Is there a free idea validation tool here?

Every new account gets credits for one complete run, no card required. There is no free tier after that. A run spends real money on search and model calls. A tool that pretends otherwise pays for it by searching less.

How is this different from asking ChatGPT?

A general model answers from what it read during training. It cannot go and look. It also agrees with the framing you hand it, so the same idea described two ways gets two verdicts. The difference here is search and sources, not intelligence.

Can it check an idea with no rivals?

It can tell you whether people discuss the problem, which is the more useful half. An empty field means one of two things. Nobody has built it, or people tried and the threads are full of dead product names. Read the threads to tell which.

What happens if it finds nothing?

You get a report that says so, with the searches it ran. That costs the same as any other run, and it is often the most valuable one. No market need is the most cited reason startups fail. It is cheaper to learn that now.

Does it work outside the US?

The threads it reads are mostly in English and skew North American. A market that discusses its problems in another language will come back thin. The business records behind the money check are anonymised US records.

Can I see the sources it didn't use?

Every quote shows the thread behind it. Findings we could not trace to a page are shown, marked as unverified, and ranked lower. The report does not yet list searches that found nothing usable. That is a fair thing to want and is on the list.

AI Idea Validation Tool That Cites Every Claim

ProofMachine is an AI idea validation tool. It starts with one question: has anyone said, in public, that they have this problem? It reads the forums your buyers post in. It quotes them word for word, with a link to each thread. Then it gives a verdict that is allowed to be no.

The useful part of a validation report is the part that disagrees with you.

Most tools in this category can’t produce it. You paste in an idea. A language model thinks about it. A score comes back, somewhere above seven. That number sums up how the idea reads, not whether anyone wants it. It is the answer a kind friend would give, only faster. The same thing happens when you ask ChatGPT directly. We compared six of these tools. Five of the six work this way.

This page describes the other approach, and where it runs out.

What a run does

You describe an idea in a sentence or two. The run then looks for people who have already complained about the problem it solves.

That search is the whole product. It reads public threads: subreddits, review sites, niche forums. That is where angry people have already typed three paragraphs at midnight. It pulls the passages where a real person describes the pain, and keeps the URL of each thread.

A run takes about four minutes. It happens on our side, so you can close the tab.

What comes back

Five checks. Each is answered from what the run found, not from a guess. Every point sits beside the count it came from.

The problem is real. People describe it without being asked, in their own words. Here are the threads. We count how many different pages the complaints came from.
Businesses like this make money. How many similar businesses make more than $5,000 a month.
They’d pay to fix it. How many real people named a price about their own money. This is the check that fails most often.
You can find these people. Where they gather, and how many places.
The field is wide open. Or it isn’t, and here are the rivals already doing it. Fewer rivals, more points.

The points make a score from 0 to 100, printed beside the verdict. The score sets the word: Strong, Mixed or Not yet. If a needed input is missing, the verdict is Not enough evidence. The word and the number can never disagree. A Proof Kit runs all five checks. A Validation Kit runs checks 1, 3 and 5, and scales them to 100.

This page warned you about scores from tools. So here is what ours is made of: counts from the run. No point in it is a model’s opinion of your idea. A check the run never measured scores zero and says how to earn it. So most ideas cannot score 100 until a buyer names a price.

The money check reads from a set of about 220,000 anonymised business records. Not from a model’s guess. The records carry no company name, website or contact details. They exist to give a rate, and a rate is what you get back. Only a Proof run gathers these. A Kit run scores 0 on this check and says so. Rates are dull, and they end arguments. If four in ten businesses in a category make real money, that is a different world from one in forty. No amount of thinking about your idea moves that number.

Findings we could not trace to a page are still shown. They are marked as unverified and rank lower.

Below the checks sits The Call: Go, Maybe or No. A No is earned only on checks the run measured. If the run never checked prices, rivals or earnings, The Call says so instead.

The report ends with a panel called The way up. Up to five suggested changes to your brief. Each one rewrites one line you typed: what it does, who it’s for, the problem, or how it makes money. Each names the check it would help and carries the quote that prompted it. None has been tested yet. One button re-runs the idea with the changes you pick. That is a new run, and it costs credits.

Where the quotes come from

Public writing by people who did not know they were being read for market research. Nobody performs for a survey they don’t know they’re taking.

Every quote in a report carries a link. Click it and you land on the thread. If a quote looks too convenient, check it. If the thread doesn’t say what the report says, that is a bug we want to hear about.

One fair objection: forum complaints come mostly from people who are angry and online. People who quietly put up with a problem don’t post about it. Neither do people who already pay someone to solve it. A complaint proves a problem exists. It does not prove a market does. The validation framework covers the steps that come after this one. The honest ones involve asking someone for money.

What it won’t tell you

It won’t tell you whether you should build it. That comes down to your appetite and your runway. No tool can see either.

It won’t find evidence in a market that doesn’t write things down. Some buyers, like plumbers, builders and big-company buyers, do almost all their business offline. A run against that kind of idea comes back thin. Thin is the honest answer, not a padded one.

It won’t stop you from building the wrong thing carefully. No market need is the most cited cause of startup failure. Good engineering does not cure it.

It won’t replace talking to people. It shortens the list of who to talk to, and gives you their words to open with. That saves weeks. It is not the conversation.

And it won’t reward you for a good pitch. A better-written idea changes nothing about what strangers posted last year.

Who should skip it

If you have shipped and have users, you have better data than this. Read your own support tickets.

If your idea depends on a market shift that hasn’t happened yet, there is nothing to find. No one complains about a problem nobody has yet. That silence is not evidence, and a run will say so.

If you want a document to show an investor, buy a report generator instead. They are cheaper, faster and prettier, and several are free. A verdict that says no is worth something to you and nothing to a pitch deck.

How to run one

Sign in, describe the idea, and come back in about four minutes. New accounts get one Evidence Pass free: the complaints, each linked to where it was said. That is enough to see whether the pain is real before you pay for the full report. After that, runs cost credits. The pricing page has the current numbers.

Bring the fuzziest version of the idea. Precision at this stage is mostly invented. The search works better on a plain description of the problem than on a sales pitch.