KoltrixTone control for AI drafts: formal, friendly and firm
The same customer reply written formal, friendly and firm, with notes on what changed, prompts that steer AI tone, and how to strip over-apology and filler.
On this page(10 sections)
Left to its defaults, an AI drafting tool tends to produce one voice: warm, apologetic, slightly corporate, and about twice as long as it needs to be. That voice suits almost no real conversation perfectly.
The fix is not a better model. It's knowing which tone a situation needs and telling the model clearly. This post takes one customer email, writes the reply three ways, explains what changed in each, and then covers the prompts and edits that get you there.
The incoming email
Here's a realistic message to a SaaS support inbox:
Hi, I was charged again this month even though I thought I cancelled in March. Can you refund it? This is the second time this has happened. — Alex
The facts, for this example: Alex cancelled a different workspace in March; the charged workspace is still active. The policy allows a one-time courtesy refund, and Alex already received one last year.
So the right answer is "no refund this time, here's why, here's how to cancel properly." The tone is what varies.
Version 1: Formal
Hello Alex,
Thank you for contacting us about this month's charge.
Our records show that the cancellation in March applied to the workspace "Design Team." The charge this month is for a separate workspace, "Alex Personal," which remains active.
As a courtesy refund was issued on this account last year, we are unable to refund this charge. To prevent future charges, the "Alex Personal" workspace can be cancelled under Settings, then Billing. Cancellation takes effect at the end of the current billing period.
Please let us know if you have any further questions.
Kind regards, Sam, Support
What makes it formal: complete sentences, no contractions, passive constructions ("a courtesy refund was issued"), neutral sign-off. It suits enterprise customers, procurement contacts, or anyone who wrote formally first.
Version 2: Friendly
Hi Alex,
Thanks for flagging this, and I get why it's frustrating to see a charge you didn't expect.
I looked into it: the March cancellation was for your "Design Team" workspace. This month's charge is for "Alex Personal," which is still active, so that's where the confusion came from.
I can't refund this one, since we already did a one-time courtesy refund on your account last year. But you can cancel "Alex Personal" under Settings → Billing, and you won't be charged again after this period ends.
If you'd like, reply here and I'll double-check that nothing else is still active.
Sam
What makes it friendly: contractions, first person ("I looked into it"), acknowledging the feeling once without groveling, an offer of extra help. The "no" is still clear, which is the key: friendly is not vague.
Version 3: Firm
Hi Alex,
This month's charge is for the workspace "Alex Personal," which is active. The March cancellation applied to a different workspace, "Design Team."
We issued a one-time courtesy refund on this account last year, so this charge isn't eligible for a refund.
To stop future charges, cancel "Alex Personal" in Settings → Billing before your next renewal date.
Sam
What makes it firm: short, factual, no apology, no softening phrases, no offer to revisit. It's not rude. It's appropriate when a customer has pushed repeatedly on the same request, or when the message was abusive and you want to stay professional without inviting more.
What changed across the three
| Element | Formal | Friendly | Firm |
|---|---|---|---|
| Greeting | "Hello Alex" | "Hi Alex" | "Hi Alex" |
| Acknowledgment | Thanks for contacting | Names the frustration once | None |
| Explanation | Full, neutral | Full, conversational | Minimum needed |
| The "no" | Clear, passive voice | Clear, first person | Clear, direct |
| Extra offer | Generic | Specific and genuine | None |
| Length | Medium | Medium | Short |
Notice what stays constant: the facts, the decision, and the next step. Tone changes how the answer feels, never what the answer is.
When firm is the right choice
Teams often avoid firm replies because they feel unfriendly. But there are situations where softness causes harm:
- Repeated requests after a clear no. Each soft reply signals the answer might change.
- Abusive messages. Stay calm and brief. Long, warm replies to abuse tend to invite more.
- Policy you can't bend, such as security procedures or legal requirements. Hedging creates false hope.
- Requests to bypass verification. "We can't change the account owner without confirmation from the current owner" should never sound negotiable.
Prompts that steer tone
Vague instructions ("be professional") give vague results. Describe the tone through concrete behaviors:
Write a reply in a friendly tone:
- Use contractions and first person.
- Acknowledge the customer's frustration in one short clause, not a paragraph.
- State the decision clearly in the second paragraph.
- No more than 120 words.
- Do not apologize more than once. Do not use "unfortunately."
Write a reply in a firm tone:
- State facts and the decision directly.
- No apology, no softening phrases, no offers to reconsider.
- Three short paragraphs maximum.
- Remain polite: no sarcasm, no blame.
It also helps to give the model the decision explicitly ("We will not refund this charge because...") rather than asking it to decide. Tone control works best when the model is phrasing a decision, not making one.
Filler to cut from any tone
Whatever tone you choose, these phrases usually make drafts weaker. Search for them during review:
- "I hope this email finds you well"
- "I sincerely apologize for any inconvenience this may have caused"
- "Please don't hesitate to reach out"
- "I completely understand your frustration" (when followed by not addressing it)
- "Unfortunately" at the start of every other sentence
- "Just wanted to" and "just checking"
Over-apology is the most common AI tone problem. One acknowledgment is empathy. Three is anxiety, and it can sound like admitting fault you don't have.
A quick tone-picking guide
- Customer wrote formally, or it's a contract/procurement conversation: formal.
- Most everyday support: friendly.
- Third request for the same exception, or abusive language: firm.
- When in doubt: friendly, but shorter than your first draft.
Key takeaways
- Tone changes how a reply feels; the facts and decision stay identical.
- Describe tone to the AI as concrete behaviors, not adjectives.
- Give the model the decision and let it handle phrasing.
- Firm is a legitimate tone for repeated asks and abuse, as long as it stays polite.
- Cut stacked apologies and stock filler from every draft, whatever the tone.
Start with Koltrix
Your domain, one inbox, and an API that sends.
A team inbox where AI sorts and drafts (nothing is sent without your click), plus the transactional API and SMTP relay your product sends with. 7 days free, no card.
