On the house

Who's who in AI, for normal people

support@thecontentupgrade.com (Mark Tandan) — Thu, 24 Sep 2026 09:00:00 +0000

If you've ever nodded along while somebody said "just use Claude for that" and had no idea what a Claude was, this one's for you.

Six names. Who they are, what they sell you, and the one thing I'd want a friend to know. The facts were checked against the companies' own pages in September 2026. The opinions are mine, and I'll flag them so you can ignore them.

OpenAI, which makes ChatGPT

Who: the company that kicked this whole thing off for normal people, at the end of 2022. What they sell: ChatGPT. There's a free version, and the paid one most people mean, Plus, is $20 a month. One thing to know: when a friend says "AI," nine times out of ten they mean this. It's the default, the way "Google it" means search. Opinion: it's a very good generalist, and it's also the one most likely to hand you the smooth, samey draft, precisely because so many people are pulling from it. Nothing wrong with the tool. Everything to do with how it's used.

Anthropic, which makes Claude

Who: in their own words, an AI safety and research company. What they sell: Claude. Free version, and Pro is $20 a month billed monthly. One thing to know: a lot of people who write for a living quietly prefer it for drafting. Opinion, and I'll own it: I build my own stuff with it. It still pulls to the middle if you let it. They all do.

Google, which makes Gemini

Who: Google. What they sell: Gemini, free, with a paid tier called Google AI Pro at $19.99 a month. One thing to know: it's already inside things you use, Gmail, Docs, the search box, whether you asked for it or not. Opinion: if you live in Google's world it's the path of least resistance, and the free tier is more capable than most people realise. I run a lot of my own tests on the free one, on purpose.

Microsoft, which makes Copilot

Who: Microsoft. What they sell: Copilot, an assistant built into Windows, Edge and the Office apps, Word, Excel, Outlook and the rest. There's a free version on the web and in the app, and more of it comes with a Microsoft 365 subscription. One thing to know: if your workplace runs on Microsoft, this is the one your IT department will let you use, and that's not a small thing. Opinion: it's the AI you have, rather than the AI you chose. That's fine. The instruction you give it matters more than which one it is.

Meta, which makes Meta AI

Who: Facebook's parent company. What they sell: nothing, in the sense that Meta AI is free. Their own words: "your free personal AI chatbot and assistant." It lives in the Meta apps, on the web, in its own app, and in their glasses. One thing to know: the price is zero for the same reason Instagram is zero. That's not a scandal, it's the deal you already signed up for. Opinion: fine for a quick question in a chat. I wouldn't write anything that mattered in it, and I say that as a guy who won't run ads on their platform either. Bias declared.

Perplexity

Who: a smaller company, a startup by comparison, and the odd one out on this list. What they sell: their own description is "a free AI-powered answer engine," and the paid version, Pro, runs around $20 a month. One thing to know: it's less "write me a thing" and more "answer this and show me where you got it." It cites its sources. Opinion: for looking something up, it's the one I'd hand a friend who doesn't trust AI, because you can click through and check. For writing in your own voice, not the tool.

A few things across all six

They're all about $20 a month at the paid tier, and they all have a free version that's better than most people think. If you're paying for two of them, ask yourself why. If you're paying for one and never touch the paid features, same question.

Every one of them, free or paid, pulls toward the average unless you tell it not to. That's not a knock on any of them. It's how the whole category is built. The fix is the same on all six: tell it how you talk, up front, before you ask for anything. I wrote about how to do that in the voice doc post, and it's free.

And none of them is going anywhere. Pick one, learn it a bit, and stop reading comparison articles. Including this one.

Which one do you actually use? Not which one you pay for. The one you open. Start there.

Six names, one habit: they all pull to the middle until you tell them not to.

That's the free drink for today. If you'd like the next one, the form below is the only thing here that wants your email.

Rule 5. Harden the F up.

support@thecontentupgrade.com (Mark Tandan) — Wed, 23 Sep 2026 09:00:00 +0000

There's a page on our site with seven rules on it instead of an FAQ. Six of them are about the app. Rule 5 is me.

Rule 5: Harden the F up.

It isn't mine. It's borrowed, with credit, from the Velominati and their book, The Rules: The Way of the Cycling Disciple. Their Rule 5 is three words long and it's the one every cyclist knows, whether they've read the book or not. Phil Gaimon knows it. Anybody who's ever been dropped on a Sunday club ride and had to grind home alone knows it. It isn't cruel. It's a reminder that the hill doesn't care how you feel about it, so you might as well keep pedalling.

On our page it covers the founder, because the founder swears like a trucker. Often completely unprovoked. I've been that way since my first job behind a bar, and every attempt to fix it has been, let's say, partially successful. Ask anyone who's built software with me. There are screenshots. They're not flattering.

So that's the earmuffs warning, and it's real. But I want to talk about the other half of Rule 5, the half that's actually useful to you, because it's why the app talks to you the way it does.

The app is blunt

Not rude. Blunt. When you paste something in, it doesn't tell you it's great. It tells you what it cut and why. When you hand the Email Negotiator a rant about your cable company, it takes the rant apart, keeps the facts, throws out the rest, and says so, on the screen. Cut, because a support desk stops reading here. That kind of thing.

The first time, that stings a bit. I know, because I built it and it stung me.

Why I think that's kinder than the alternative

The alternative is the tool you've already got. The one that says "Great question." The one that tells you your draft is compelling and then hands you 400 words of compelling nothing. The one that would happily let you send the angry email, the padded ask, the listing with "stunning" in it three times, and never once mention that none of it is going to get read.

That isn't kindness. That's a machine that wants you to keep chatting.

Real kindness, the kind I got from the good people in my life, sounds more like: that's not going to work, here's why, here's what would. Blunt, then useful. A friend across a table, not a customer service script.

My best clients hired me for that. Not the writing. Plenty of people write. They hired me because I'd say "yeah, that's good in theory, but it's not going to work, and here's why," and then stay to fix it. It cost me a few jobs over the years. It kept the ones worth keeping.

So the app does what I do. It's blunt about your draft so you don't have to find out the hard way from the person you sent it to. And it's polite. The software never swears, I checked. That part's on me, and me only.

A caveat, because I hate absolute rules even when they're mine. Blunt is for the work. Never for the person. If anything I build ever makes you feel small, that's a bug. Write to support, I'll fix it and I'll apologise. The hill doesn't care how you feel. I do. Those are different things.

Does it bother you, a tool that pushes back? Some people love it. Some people close the tab. Both are fine, and you'll know within a minute which one you are.

Rule 5. Three words, and kinder than it sounds.

Want to feel it? The Email Negotiator is at the top of the blog. Paste in the angriest email you never sent and watch what it keeps.

What a voice doc is, and how to write a rough one in 20 minutes

support@thecontentupgrade.com (Mark Tandan) — Mon, 21 Sep 2026 09:00:00 +0000

A voice document is a page that tells a machine how you talk.

That's all. One page, maybe two. You write it once, you paste it at the top of any AI chat before you ask for anything, and every draft that comes back starts from you instead of from the average of the internet.

I think everybody should have one. Not everybody who buys my software. Everybody. If you use ChatGPT for work emails twice a week, you should have one. It's the best return on twenty minutes I know of in this whole AI business, and I'm about to give you the method, the real one, because keeping it back would make me the guy I complain about.

Five sections. Rough is fine. Rough is better, actually, because a tidied-up voice doc is worthless. The machine needs your fingerprints, not your CV.

Section 1: who I am and who I'm talking to

Two lines. What you do. Then one person, singular, who reads what you write. Not "my customers." One of them, with a name in your head. If you write for a group, pick the one person in it who matters most and write to them.

Section 2: the dials

Answer these fast, first instinct. Casual or formal? Contractions, always or never? Humour, and what kind? Self-deprecating, dry, none? How blunt are you, honestly? Would you tell a customer they're wrong? Exclamation points, yes or no? Mine's never. They read desperate.

Section 3: the ramble

This is the part that does the work. Open the voice memo app on your phone. Talk for five minutes about what you do and what people get wrong about it. Don't plan it. Then talk for two more minutes about a time you had to give somebody bad news. Then one more minute on something you love that has nothing to do with work. Transcribe it. Your phone does this for free now. Paste the transcript in raw. Ums, half sentences, the lot. Do not clean it up. I'll say it again because everyone does it: do not clean it up. The mess is the voice.

Section 4: the never list

Words and phrases you'd never say. Mine has "leverage" on it, and "it's not just X, it's Y," and any sentence with three adjectives in a row. Add "no em dashes" if you want to be able to tell your own writing from the machine's. Then add the guardrails. Anything you're not allowed to claim. Anything you'd be embarrassed to see under your name.

Section 5: go-to phrases

Five or six things you actually say. Not what you wish you said. Ask a friend if you're stuck, they know them better than you do.

Then put this at the top, word for word if you like:

"This is my voice document. Follow it for everything you write in this conversation. Where it conflicts with your default style, this document wins. Before you give me any draft, check it against the never list and fix violations first."

Done. Twenty minutes, and you might never write a two-hour prompt again.

A few things I've learned from building a lot of these

The ramble does more than every typed answer put together. If you only do one section, do that one.

Include the range. Somewhere in there, say the same thing two ways: how you'd pitch it to an old friend you wanted to impress, and how you'd explain it to your grandparents. That gives the machine a top and a floor, and it'll stop making you sound bigger than you are.

Keep it somewhere you own. A text file. Not inside one chat window, not inside one app. It's yours, it should be portable, and you should be able to walk into any AI on earth and paste it.

And revisit it in a month. You'll cringe at one line and you'll be right.

Is this what my software does? Yes, a more thorough version, by asking you the questions out loud, with some things I've learned over the years about what actually gets read layered on top. But the rough one above is honestly most of the value, and it costs you nothing but twenty minutes and a bit of nerve talking to your phone.

So, will you do it? Today, not someday?

Your voice, written down, is the one AI asset nobody can sell you. Go make it.

If you build one and it surprises you, I'd like to hear about it. The form below is how you'd hear from me. The other direction is support@thecontentupgrade.com.

The bartender rule

support@thecontentupgrade.com (Mark Tandan) — Sun, 20 Sep 2026 09:00:00 +0000

My first sales job was behind a bar.

I was pretty good at it, wherever I worked, and not because I could flip a bottle. I couldn't. No hand-eye coordination at all. Still don't.

What I could do, I learned from my first real bar manager, a friend of my dad's. Almost 40 years ago now. He gave me one speech and I've been living off it since.

Your only job is to keep them here.

That was the speech. Not sell more drinks. Not be charming. Keep them here. If the place looks busy, more people come in, and more people stay. And they're not really here for you anyway. What they come back for is each other.

So John walks in after work. Alone, beat up, ten minutes to himself before he goes home to whatever's waiting for him there. You put a drink in front of John. Then you introduce him to the couple beside him. "Bob, tell John that thing you were telling me." And you walk away.

John leaves feeling a little better about his day. John comes back.

That's the job, all of it, and it's the only thing I've done for 30 years. In bars, in front of rooms, on the phone, and eventually in writing for other people. Get the room talking to each other. Face to face it's easy. Online it's harder. Same job.

Which explains what this blog is

It's a free drink.

You walked in. Maybe somebody sent you, maybe you were fighting with ChatGPT at midnight and typed something rude into Google. Either way, you're here, and the first thing I do is put something in front of you that costs you nothing. A tool that argues your bills down. A few things I've learned the hard way about getting a machine to sound like a person. The odd funny thing, because I've been called an entertainer since my first serving job and it's a hard habit to kick.

And then, if I'm doing my job right, I introduce you to the person beside you and walk away.

That's the part I'm still working out online. In a bar it's easy. "Bob, tell John." Here it's slower. But the idea's the same: I'd rather you talk to each other than listen to me. If something on this blog helps you, tell the person who'd get the same use out of it. That's you talking to Bob. I don't need to be in the conversation.

So what do I want from you?

Nothing. Yet.

I mean that, and I know how it sounds, because everyone on the internet says it right before the pop-up. So let me say it more precisely. There's nothing on this page to buy. There's a form at the bottom if you'd like the next post, and that's optional, and you can leave whenever you want. That's the full list of asks.

Somewhere down the line I'll have something to sell you, and I'll say so plainly on that day, in a place you can't miss. It'll be the same deal as the bar. Buy or don't, you're welcome to stay either way.

Until then, the drink's on me.

What did that bar manager know that the internet forgot? That people come back for each other. Not for the guy pouring.

Your only job is to keep them here. Everything else follows.

The free tool's at the top of the blog. That's the drink. Everything else is up to you.

You're an island. This just brings it out.

support@thecontentupgrade.com (Mark Tandan) — Sat, 19 Sep 2026 09:00:00 +0000

How many people on earth sound exactly like you?

Same words, same jokes, same way of getting to the point or wandering around it. Even if you're a Swiftie and you've got a lot of company. What, 2? 3?

Now count the tens of thousands of people running the same AI off the same template packs. Templates built, on purpose, to make everything sound the same. Same openers. Same "I hope this finds you well." Same three bullet points with a bolded word at the front of each.

You're an island. Every one of us is. The problem was never that you don't have a voice. It's that the moment you hand your idea to a machine, the machine starts sanding it.

Nothing adds a voice to you

I want to be careful here, because there's a version of this that sounds like a pitch and I don't want to give you that one.

Not my software, not anyone's, not a course, not a coach. If someone tells you they'll give you a voice, walk. You had one before you could spell. It's the one your friends recognise in the second sentence of a text. It's the reason your mum can tell which kid wrote the birthday card.

So the job, the only job, is stopping the sanding.

Here's how the sanding works, in plain English. You write, or say, something that's very you. Rough, maybe. Then you ask an AI to "clean it up." And it does, in the sense that it replaces every word that was specifically yours with the word that's most common. Your "yeah, no, that's a hard pass" becomes "unfortunately, I'll have to decline." Technically better. Nobody would ever guess it was you. The rough bits were the fingerprints.

Now flip it. Give the machine one instruction up front, before it starts. Something like: "This is how I talk. Keep it. Fix spelling, fix the one sentence that runs on, leave everything else." Then paste your rough thing in and get out fast.

What you get back isn't a new voice. It's yours, with the dents left in. In my opinion that's the only thing worth paying for in this whole category, and it's also the one thing you can do for free with a decent instruction and a bit of discipline.

Why it matters, and neither reason is about marketing

One, people buy you before they buy whatever you're selling. Always have. That's close to forty years of face-to-face selling talking, not a theory. The listing with a little of you in it gets the call. The email with a little of you in it gets answered.

Two, it feels better. I didn't expect that one. When my own drafts started coming back sounding like me on a good day, something in the back of my head went quiet. The imposter part. I'm 56 and it still surprises me that reading my own words, done well, does that.

What would it look like if the next thing you sent sounded like you? Not the polished you. The actual one, the one people already like.

You already own the winning piece. You've owned it the whole time.

You're an island. Everything I build, and everything on this blog, just brings that out.

Want to see the sanding stop on something real? The free tool at the top of the blog does exactly that to an angry email. Try it on yours.

Why we're in a hurry

support@thecontentupgrade.com (Mark Tandan) — Fri, 18 Sep 2026 09:00:00 +0000

Anything this different gets copied. That was never up for debate.

What I can do is tell you what I actually know, because I'd rather you hear it from me than work it out later and wonder what else I skipped.

We ran a scan. July 12 to September 14 this year, 64 days, about 30 searches plus a direct read of product pages, GitHub, the Product Hunt leaderboards and the vendors' own release notes. If you want the source list, write to support and it's yours, it's a real list. And treat the count as a floor. Nobody reads every feed.

What surfaced in 64 days

Five new products or features that touch what we do. Five in nine weeks. Roughly fifteen credible products in total now touch some part of the job.

How many do the whole job? Meaning: learn your voice by talking with you, hand you a document you own and can take anywhere, run on the AI you already have, carry twelve years of judgment about what actually gets read, and charge once.

Zero. I checked twice, because I wanted the answer to be zero and that's exactly when you should check twice.

The market has split into two halves that haven't met. One half interviews you and then keeps what it learned inside their subscription. The other half hands you a document you own but builds it from writing samples and has no opinion about whether the writing works. Neither half is in any hurry to become the other, because their business model is the reason they're the half they are.

So why the hurry on my end?

One of the five. On September 7 OpenAI shipped a feature called Writing Styles. It reads your connected Gmail, Drive, Slack and SharePoint and learns your phrases and sign-offs on its own. Right now it's on ChatGPT Work plans only, not in the ordinary chat most people use, not on free, and what it learns stays inside ChatGPT by design. That's the one that matters, because it's the only one that could make the problem go away for a lot of people instead of competing on the fix.

Will it? My honest read: it learns the voice you use in your sent folder, which for most people is the voice that says "per my last email." It has no idea what gets a bill reversed or a listing read. And it doesn't let you leave with anything. Those are three real gaps. They're also gaps a very large company could close if it decided to.

Our best guess at the window, and it's a guess with homework behind it, is somewhere between nine and fifteen months. Bounded on one side by that feature reaching the free tier and actually working. Bounded on the other by somebody with a real audience bolting a spoken interview onto a tool that already has the judgment part. There's one company well placed to do that, and conveniently, they just moved their business toward monthly subscriptions, which a one-time product would undercut. That conflict is a better moat than any technology I've got.

What I'm not going to do

Give you a launch date or a promise. I've done that before, too early, and I ate it. Rushing to be ready and rushing to sell are different things, and I got them confused once.

What I am going to do is keep building the part that's hardest to copy. The tech, sure, somebody will get there. Twelve years of knowing what a person actually says yes to, and the taste to leave the rest out? That isn't on GitHub.

And your voice? That one can't be copied by anyone. Not by me, not by them. It's the only part of this that was never for sale.

Does the hurry make sense now? It's not fear. It's a window, and I can count.

Five copies in nine weeks, zero doing the whole job, and a clock I can hear.

If you'd like to know what happens next, the form below is where I'll say it. No dates, until there are dates.

Done is better than perfect

support@thecontentupgrade.com (Mark Tandan) — Thu, 17 Sep 2026 09:00:00 +0000

Five years ago I wouldn't have let a blog post like this one ship.

I'd have spent two more days on it. Tweaking, tuning, agonizing over a comma in paragraph four that nobody on earth was ever going to notice. I was a perfectionist. Didn't enjoy it. Took me about 50 years to figure that out, which is a long time to learn a five-word lesson.

Done is better than perfect.

It's my favourite line of the last two years, it's on the T-shirt, and it's the closest thing I have to a philosophy. I say it in Spanish too, badly. Hecho es mejor que perfecto. My Spanish is terrible. It's also functional. I can tell a bad joke in it and make a room smile, and I'd never have got there if I'd waited until I could conjugate properly.

Where it came from is nothing fancy. I built software to write faster and the software started handing me drafts that were good enough, fast. I could keep polishing, or I could go for a bike ride. I went for the ride. Since then I winter in Panama. Correlation, not causation, but I'll take it.

Why I'm telling you this on a blog about AI

Because I think perfectionism is exactly how people lose an afternoon to ChatGPT.

Think about the last time it happened to you. You asked for one thing. What came back was 80% there. And instead of taking the 80% and fixing the rest yourself in four minutes, you went back in. "Make it warmer." "Less corporate." "Can you try a version where..." Three hours later you've got eleven drafts, you're annoyed, and you send the first one anyway.

That's not the machine's fault, and it's not yours either. It's the perfectionist's trap with a new coat of paint. The machine will always give you another version. It never gets tired. You do. And the eleventh draft is almost never better than the second. It's just later.

My rule, and don't follow it as a set rule

Two rounds. Not three. You ask, you read, you ask one more time if you have to, and then you take what you've got and finish it by hand. Your hand. Your words on the rough spots. Send it.

Will it be perfect? No. Will it be done, in your voice, in the time it was supposed to take? Most of the time, in my experience, yes. And done and sent beats perfect and still in drafts every single day of the week.

A caveat, because I don't want to be the guy who tells you to ship crap. Some things deserve the extra hour. A contract. Something going to ten thousand people. A note to somebody who's grieving. Take the hour. Perfectionism is a problem when it's the default, not when it's a choice.

Everything else? The Tuesday email, the listing, the post, the note to the team. Two rounds and out.

And if two rounds gets you nowhere? Close the tab and write it yourself. You could always do that. You've been doing it your whole life, and the machine was supposed to save you time, not eat it.

What would you do with your afternoons if you got them back? That's the question I'd ask. I got a bike ride and a second language I speak badly, and I'd take that trade again tomorrow.

Done beat perfect. It usually does.

If that line works for you, there's a small form below and it's the only ask here. If it doesn't, no problem, and thanks for reading this far.

The free tool that argues your bills down

support@thecontentupgrade.com (Mark Tandan) — Wed, 16 Sep 2026 09:00:00 +0000

Ever been billed the wrong amount and got nowhere?

Promo ends, the bill jumps, you call, you wait, a very nice person tells you there's nothing they can do. You write an email at 11pm on the fourth try and it reads like a hostage note. Nobody answers it.

I've been that guy. Which is why the first thing on this blog is a free tool and not an opinion.

It's called the Email Negotiator. It's at thecontentupgrade.com/email-negotiator. No account, no sign-in, nothing to buy, and nothing you type leaves your browser. I want to be straight about what it is and isn't, because the internet is full of "free" things with a hook in them.

What it does

You answer six questions. The page turns your answers into a brief. You copy that brief and paste it into whatever AI you already use. ChatGPT, Claude, Gemini, the free one, doesn't matter. Your own AI writes the email. The tool never writes it for you and never sees it.

What comes back is one short, factual email. It names your account, one problem, one specific fix, one date, and one true next step. Everything else you wanted to say gets cut, and it tells you why it got cut. A rant makes a support desk stop reading. A clean ask with a date on it gets forwarded to somebody who can settle it.

What it's for

A bill that went up without a reason. A fee that showed up out of nowhere. A refund that's been "processing" for six weeks. A flight they cancelled and never paid back. An insurance renewal that jumped. A landlord, if you pick "something else" and don't expect it to quote you a law, because it won't invent one.

The tool picks a route based on which line sounds most like your problem, and the route carries the one next step you're actually entitled to, with the date it was checked. Where nothing's verified for your country, it names no agency and no law. It doesn't make one up to scare anybody. A made-up law is how a good email gets binned.

A run-through, generic, so you can see it work

Question 1: what happened, told the way you'd tell a friend. You type, or tap the mic and talk. "My internet was supposed to be a set rate for a year, the bill's been higher for two months, I called twice and got told it's a system thing." Ramble all you want. It reads through the filler to the facts.

Question 2: which line is closest. Internet, cable or cell bill, US. That picks the route.

Question 3: the account number and the name on it. That goes in your subject line and your first sentence, so the email gets routed to a person instead of a queue.

Question 4: the numbers and the dates. What you were promised, what you got charged, when. Only what you can point at. Anything you can't back up stays out, on purpose.

Question 5: what you've already tried. Two calls, the dates, the rep's name if you wrote it down. Haven't tried yet? Say so. You get a short first email to normal support instead of an escalation, and the escalation waits in your pocket, stronger for having a documented first try behind it.

Question 6: what would fix it. Leave it blank and it asks for the smallest fix that covers what you lost, built from your own numbers. Nothing padded, because a padded ask is the fastest way to get dismissed.

Then you hit compile, copy, paste into your AI, and read what comes back against your own records before you send it.

What I can't promise you

What the company does next. Nobody can, and anyone who promises you an outcome on somebody else's billing department is selling something. What I can tell you is the email you send will be hard to dismiss, and in my experience that's usually the whole difference.

Why free? Because I was a bartender before I was anything else, and the first drink is on the house. More on that in another post.

Go argue with your bill. Then come back and tell me how it went.

One calm email to the right desk does more than four angry ones to nobody.

The tool's here: thecontentupgrade.com/email-negotiator. That's the only link on this page that wants anything from you, and all it wants is your bill.

Why every AI draft sounds like the same person

support@thecontentupgrade.com (Mark Tandan) — Tue, 15 Sep 2026 09:00:00 +0000

Ever pasted something into ChatGPT, read what came back, and thought: that's fine, but it isn't me?

Same. For about two years, on and off, that was my whole relationship with the thing.

Here's the deal, and I'm not a computer scientist, so take the plumbing with a grain of salt. An AI is built to hand you the next most likely word. Not the best word. Not your word. The one that shows up most often after the word before it, across everything it ever read.

Which is a lot. More than any person alive has read.

So when you ask it for an email, a listing, a post, a note to the boss, it reaches for the safe version. Competent. Clean. Nothing wrong with it. And nothing you'd be proud to put your name on either.

That's why every AI draft sounds like the same person. Because in a way, it is the same person. An average of everybody. And an average of everybody sounds like nobody.

The part that took me longest to accept

Prompting it harder makes it worse.

You know the routine. "Make it sound more like me." "Less formal." "No, warmer." "Okay, now shorter." Twenty minutes in you've got seven versions and none of them are you. The longer that thread runs, the harder the thing pulls back to the middle. Every correction you type is one more jar on the shelf, and the kid in the candy store keeps walking out with the same bar every other kid picked.

And that's not me selling something. That's how they're built.

So what's the fix?

Not a better prompt. A better question, asked before you type anything.

What would you actually say out loud?

Picture the person the email is for. Your landlord. A client. The guy who runs the Tuesday league. Now say the thing to them, across a table, coffee in your hand. Out loud, if nobody's around to judge you. Then type that. Word for word, ums included if you want.

Give the AI that, and give it one job. "Tighten this up. Don't change how I talk. Fix the spelling." In and out.

What you're doing there is handing it one jar instead of a whole store. Same brain, one direction. It stops choosing and starts working.

And read what comes back out loud too. The ear catches what the eye forgives. If a line clatters when you say it, it wasn't yours, and you'll hear it in a second. I've edited by ear for paying clients for 12-plus years, long before any of this. Turns out it works on the machine as well as it works on people.

Try it once on the next thing you have to write. I'd bet the draft comes back closer to you than anything you've had off it in a year. Not perfect. Closer. And closer is a lot, when the alternative is sounding like a press release.

A couple of caveats, because I hate saying never. For a two-line reply, the AI's fine as it is, leave it alone. And if you've got nothing to say, this won't give you something. It never will. You bring the idea. That part stays yours, and honestly, that's the good news. The idea was never the problem. The sanding was.

Does this even make sense for you? If your AI drafts already sound like you, great, skip the rest of this blog and go enjoy your afternoon.

If they don't, there's more here, and none of it costs anything.

Your voice was never missing. It was getting sanded off.

If you'd like the next one when it's up, the little form below is the only ask on this page.

Why we're a little concerned about AI, and what we're doing about it

support@thecontentupgrade.com (Mark Tandan) — Mon, 14 Sep 2026 09:00:00 +0000

We use AI all day. We built a product on it. So this isn't the opinion of somebody who's scared of the thing because they don't understand it. It's the opinion of somebody who understands it a little too well.

Here's what bothers us.

The people who built AI have put a number on it going badly for all of us. As of September 2026: the godfather of AI, Geoffrey Hinton, says one in ten within a decade; the CEO of Anthropic says one in four; the largest survey of working AI researchers says one in twenty, and the professional forecasters who are paid to be right say one in 263.

Take whichever number you like. Even the smallest one is a number nobody would accept on a plane.

For balance: Yann LeCun, who shares the Turing Award for the deep learning that started all this, puts it below one in ten thousand. Smart people disagree. That's part of the problem. When the smartest people in the room can't agree on whether the thing is safe, the rest of us are guessing.

What a small company can actually do

We're not a lab. We're a small company with one app. Two things are in our control.

One. The tools we build run on your own AI, keep what you type on your side, and put you, not the machine, in charge of what sounds like you.

Two. Awareness. The more normal people see what this can do, right now, today, the harder it gets for the platforms and the governments to shrug. That's most of what this blog is for.

We're not asking you to panic. We're asking you to pay attention, and to keep your own voice while you do.

Pay attention. Keep your own voice.

The free tool's at the top of the blog. Still on the house.