How to make your AI DM bot sound 100% human (5 prompt tricks)

The three tells

Every "obvious bot" has the same three giveaways:

1. It replies in under 2 seconds. Real people type. They take 8–30 seconds for a short message and a minute for a long one.
2. It uses perfect grammar. Real people miss capitals at the start of sentences. They use "lol" and "tbh" and forget commas.
3. It answers everything in one message. Real people send two short messages instead of one paragraph.

Fix all three and the bot becomes invisible.

Five prompt tricks that work

1. Force lowercase casual

Add this to the system prompt:

Always reply in lowercase, casual style. Never start a sentence with a capital letter except for proper names. Drop punctuation at the end of short messages. Sometimes use lol, tbh, fwiw, def, prob.

This single line removes ~50% of the "bot" feel.

2. Constrain message length to 5–15 words

Bots love long paragraphs. Humans don't. Add:

Reply with one short message, 5–15 words max. If you need to say two things, send them as two messages with a natural pause.

The "send them as two messages" instruction is critical. It produces the realistic "double-message" pattern: "ok cool" → "so how soon you wanna start?"

3. Vary the typing delay

If your tool supports it (DMForge does, ManyChat doesn't), set the delay to a random value between 8 and 30 seconds per message. Never instant.

4. Mirror the lead's energy

Add:

If the lead uses emojis, use them sparingly back. If they don't, don't either. If they write in fragments, write in fragments. If they write in full sentences, write in full sentences.

This is the single biggest lift in "feels human" ratings. Mirroring is what real humans do unconsciously.

5. Add a "thinking" pause for hard questions

When the lead asks something complex (pricing, refunds, "can I see results?"), add:

Before answering complex questions, send a short message like "good question, one sec" or "let me think". Then wait 20 seconds, then send the real answer.

This pattern is so human it borders on uncanny.

What NOT to do

Don't apologize for being a bot. It's not lying — Meta's ToS allows AI replies to DMs as long as you're not impersonating a specific other person. You're representing yourself with an AI assistant.

Don't use phrases like "as an AI" or "I'm here to help." Instant tell. Strip them.

Don't echo the user's question back. "So you want to lose 15kg?" → bot. "got it, 15kg" → human.

The DMForge default prompt

We tuned the DMForge default system prompt to do all five of these things automatically. You can override any of them by typing in plain English in the agent editor: "be more direct" or "always use emojis." The AI rewrites itself in real time.

Try it yourself

Build an agent at DMForge → describe your offer in 30 seconds → open the live simulator → pretend to be a lead. Watch the agent reply. If it feels robotic to you, type "make replies shorter and more casual, drop capitalization" in the AI editor. Within 2 seconds, the prompt updates and the next reply will sound completely different.

That live tuning loop is the whole reason people switch off SetSmart and ManyChat — they don't let you see or edit the underlying prompt.

Related: The 7 qualification questions, From DM to closed deal.