The Best Time to Use AI Is Right Before You Hit Send

Most of the advice about AI is about the blank page. Use it to write the email. Draft the proposal. Start the post. And sure, it's fine at that.

But there's a quieter use that's arguably more valuable, and almost nobody talks about it. It's not the first draft. It's the last one. The moment right before you send the thing that matters.

Think about the messages that actually carry weight in your week. The proposal that decides whether you win the client. The reply to the customer who's upset. The email to your team that sets the tone for a whole project. The quote with a number on it you can't take back. You already know how to write those. The problem isn't writing them. The problem is that you're too close to them to see what's wrong.

That's the job AI is quietly great at. Not writing for you. Reading for you.

Why you can't proofread your own stuff

There's a reason you catch the typo the second after you hit send, never the second before. When you write something, your brain fills in what you meant instead of what's actually on the screen. You read the intention, not the words. Everyone does this. It's not carelessness, it's just how brains work with their own writing.

The same blindness applies to more than typos. You can't easily see:

• The sentence that reads as rude when you meant it as direct.
• The obvious question your reader will have that you never answered.
• The ask that's buried in paragraph four when it should be the first line.
• The assumption you made that the other person doesn't share.

A colleague could catch these. But you don't always have a colleague handy at 9pm when you're finishing the proposal. AI is the colleague who's always awake, never annoyed you asked, and has no ego in the outcome.

The move: ask it to push back, not polish

Here's the mistake people make. They paste in their draft and say "make this better." That's the wrong instruction. "Better" gets you a blander, more corporate version of what you wrote, usually stuffed with the exact buzzwords you'd never say out loud.

You don't want it to rewrite. You want it to react. Ask it to be a critic, not an editor. The prompt changes depending on what you're sending:

See the difference? You're not asking AI to write. You're asking it to stand in your reader's shoes and tell you what they'll see that you can't. It'll point to specific lines. Then you decide what to change. You stay the author. It's just the test audience.

A quick, honest example

Say you've written a short note to a client whose project is running late. Your draft says:

Feels fine to you. You were being efficient. Run it through the "read this as the client having a bad day" prompt, and you might get back something like:

• It says the timeline shifted but not why, which can read as an excuse or a shrug.
• There's no apology or acknowledgment that a delay affects them.
• "Let me know if you have questions" puts the work on Sam. It doesn't offer a next step or reassurance.

None of that is you being a bad communicator. It's just the stuff that's invisible from the inside. Now you can decide: maybe you add one line of context, one of ownership, and a clear "here's what happens next." Two minutes. Same message, completely different reception.

That's the whole game. Small fixes, caught before they cost you.

Where the payoff is biggest

You don't need to do this for every email. That would be exhausting and unnecessary. Save it for the handful where the stakes are real, and they tend to fall into four buckets.

Sales and proposals. A second set of eyes on the thing standing between you and a signed deal is worth ninety seconds every single time. Ask what would make the buyer hesitate.

Hard conversations. Layoffs, price increases, saying no, delivering bad news. These are exactly the messages where tone matters most and where you're least able to judge your own tone.

Anything with a number in it. Quotes, invoices, scope. Ask AI to check whether the terms are clear and whether anything could be misread later. It won't catch a wrong price, but it'll flag a vague one.

Public-facing writing. A post, a listing, an announcement. Ask "what's the least charitable way someone could read this?" before the internet does it for you.

The catch, because there's always one

Let's stay honest, because this isn't a magic wand.

AI is a simulated reader, not your actual reader. It's a good stand-in for the average person's reaction, but it doesn't know Sam personally, doesn't know your history with that account, doesn't know the inside joke that makes a line land. Use its reactions as a prompt to think, not as a verdict.

It can also be a bit of a people-pleaser. If you ask "is this good?" it'll usually say yes. That's why the prompts above force it into a specific critical role. "Tell me what's confusing" gets you real feedback. "Is this good?" gets you a pat on the head.

And it will sometimes flag things that aren't problems, or miss your intended tone entirely. That's fine. You're not obligated to accept a single note. It's a test reader, not a boss. The value isn't in doing what it says. The value is in getting a fresh read that jolts you out of your own head for thirty seconds before you commit.

Try it on the next thing that matters

You don't have to change how you work. Keep writing your own emails and proposals in your own voice. Just add one step at the end, on the messages that count.

Before you hit send on the next important one, paste it in and ask AI to read it as the person receiving it. Ask what's confusing, what's missing, what could be misread. Then take the two or three notes that ring true and ignore the rest.

It's the cheapest insurance policy in your business. The email you can't unsend is a lot more expensive than the thirty seconds it takes to have someone check it first.

If you're trying to figure out where AI actually fits into your business or daily workflow, that's exactly what we help people do at Humanity AI.