Do I have to change anything on my website because of decision models?

Nothing needs changing for them specifically. They do not crawl, never appear in your logs and require no file or tag to be declared; what does matter is what survives of your page once a crawler turns it into text.

Can I appear in a decision model's answers?

A decision model writes no answers, so there is nowhere to appear. It returns a choice among options declared in advance, with its probability.

Does Primafonte measure citability for decision models?

The index measures whether a website can be read, understood and cited, and that same foundation decides what text reaches a model that scores. There is no separate check for them, and there will not be one until measured evidence exists.

From reading you to scoring you

«From reading you to scoring you» names the shift by which a website stops being read as prose by a model that writes and starts being evaluated with a typed question by a model that decides.

More than fifty pieces about Jev, the first general-purpose decision model, appeared in seven days, and every one treats it as a tool you use. This page takes the opposite view: what it implies for the page you publish.

What exactly changes?

What changes is who asks and what they ask. A model that writes reads your page and may cite it inside an answer; a decision model receives your text already extracted and answers a closed question — «does this page cover this topic?», «on a scale of 1 to 5, how much?» — with a probability and a confidence.

The result is not a sentence about you, it is a box ticked. Nobody cites you, nobody links to you and nobody sees your name: somebody obtains a number about your page and acts on it, and you never find out because the model never touched your server.

Why now and not before?

Judging pages by machine was always possible; paying for it was not. Jev's input costs $0.042 per million tokens and its output is not charged at all.

An audit published by Boring Tools Kit shows it in figures: 4,816 judgments across 1,204 pages in under 3 minutes, for less than $1. That works out at roughly two ten-thousandths of a dollar per judgment, and what used to be a consultancy project becomes a query that fits inside a nightly cron.

What does not change?

Almost everything people assume changes does not change. Saying so first matters, because the confusion is already circulating and it costs money:

Your robots.txt has no effect: a decision model has no crawler and never appears in your logs.
There is no user agent to allow or block, and no bot verification to request.
There is nothing to declare: no new file, no tag, no particular markup.
It does not cite you, so there is no «appearing in its answers»: it writes no answers.

What are we not going to tell you to do?

Nobody here is going to sell you an optimisation tactic for decision models, because none has been measured. This index is built from checks that have evidence behind them, and on this subject the evidence is one week old with no independent study.

Nor is anybody going to teach you how to move a verdict. Jev's documentation admits that text written to steer the model can change the answer, and that is exactly where measuring a website parts ways with manipulating a judgment: the second works until somebody looks, and by then it is not your website, it is your trap.

The only honest thing that can be stated today is plain common sense and needs no product: a decision model works on the text a crawler pulled out of your page, so the quality of that extraction is what ends up being judged.

How do you know whether this affects you?

Look at what survives of your page once it becomes text. The same work that makes a website legible to a model that writes — making sense without the navigation, headings phrased as questions, claims carrying their figure and their source — is what decides what reaches the input field of a model that scores.

That is the measurable part, and the Citability Index already measures it with 33 checks. A one-week-old category does not need a new methodology: it needs nobody inventing one.

Where this comes from

Checked against the vendor's documentation on 22 September 2026, one week after launch. Where the source is a third party, it says so.

TypeSafe AI — Concepts: State
Input is supplied by the caller and only text is accepted: that is why the model never touches your server.
TypeSafe AI — Models
The price the per-judgment cost comes from: $0.042 per million input tokens, output free.
TypeSafe AI — jev-1.13 known issues
Admits that text written to steer the model can move the answer.
Boring Tools Kit — The cost of an SEO audit in 2026
Third party: 4,816 judgments across 1,204 pages in under 3 minutes for less than $1.
Firecrawl — What is Jev?
The crawler turns the page into markdown, and that text is what gets judged.

Frequently asked questions about From reading you to scoring you

Do I have to change anything on my website because of decision models?: Nothing needs changing for them specifically. They do not crawl, never appear in your logs and require no file or tag to be declared; what does matter is what survives of your page once a crawler turns it into text.
Can I appear in a decision model's answers?: A decision model writes no answers, so there is nowhere to appear. It returns a choice among options declared in advance, with its probability.
Does Primafonte measure citability for decision models?: The index measures whether a website can be read, understood and cited, and that same foundation decides what text reaches a model that scores. There is no separate check for them, and there will not be one until measured evidence exists.