About Intent Amplifier
Why we built Intent Amplifier
We are a small team with one stubborn habit: we measure our own ideas honestly, and we say so in public when they lose.
Intent Amplifier is a small button that turns a rough prompt into the one you meant to write, right inside the AI chat sites you already use. This page is the story of why it exists, one large thing we got wrong, and how you can be part of what comes next.
Where it started
It started with a feeling anyone who uses AI knows. You open Claude, ChatGPT, or Manus, you type something rough like "help me write a sales page," and what comes back is fine. Technically correct, a little hollow, missing the one thing that would have made it land. So you follow up. And follow up again. Ten messages later you have talked the model most of the way to what a good prompt would have gotten you on the first try.
We were tired of that loop. Not because the models were weak, they were getting better every month, but because the burden was always on us to know how to ask. The people who got great results were not using a secret model. They were bringing better judgment to the same empty box: the instincts of a marketer, a negotiator, a teacher, an engineer. The model was never the missing piece. The way of asking was.
So we asked a different question. What if that judgment could ride along with your prompt, quietly, so you did not have to already have it yourself?
The bet we made, and what happened to it
Most attempts to make AI smarter reach for one of two levers: a bigger model, or your private data to train on. We reached for a third. The best experts in the world have already published how they think, so we read them, organised what they said, and slipped the relevant piece into your prompt.
We built the whole thing. Fifty nine experts, 1,224 individual pieces of their thinking, sorted and scored so the right one could be found in a fraction of a second. Then, before relying on it, we measured whether it actually made anyone's answer better.
Across 108 blind comparisons, prompts carrying an expert's knowledge were preferred 28.7% of the time. Prompts carrying a complete stranger's personal background scored 26.2%. Your own context, about you, scored 93.8%.
It was not a bad library. It was material about somebody else, and material that is not true of the person asking makes an answer worse no matter how good it is. So we switched it off. It is still there, still reading your request to work out which specialism it belongs to, and none of it reaches your prompt any more.
That was the hardest month of the project, and it is also the part we are proudest of, because we ran the test that could kill our own best idea and then honoured what it said. All of it is published here, including the parts that do not flatter us.
What we're building toward
What we are building toward now is narrower and, we think, more honest. Every time you ask an AI for something that matters, we want the request that leaves your keyboard to be the one you actually meant: everything you said still in it, and nothing invented about you added to it.
Today that lives in one button on more than twenty AI chat sites. It reads your request back to you before it writes anything, so you can correct it. It asks the one or two questions that would genuinely change the answer. And when your request already looks complete, it says so and offers to send it as it is, because we measured that too, and on those requests we make things worse.
What we believe
A few beliefs shape every decision we make.
- •It should be invisible. The amplifier never lectures, never names a technique, and never shows off. You get a sharper prompt, not a lesson on why it is sharper. It should feel like your own good instincts on a better day.
- •Your words are yours. Nothing you type is trained on, and your own sentences are carried into the finished prompt word for word rather than summarised away. Sharing anything is off by default. You opt in, never out.
- •Honesty over hype. We publish the measurements that went against us, and there are plenty. When your request already looks ready, the product says so and offers to step out of the way rather than manufacturing a reason to be useful.
- •Better together. Almost everything we know about this product came from testing it against real requests and losing. The fastest way to make it better is to tell us where it fell short for you.
Where you come in
The team behind Intent Amplifier is small, and that is on purpose. It means the people who use the product shape it directly.
If the amplifier wrote something that landed, tell us. If it wrote something that missed, tell us that first. An honest miss is a gift; it is the signal that makes the next version better, and it is how nearly every decision on the evidence page got made.