Most websites show everyone the same thing and leave you to search through menus. Our website tries to think along: it asks questions, writes on from your answers and suggests fitting next steps. Here is how that works - and where we deliberately draw a line. The few tips you come across on the website are taken word for word from this article.
The assistant
Top centre - at the bottom edge on a smartphone - sits our AI assistant. Ask it whatever you want to know; it answers only from our own content and shows the matching pages.
In the panel everything is in the order of the conversation: your question at the top, the answer below, then the matching content, the input at the very bottom. Each suggestion says in a few words how it relates to your question. Tap one and the website shows the page, while your question stays at the top - so you always know where you came from.
No conversation gets lost: you find your history and can ask on from any earlier question.
Questions instead of forms
Wherever you can answer - a claim, a question, a slider - the page adapts: your answers decide what comes next and which content we consider more relevant for you.
On the home page it starts with a claim that rather speaks against providers like us. Agree, disagree or say “it depends” - from there the website writes on section by section, with sources in the margin. You never have to name your role; which claim speaks to you says enough.
At the end of an article or a reference, “Your move” asks where to go next. Your choice leads to new, fitting content - and feeds into the picture the website has of your visit.
Articles and references as a magazine
Articles and references appear as a magazine edition: the same reviewed content, prepared by AI for reading on screen - with a lead, quotes, graphics and sources. The approved original is always one click away. Each edition is created once and then delivered the same to everyone; it only becomes personal at “Your move”.
Discover more
At the bottom, the AI suggests where to go next - and tells you when reading is worthwhile instead of retelling the title.
Each suggestion says how it relates to what you are reading: it goes deeper, branches off in another direction or is the next step. With “go deeper”, “next step” and “branch off” you decide which direction the next suggestions take. If the AI does not answer, we show no substitute list, only the way to the overview.
Your focus: the picture we form
While you read, the website forms a picture in the background of what matters to you: rather time, quality or cost. It is based on the pages you read and for how long, the claim you answered, and where you went next - deeper, one step further or in a completely different direction.
We do not hide this picture. Under “Your focus” in the menu you see all of it: how we read you and why, your path as a short story with its highlights, and a brief perspective on your visit. One click on “Forget everything” deletes it.
We show this on purpose. A website that pretends to know nothing about you is rarely honest - one that shows what it assumes lets you decide for yourself.
Why our AI takes a brief moment
Our AI never answers faster than the blink of an eye - on purpose: an answer that comes instantly looks prefabricated.
A short, visible pause, by contrast, makes it credible that something is really being created for you. Research calls this the “labor illusion”: work you can see is valued more highly (Buell & Norton, Management Science 2011). For chatbots, a study found that a dynamic response delay leads to higher satisfaction than an instant one (Gnewuch et al., ECIS 2018).
We say this openly because a pause you keep quiet about would be a trick. If it is too long for you, switch it off on the Your focus page.
Two kinds of AI
Two kinds of models work behind the website. Decision models choose from a fixed list and give a probability for each option - for example which content fits now or how much a text is about time, quality or cost. They are fast, inexpensive and cannot invent anything, because they only choose.
Language models write text: the growing page, the magazine editions, the sentence on when a suggestion is worthwhile. With us they only get our own content and checked studies as sources. Freely written sections are checked by a decision model before they are shown.
If no AI answers, we show nothing. A fixed rule pretending to be AI does not exist here.
Built to be data-sparing
- No cookies, no identifier, no account. We do not recognise you on your next visit.
- Your path, your answers and the picture of them stay only in your browser’s session storage - deleted with “Forget everything” or when you close the tab.
- Your browser remembers three settings permanently, but only if you change them yourself: “show instantly”, “No more tips” and “Think along”. With “Think along” off, the website does not remember your path at all. You reset all three under Your focus.
- Only what is needed for the one answer goes to the models: titles of the pages you read, your answers, your focus. All models run in the EU, without storage at the provider and without training - more in Where this website runs.
- We count anonymously how often pages and features are used - without an identifier and without a history per person. Details are in the privacy policy.
Technology in brief
The pages themselves are generated statically and are fast; the AI building blocks are small web components that only switch on where they are needed. The assistant builds its answers from checked building blocks instead of generating arbitrary HTML, and may only link to pages of this website. How we limit abuse and costs is described in Public AI without a cost trap.
Our thinking behind it: a website that thinks along has to be honest above all - about what it assumes, what it knows and what it deliberately forgets.
