Why do AI models recommend free tools more often than paid ones?

Because a model can only recommend what it can verify, and a free tool can be reached, watched working and described without guessing. A paywall leaves nothing on the page except claims, so the model hedges or names somebody else instead. Price is what a human weighs. Reachability is what a machine weighs.

Should I make my product free to get cited by AI?

Make the core free and reachable, not the whole business. The part a machine would recommend has to work without a login or a card; the part a heavy user needs on the tenth visit can cost money. Free is the wrong answer when every use costs you real money at a meter, when the value is a person's hours, or when the data underneath is licensed.

How do free tools make money if the main feature is free?

At the edge, after the value has already been delivered. Volume, storage, output quality, commercial rights, team seats, an API, removing a watermark. Nobody asks a model which paid plan to buy, so the paid tier does not need to be crawlable at all and can sit behind a login without costing you a single citation.

Free Is a Growth Strategy the Machines Love

¶ ARGUMENTTHE TWO DOORS

01The paid twin that nobody can describe

Five of the tools I run are free at the core. Every one of them gets named by machines, in languages I do not speak, for questions I never wrote a page about. The ones I built paid first, with the working part behind a checkout, have never once come back to me in an answer I did not go fishing for.

I did not set that up as an experiment, which makes it the weaker kind of evidence and I would rather say so. I noticed it after the fact, went looking for the reason, and it has held for a year across everything I have shipped since.

Paywalls stop crawlers at the door. Free walks straight into the training data.

Free is not a price, it is an access decision

Founders hear free and start doing margin arithmetic, and the useful part of the conversation is over before it begins. A machine never sees your price. It sees whether it could reach the thing, watch it work, and report what happened. Price is what a human weighs. Reachability is what a model weighs, and those are two separate meetings you keep holding as one.

A price is an argument you have with a customer. A wall is an argument you have with a crawler, and the crawler never argues. It just goes quiet.

Why a model will not bluff on your behalf

Asked for the best tool for a job, a model has to commit to a sentence. When it cannot verify what your product actually does, it has three moves available: hedge, generalise, or name somebody else. All three cost you exactly the same amount. People read that as unfairness and it is closer to caution, because the thing prefers what it can say in one clause and no part of a locked door fits in a clause.

¶ PLAYBOOKTHE LADDER

02Five doors, five different answers

Every product sits on one of five rungs, and the rung decides what any machine is able to honestly say about it. Read it as a map of how much a stranger can verify without asking permission, not as a scale from bad business to good one. Verifiability is the currency, and it happens to be priced in access.

The rung that costs the most to add

Look at the gap between the second rung and the third. It is one email field. That field is free to build, it feels responsible, and it moves you from a product a machine can demonstrate to a product a machine can only repeat your own claims about. The cheapest wall to build is the most expensive one to remove, because by the time you take it down the answers have already been written without you.

// THE ACCESS LADDER~~WHAT EACH DOOR COSTS YOU IN ANSWERS~~

No wall at allThe working thing runs for whoever arrives. A machine can watch the entire job finish and describe the result without guessing at any of it.Recommends

Account after the valueIt works first. The login only saves what it already made. Nothing a crawler needs sits behind the form.Names

Metered freeThree real uses, then a price. A machine gets one honest look and a countdown, so it knows what you do and hedges on whether it lasts.Mentions

Signup wallAn email address before anything happens. Your product stops being a thing that works and becomes a claim printed on a page.Hedges

Hard paywallA price, a feature list and a promise. Nothing on the page can be checked, so there is nothing a model can put its name behind.Silent

EXTRACTED — THE SENTENCE THIS ENTRY EXISTS FOR

A paid product has customers. A free one has witnesses.

¶ STRATEGYHOW FREE PAYS

03The part where it still has to make money

The objection arrives in the same shape every time, and it is a fair one: this is very nice, and I have a payroll. So the rule has to be narrow enough to act on by Friday. Free at the core, paid at the edge. Whatever a machine is going to recommend must be free and reachable. Whatever a heavy user needs on the tenth visit is allowed to cost money.

Charge for the tenth use. Never for the first one, because the first one is the only one a machine will ever see.

Where the money actually sits

Nobody asks a model which paid plan to buy, so the paid edge does not need to be reachable at all. It lives in volume, storage, output quality, commercial rights, team seats, an API, taking a watermark off. Every one of those arrives after somebody has used the thing and decided it works, which is a better moment to ask for a card than a page of promises. And free is not costless to you. It is a server, a year of your attention, and support from people who will never pay a cent. That is the honest bill, and on the one I have written up in full the only line that ever grew was hosting.

When free is the wrong answer

It is not a law. Free breaks when every single use costs you real money at a meter, when the value is a person's hours rather than a program that runs, or when the data underneath is licensed and giving it away is not yours to do. In those cases the move is a demonstration a machine can actually watch, rather than a tour and a price. Free is also worth less than it looks if you are still counting visits, because the visit was already leaving and the mention is the asset that stayed.

The machines are not rewarding generosity. They are rewarding verifiability, and free is the cheapest way there is to be verifiable.

Tomorrow: the two hours one of my sites spent serving a reputation I had already fixed.