AI Drafts Can Sound Human on WhatsApp

AI drafts can be correct and still sound wrong. The first time we showed a Poli Digital suggestion to a sales rep, she said, "That doesn't sound like me at all." It answered a delivery question, but its stiff, FAQ-like wording did not match how she spoke to leads.

The main problem with AI drafts is the distance between formal output and a human voice. WhatsApp feels direct and personal. A reply that sounds like a support ticket breaks that expectation in one sentence.

Draft quality is not fixed. Give the system better inputs and the suggestions improve. Here is what works.

Why WhatsApp Drafts Miss the Mark

WhatsApp falls between business email and casual conversation. It is more formal than a text to a friend, yet much less formal than a support ticket. A new lead might write: "Oi, vi o produto de voces no Instagram, tem como me falar mais sobre o plano?" The tone is warm, direct, and slightly informal.

Generic business writing can turn that into: "Thank you for your inquiry regarding our product. I would be happy to provide additional information about our plan options."

Nobody talks that way on WhatsApp. The lead was casual, so answer casually.

This is a tone calibration issue, not a model flaw. The system cannot know how your team communicates until you show it. Unconfigured drafts often fail because nobody defined what "good" sounds like.

What Tone Profiles Do

A tone profile sets three things: your register, formal, semi-formal, or casual; your preferred sentence length; and the phrases your team uses. In Poli Digital, tag 20 to 30 sent messages as positive examples so the model can calibrate.

Choose those examples carefully.

Messages that got a positive reply or moved the conversation forward are better examples than messages met with silence. A technically accurate message can still land badly.

Skip your most templated messages. If every new lead gets the same greeting, the AI will copy that wording. You want natural variation, not one repeated script.

For small Brazilian sales teams, semi-formal usually works well: complete sentences, little corporate jargon, and clear next steps. Aim for friendly, direct, and specific.

Four Draft Types to Review

AI suggestions are not equally dependable. From our work with teams handling different message volumes, we see a clear split between drafts that are close to ready and drafts that need more editing.

Factual answers are the most reliable. For questions about price, availability, or a feature, the task is narrow: find the right information and state it clearly. Usually, only small tone edits are needed.

Follow-up nudges for a quiet lead are moderately reliable. The structure is often right, but the wording may sound salesy or formal. Allow time to edit a sentence before sending.

Objection replies need more care. When a lead says the price is too high or they need time to think, the draft may sound defensive. Read it, then often rewrite the opening completely.

Never send an empathetic escalation draft without human review. In a tense moment, the AI lacks enough context to judge tone, and a poor reply can end a recoverable relationship.

This is not a case against AI drafts. It is a guide to where they save time and where a human check is worth a few extra seconds.

The Two-Second Draft Check

The most effective teams treat AI suggestions as starting points, not finished replies. In practice, that requires a simple review habit.

Before sending, read the first sentence silently. If you would say it to this person, send it. If it sounds like a form letter, rewrite that sentence first.

The opening sets the register. A natural first sentence usually carries the rest along. Starting with "Thank you for reaching out" tends to make the whole reply more formal.

Some teams share a list of openings to avoid: "I hope this message finds you well," "Thank you for your inquiry," and "As per our conversation." These are not always wrong, but on WhatsApp they can signal an uncalibrated AI opening and create distance.

When to Disable Drafts

AI suggestions do not fit every conversation. Knowing when to switch them off matters as much as knowing when to use them.

Turn drafts off during a negotiation with a warm lead once the discussion becomes specific to their situation. The AI lacks context, and reviewing suggestions may slow you down.

Turn drafts off after frustration or disappointment. Even a technically suitable suggestion can feel canned when the customer needs personal attention.

If a rep knows the customer and has spoken with them five times, the draft may miss that relationship's register. The message should feel personal, and matching it may require significant editing.

In Poli Digital, disable drafts for one conversation without changing global settings. The rest of your queue stays unchanged.

Read the Customer's Tone First

Before accepting a suggestion, study the customer's wording. Short message, no punctuation, conversational phrasing? Match that energy. Longer message with details and a direct question? A more structured reply fits.

The draft reflects your team's general tone. The customer's message signals the tone they want back. A small edit to match that signal can make the reply feel much more personal.

WhatsApp drafts also need different calibration from email drafts. Even in Poli Digital's unified inbox, channel context matters. A reply that works in email may feel too formal in a WhatsApp thread with the same customer.

Improve Drafts Over Time

Drafts improve as the system sees more sent messages. When a rep substantially edits a suggestion, the final version gives better calibration feedback than the original.

Teams that edit drafts regularly and sometimes flag unhelpful suggestions see better results after 3 to 4 weeks than teams that accept every draft or ignore them.

The aim is not fully automated replies. It is a first draft fast enough for reps to focus on judgment, context, and relationship decisions instead of sentence construction. Human review remains part of sounding like a person who cares about the conversation.