How to Make AI Sound More Human, and Like You: A Guide for Agents

For real estate agents, insurance agents, loan officers and travel advisors whose AI drafts read like a form letter.

Why your AI drafts sound like a robot

You ask for a quick follow-up email and get back something that opens with “I hope this message finds you well,” promises to “go above and beyond,” and closes with “Please don’t hesitate to reach out.” Nobody you know writes like that. Your clients can tell.

The reason is not mysterious. When a chat tool knows nothing about you, it writes the safest, most average version of the message it can. Average tone, average vocabulary, average enthusiasm. It has no idea you sign off with your first name and a phone number, that you never say “dream home,” or that your clients are used to three short sentences, not three paragraphs.

That is a context problem, and it has a context fix. But first, a warning about the advice you’ll find when you search for one.

The “sound more human” advice was written for bloggers

Search “how to make AI sound more human” and many of the top results offer some version of the same list. Vary your sentence length. Cut the filler. Add personality. Add personal stories and experiences.

The first three are fine. The last one needs a line drawn through the middle of it.

A true, relevant detail from you is fine. “We talked about the commute on Tuesday” is real if it happened. The problem starts when you ask the AI to add a personal detail you haven’t supplied, because it may invent one. An agent who asks AI to “add a personal touch” can get sentences like “I fell in love with this kitchen the moment I walked in,” “You’re going to love the peace of mind this policy gives you,” or “I just know you’ll have the trip of a lifetime.” Every one of those is a claim. Did you walk into that kitchen? Does the policy actually cover what the client is worried about? Is that trip confirmed?

For an agent, sounding human does not mean sounding warmer at any cost. It means sounding like you, while saying only what is true. Keep those two jobs separate and the rest gets easy.

Five fixes that work for agents

1. Write your voice down once

Give the tool the same things a new assistant in your office would need on day one:

• Your voice in three words. “Warm, plain-English, calm” or “direct, friendly, no fluff.” Three words give it something to aim at. A blank gives it the brochure.
• Your sign-off, exactly. Name, company, the contact line you actually use.
• Phrases you use. The way you open, the way you say “call me.”
• Phrases you never use. “Dream home.” “Won’t last long.” “Act now.” “Lowest price guaranteed.” “Dream vacation.” “Please don’t hesitate to reach out.” Your list will be different. Write it anyway.

Save it where the tool will reuse it, in a Project or your account’s saved instructions, so you are not retyping it every time. If you want the details on how each tool handles that, the Claude vs ChatGPT comparison for agents covers it.

2. Show it two or three real messages

Descriptions of a voice only go so far. Examples go further.

Pick two or three messages that sound like you on a good day, and only ones your office allows you to use this way. Remove names, addresses, numbers and anything identifying or sensitive, including circumstances a client would recognize. If you can’t use real messages, write two short samples yourself in the voice you want. Paste them in and ask:

That last line matters for two reasons. The one-line description tells you whether it understood you before you rely on it. And “copy my style only” lowers the odds of a detail from an old message showing up in a new one. It doesn’t rule it out, and that kind of slip is hard to spot because it reads like something you would say. The fact check at the end is your best chance of catching it.

3. Give it the specifics

Generic writing is usually a symptom of generic input. “Write a follow-up to [CLIENT] after their showing” gives the tool nothing to work with, so it reaches for filler, and filler is where robotic phrasing lives.

Give it the real details instead: what they said they liked, what worried them, what happens next, and when. Then add the rule that does more for your voice than any style tip:

A specific message sounds human because specifics are what people actually write about. Specifics don’t stop invention on their own, which is why that rule is there and why the fact check still comes last. What they do is give the tool something real to say, so it has less reason to reach for stock phrases.

4. Fix the tone with one short line

When a draft is close but off, don’t rewrite the prompt. Add one short instruction to the same chat:

• “Warmer.”
• “Shorter, three sentences.”
• “Less salesy.”
• “More formal.”
• “Write it like I’d text a client I like.”

Then add the part people forget: “Keep every fact unchanged.” A tone fix should change how something is said, not what is said. That instruction lowers the risk. It doesn’t lock anything. “Warmer” can still sneak in reassurance you never offered, and reassurance in this business can read like a promise, so the new draft gets checked again.

5. Read it for voice, then read it for facts

Two separate passes, because one pass does both jobs badly.

The voice pass. Read it the way the client will. Would you say this out loud? Is the sign-off yours? Did any phrase from your never-use list sneak back in?

The fact pass. Put the draft next to your notes. Every number, date, feature, status and promise in the draft has to be on your list. If a sentence makes you sound more confident, more certain or more personally involved than you are, it goes, no matter how human it sounds.

What it looks like

Here is a small example, written for this article to show the difference. It is not a recorded AI output.

The facts the agent supplied (insurance renewal, details anonymized):

• Client: [CLIENT], homeowners policy renews [RENEWAL DATE].
• Premium is going up. The carrier’s renewal notice gives the new amount: [AMOUNT FROM NOTICE].
• The agent can shop the policy with other carriers if the client wants.
• Next step: reply or call [PHONE] by [REPLY-BY DATE].
• Voice: warm, plain-English, calm. Sign-off: “Here when you need me, Dana.”

The robot version:

Count the problems. Five stock phrases, from “Dear Valued Client” to “Please don’t hesitate to reach out.” “Your coverage remains comprehensive and you’re fully protected” is a coverage promise nobody supplied. “Industry-wide” is a reason nobody supplied. “The best possible rate” is a promise too.

A version based on the supplied voice instructions:

Shorter, warmer and more human, and it adds nothing the agent didn’t give it. That is the whole trick.

Where the Boltz Creations workflow guides fit

Every guide in the series starts with a context block that asks for your voice, your sign-off and the phrases you never use, and each one gives you a way to fix a draft that misses it.

• Claude Workflow for Real Estate Agents: voice words, banned phrases and sign-off in the Agent Context Block, plus copy-and-paste repairs for drafts that come out too formal, too generic, too long or off-brand.
• Claude Workflow for Insurance Agents: a “teach Claude your voice” workflow built on your own anonymized messages, a snippet library, and one-line tone fixes.
• Claude Workflow for Mortgage Loan Officers: a voice workflow that tells the tool to copy your style only and not carry names, numbers or borrower details over into other tasks.
• AI Workflow System for Travel Agents: a voice library workflow that turns five of your real messages into bullet points for your context block, and a supplier-description rewrite that keeps every fact and drops the marketing fluff.

They are drafting aids, not a substitute for your review. Every draft is still yours to check before it goes out.

For more on the same problem from a real estate angle, see Claude for Real Estate Agents: A Practical System. For insurance, see How Agents Can Fix Generic AI Insurance Emails.

The short version

Write your voice down once. Show it permitted examples and tell it to copy style, not facts. Give it specifics so it has something real to say. Fix tone with one line and tell it to keep the facts. Then read twice, once for voice and once for truth. The instructions lower the risk, and your final check is still necessary. A draft that sounds exactly like you and says one thing that isn’t true is not a human-sounding draft. It’s a problem with your name on it.