-sounding details with no evidence. |
| Tone | It is respectful, clear, and consistent with the desired professional voice. | Presumptive familiarity, pressure, exaggeration, or a robotic script. |
HubSpot’s AI-Powered Email Content Suggestions That Actually Convert Leads provides a related email checklist. Read it as a comparison, not as a requirement to include every item in every sales message: for example, a plain reply to a prospect does not need an internal link.
AI-powered email content suggestions that actually convert leads
Read HubSpot’s quick quality checklist for an example of turning broad editorial goals into checks a reviewer can apply before sending.
Under “AI Email Content Generation Guardrails,” find “Layer 2: Compliance,” then “Quick Quality Checklist.” Read the checklist, focusing on relevance, verified claims, tone, and a clear call to action. Continue through the paragraph on expert review for sensitive topics. In the “Example disallowed list” immediately below, notice how phrases such as “Guaranteed results” make a tone rule easier to enforce.
The checks work best when you record a reason, not just a rating. “Tone: revise—‘lock in our work now’ pressures a prospect who has only requested an outline” gives a colleague something to fix. “Tone: bad” does not.
Make the decision explicit
Use three outcomes:
- Use only when all four checks pass for the intended use.
- Revise when the draft is grounded in the evidence but lacks relevance, useful detail, or the right tone. Review the revised version again.
- Hold when a material factual claim is unsupported, contradicted, or cannot yet be checked against an available source. Remove or verify it before the message goes out; polishing the wording is not enough.
An unsupported claim is not necessarily false. The point is that you do not have grounds to send it as fact. For an internal draft, you may instead mark it clearly as unverified; for prospect-facing copy, do not quietly leave it in.
Start with claims, then judge usefulness
Accuracy comes first because an elegant message built on a false premise is still unusable. Read the draft sentence by sentence and ask what each factual claim rests on. Pay particular attention to commitments, dates, numbers, results, and statements about what a buyer has decided. Distinguish “the VP is expected to review options next week” from “the VP will sign off next week.” They describe different stages of a sale.
The image below depicts a fuller fact-checking process, from identifying a claim through documenting and updating a conclusion. For this lesson’s sales review, its immediate lesson is simpler: identify the claim and its context before deciding whether the available evidence supports it. We will return to more extensive source verification later in the course.
Once factual claims are checked, ask whether the supported material belongs in this output. A detail can be accurate but irrelevant to an operations lead who asked for a method outline. Conversely, “market changes create exciting opportunities” may be harmless in isolation but too generic to help that reader understand why you are writing.
Specificity is not the same as adding numbers or names. “Two regional markets” is specific because it appears in the notes. Naming those markets without evidence would make the message more detailed and less trustworthy. If the notes do not support personalisation, the honest option is a more general draft or a request for better source material—not manufactured detail.
Finally, read the message as its recipient might. Does it assume a close relationship that does not exist? Does it push for a deal when the buyer asked only for information? For an internal sales-manager update, a direct, compact tone may be suitable; for a prospect-facing email, the same shorthand might feel abrupt. Tone is judged against audience and purpose, not against a universal rule that every sales message must sound cheerful.
Apply the checklist to a Harborline email
Here are the fictional source notes from the previous lesson:
Harborline Components is exploring two regional markets. An operations lead said the team is concerned that its demand forecasts use outdated reports. The operations lead requested a short outline of Northstar Research’s method and indicative scope. The VP of Sales is expected to review options next week. Budget, decision authority, procurement timing, and pricing were not confirmed. Northstar provides market-sizing research and buyer interviews.
Suppose an AI tool drafts this reply to the operations lead:
Subject: Ready to start your approved expansion project
Hi,
Congratulations on expanding into two new markets! Your outdated forecasts are costing you sales. Northstar’s proven method will increase revenue by 30% in 60 days. With your VP signing off next week, let’s lock in our work now. I can send a proposal Friday.
Best,
Northstar
The email is short and has an apparent next step. Neither quality makes it ready to send.
Accuracy — hold. “Approved expansion project,” “costing you sales,” a 30% revenue increase, and “VP signing off” are not established by the notes. Harborline is exploring markets; its VP is expected to review options. The precise result sounds persuasive but has no supporting evidence.
Relevance — revise. The operations lead requested a method and indicative-scope outline. Pressing for a proposal and a commitment to start work skips over that request. The draft has noticed the opportunity but not responded to the person’s immediate need.
Specificity — revise. The message mentions two markets and forecasts, yet most of its apparent detail is unsupported. It does not say what the requested outline will cover. That is a useful reminder that a draft can contain many specifics while failing a supported specificity check.
Tone — revise. “Congratulations” presumes the expansion has happened; “let’s lock in our work now” adds pressure before any purchase is confirmed. The voice is more aggressive than the situation warrants.
A suitable revision would be:
Subject: Outline of Northstar’s method and scope
Hi [Operations Lead],
Thanks for requesting a short outline of our method and indicative scope. You mentioned that Harborline is exploring two regional markets and is concerned that its demand forecasts rely on outdated reports. I’ll send an outline of how our market-sizing research and buyer interviews could inform your assessment. Is there a particular question you’d like it to address before the VP reviews options next week?
Best,
[Sender]
This version treats the prospect’s statements as statements, the seller’s intended action as an action, and a possible benefit as a possibility—not a guaranteed result. It answers the request and gives the recipient a specific, low-pressure way to shape the outline. A human sender should still confirm the wording reflects what they intend to provide and fits their organisation’s voice before using it.
The review is not a contest to produce the most elaborate email. Its purpose is to find the smallest message that is accurate, relevant, concrete enough to be useful, and appropriate for the relationship.
Conclusion
A dependable review starts with the task and source notes, not with the AI draft’s fluency. Check accuracy for supported claims, relevance for fit with the reader’s need, specificity for useful details that the evidence actually permits, and tone for professional judgment. Record the reason for each failure and decide whether to use, revise, or hold the output. Material unsupported claims are a hold, even when the rest of the writing is polished.
Next, we’ll consider what information should not enter a public AI tool in the first place: how to anonymise sensitive client, prospect, and organisational details before prompting.