AI Email Drafts That Actually Sound Like You: Voice Scoring and AI Slop Detection Explained

Introduction: The AI Slop Epidemic

Open your inbox right now. Scroll through the last 20 emails you received. How many of them contain phrases like:

• "I hope this email finds you well"
• "Please don't hesitate to reach out"
• "I wanted to circle back on our previous conversation"
• "Thank you for your prompt attention to this matter"
• "I look forward to hearing from you at your earliest convenience"

If you counted more than five, you're witnessing the AI slop epidemic firsthand.

In 2026, an estimated 40% of professional emails are either fully AI-generated or heavily AI-assisted. Tools like ChatGPT, Grammarly, and Superhuman's Auto Drafts make it easy to generate email text quickly. The problem? They all produce the same generic, over-polished, unmistakably artificial prose.

The result is an inbox full of emails that all sound the same — written by different people but all channeled through the same AI filters. Recipients can tell. Trust erodes. Relationships feel transactional rather than human.

Presend was built to solve this exact problem.

What Is AI Slop?

"AI slop" is our term for email text that exhibits telltale signs of AI generation:

Excessive Formality: AI models default to formal, corporate-safe language even when the conversation is casual. "Per our discussion" instead of "Like we talked about."

Hedge Words: AI loves hedging. "I believe," "It might be worth considering," "Perhaps we could explore the possibility." Humans are more direct.

Filler Phrases: Stock openings ("I hope this email finds you well") and closings ("Please don't hesitate to reach out") that add no information but pad the word count.

Perfect Grammar Everywhere: Ironically, perfect grammar can be a tell. Humans use contractions, fragment sentences, and informal punctuation in email. AI rarely does unless specifically prompted.

Generic Structure: AI emails follow a predictable pattern: greeting → context → request → closing. Real emails are messier, more direct, and more varied.

Lack of Personality: The biggest tell. AI emails have no quirks, no humor, no personality. They sound like they were written by a committee, because they essentially were — trained on millions of committee-written documents.

How Presend's Voice Scoring Works

Presend approaches the AI slop problem with two complementary systems: Voice Scoring and AI Slop Detection.

The Voice Check

The voice check answers one question — is this you? — and answers it as a verdict, not a percentage.

How it learns your voice: because Presend stands at the send gate, every message you actually send is a labelled sample of your real writing, on that platform, approved by you. The profile builds itself. Nobody pastes samples into a settings page.

The comparison is Burrows's Delta against your own corpus, across three feature families:

• Character trigrams, which stay robust at email length
• Function-word frequencies, the strongest signal when there is length to work with
• Style scalars — punctuation, sentence rhythm, contraction rate, lowercase "i"

Each feature is z-scored against the mean and standard deviation measured across *your own* samples, then averaged. Distance therefore comes out in units of your own variation rather than against an abstract norm, and calibration is leave-one-out over your corpus.

The four verdicts:

• Sounds like you — the draft sits inside your normal range.
• Drifts — it does not, and the sentences that drove the verdict are underlined in the compose box.
• Too short to compare — under the word floor for that channel.
• Not enough of your writing yet — under five samples or four hundred words. Presend says so rather than inventing a number. A voice verdict from two samples is a horoscope.

It is deterministic: the same draft and the same profile give the same verdict every time. An earlier version scored with a model call, the number jittered between keystrokes, and people learned to ignore it.

The Slop Check

Separately, and needing no profile at all, Presend counts model tells:

• Stock phrases and clichés that appear disproportionately in AI output
• Structural tells — uniform sentence length, tidy greeting-body-closing scaffolding, wrap-up paragraphs that summarize what was just said
• Invisible characters pasted in from a chat window, which are near-conclusive on their own
• Assistant leftovers such as "as an AI language model" or "Certainly!"

The result is clean, traces, or slop, based on distinct tells rather than raw hits — one cliché used three times is a habit, not three pieces of evidence. Three unrelated tells is a signal.

The Three Rewrite Modes

When a draft drifts from your voice or reads as slop, Presend offers one-click rewrites in three modes:

More Me

This is Presend's signature rewrite mode. It takes your draft and rewrites it to match your personal voice model:

Before (slop: three tells, drifts from your voice):

Dear Michael, I hope this email finds you well. I am writing to follow up on our recent discussion regarding the Q3 marketing strategy. I believe it would be beneficial for us to schedule a meeting at your earliest convenience to discuss the key deliverables and align on next steps. Please don't hesitate to reach out if you have any questions. Best regards, Alex

After "More Me":

Hey Michael — following up on our Q3 marketing chat. Can we grab 30 min this week to nail down the deliverables? I've got some ideas I want to run by you. Let me know what works. Alex

Same message. Same intent. But one sounds like a human who knows the recipient, and the other sounds like a ChatGPT prompt response.

Shorter

Condenses your email while maintaining your voice. Removes filler, tightens language, and gets to the point faster.

Before (247 words):

A four-paragraph email explaining a simple project update with extensive context-setting and polite filler.

After "Shorter" (62 words):

A focused two-paragraph update that delivers the same information in a quarter of the space.

Less AI

Specifically targets and replaces AI-generated patterns while preserving your core message. This mode is ideal when you've used AI to help draft but want to remove the telltale signs.

It replaces:

• "I hope this email finds you well" → removed (unnecessary)
• "At your earliest convenience" → "when you get a chance"
• "I wanted to circle back" → "following up"
• "Please don't hesitate to reach out" → "let me know"
• "Thank you for your prompt attention" → "thanks"

Why Presend Rewrites Instead of Generating

Plenty of tools will write the email for you from a one-line prompt. Presend deliberately does not, and the reason is the same one that makes the voice check work at all.

A rewrite starts from something you wrote. It has your structure, your emphasis, the thing you actually meant — and the model's job is narrowed to moving it closer to how you normally say things. A generated draft starts from the model's idea of the message, and every round of editing pulls you toward its defaults rather than yours. Feed those generated messages back through the send gate and they quietly become part of the corpus your voice is measured against.

So the flow is: you write badly and fast, Presend tells you where it drifted, and one click pulls it back toward you.

Here is a draft written the way people actually write when they are in a hurry and let a chatbot tidy it up:

Dear Design Team, I am writing to inform you that we have made the decision to postpone the homepage redesign project to the next sprint cycle. This decision was necessitated by the fact that the API development work has not yet been completed, and we believe it is important to ensure that all dependencies are in place before proceeding with the redesign. I appreciate your understanding and flexibility in this matter. If you have any questions or concerns, please do not hesitate to reach out. Best regards, [Your Name]

And the same message after a More Me rewrite:

Hey team — heads up, we're pushing the homepage redesign to next sprint. The API work isn't quite ready yet and I don't want us building on a moving target. Nothing's changing about the design itself, just the timeline. I'll update the sprint board today. Shout if you have questions. Thanks!

Compare the first version to what a generic AI tool produces when asked to be professional:

Dear Design Team, I am writing to inform you that we have made the decision to postpone the homepage redesign project to the next sprint cycle. This decision was necessitated by the fact that the API development work has not yet been completed, and we believe it is important to ensure that all dependencies are in place before proceeding with the redesign. I appreciate your understanding and flexibility in this matter. If you have any questions or concerns, please do not hesitate to reach out. Best regards, [Your Name]

Same message. Wildly different impact. The first sounds like a human teammate. The second sounds like a corporate memo from someone who doesn't know the team.

Why Voice Authenticity Matters

1. Trust and Relationships

People respond differently to authentic communication. When your emails sound like you, recipients feel a genuine human connection. When they sound like AI, recipients (consciously or not) treat the interaction as less personal and less trustworthy.

2. Professional Reputation

In senior roles, your communication style is part of your personal brand. CEOs, VPs, and team leads who suddenly start sending AI-polished emails lose the authentic voice that earned them trust and respect.

3. Response Rates

Authentic emails get faster, more engaged responses. Generic AI emails get perfunctory replies. If you're in sales, recruiting, or business development, voice authenticity directly impacts your conversion rates.

4. Internal Culture

When everyone on a team uses AI to generate emails, internal communication becomes sterile. Ideas feel less inspired. Feedback feels less genuine. Culture suffers when humans stop communicating like humans.

Privacy and Voice Data

A natural question: if Presend analyzes your writing style, where does that data live?

The voice profile is computed entirely on your machine. No model call is involved: the comparison is Burrows's Delta over character trigrams, function words and style scalars, measured against your own past messages. The corpus is never synced, never attached to a scoring request, and never sent anywhere.

Quoted reply chains are stripped before storage, so the profile keeps what you wrote and never what was written to you. It is capped at sixty messages per channel, email and chat are kept separate, and PRESEND_VOICE_FORGET erases it. Your voice model exists only in your browser's local storage — if you uninstall Presend, it's gone.

Drafts you score do go to Presend's backend, but message bodies are not stored there either: what is kept is a one-way hash that cannot be reversed into your text.

Getting Started with Voice Scoring

Voice scoring activates automatically after Presend has analyzed a small sample of your sent emails (typically 20–30 messages). The more emails it analyzes, the more accurate your voice model becomes.

Once active, every email you compose will show your Voice Score and AI Slop Score in the Presend sidebar. No configuration needed — just write naturally, and Presend will tell you when something sounds off.

Add Presend to Chrome — Free →