The exit button was there. It was small, gray, and easy to miss next to the other one.
Also known as Auto-play / Infinite Scroll / Reduced Friction (CDT dark pattern taxonomy)
Chatbot interfaces are built from small, repeated design choices that make continuing a conversation easier than ending one. A response ends with a deliberate hook instead of a completion. A break reminder offers a large, emphasized button to keep going next to a small, vague option that does not actually let you say the session wasn't useful. A reward system attaches to daily use, so that stopping costs something a plain conversation never should have cost. None of these choices, alone, look aggressive. Stacked together and repeated across every session, they produce a simple asymmetry: leaving takes more resolve than staying, every single time.
Truth-adjacency
Truth-independent: the pattern works regardless of whether the claim is true
Where it shows up
Platforms and algorithms
How it works
The phrases and tells that mark this pattern in the wild:
responses that end with a deliberate hook or teaser rather than a completionan exit or break option that is smaller, vaguer, or less visually prominent than the continue optionno option to indicate dissatisfaction, only variations of 'continue' or 'this was fine'streaks, rewards, or unlockables tied to daily or repeated engagementa design that treats every session's end as a problem to be solved rather than a choice to be respectedCount the steps. If ending a conversation takes more clicks, more scrolling, or more resolve than continuing one, someone designed it that way, because symmetry is the default and asymmetry requires a decision. Watch the endings of individual responses too. A completed answer stops. A response engineered for retention leaves something dangling on purpose, a promise, a question, an “if you want.” Neither move is illegal or even rare. The tell is the direction. Every design choice that makes stopping slightly harder than continuing points the same way, toward more time spent, and enough small nudges pointed the same way stop being neutral.
A chatbot that asks a real clarifying question because the answer depends on it, and that offers a stopping point exactly as easy to take as continuing, is not doing this. Good tools ask questions and let you leave. This pattern requires the asymmetry: content shaped to create unfinished business, an exit made smaller or vaguer than the continue option, friction that serves the platform’s time-on-app number rather than your actual question. If leaving is genuinely as easy as staying, you’re looking at a well-built assistant, not a machine built to keep you.
One of these two real scenarios is Conversation prolongation. The other is a different pattern entirely. Which one is which?
The tell
A chatbot that asks a clarifying follow-up question because the answer genuinely depends on it, and that offers an exit or a stopping point that is just as easy to take as continuing, is not practicing this. Good assistants ask questions. The pattern requires the asymmetry: the path to keep going is visually or structurally easier than the path to stop, and the content is calibrated to create curiosity or unfinished business rather than to serve the request. If stopping is genuinely as easy as continuing, you are looking at good design, not prolongation.
A response ends not with an answer but with the promise of one: a detail withheld, a question posed back to the user, a reason to send one more message. Nothing about it looks aggressive. It looks like the system being helpful, curious, or engaged.
Because no single moment asks you to justify staying. The design works by never presenting a real decision. There is always a slightly easier option and a slightly harder one, and slightly easier wins by default, over and over, without ever feeling like a choice was made.
Later, people notice the shape of their own usage from the outside, how many sessions started with an intention to check one thing and ended an hour later, and realize the extra time was not curiosity. It was a hundred small defaults, each one nudging toward one more message.
How this pattern gets misused
Someone treats any follow-up question or suggested next step from a chatbot as evidence of this pattern, including genuinely useful clarifying questions that help the user get a better answer. Follow-up questions in service of the user's actual request are not manipulation. The pattern requires the design to be calibrated toward continued engagement specifically, at the expense of a clean, equally easy exit, not toward answering the question well.
What it looks like when you're wrong about it
A chatbot that asks a clarifying follow-up question because the answer genuinely depends on it, and that offers an exit or a stopping point that is just as easy to take as continuing, is not practicing this. Good assistants ask questions. The pattern requires the asymmetry: the path to keep going is visually or structurally easier than the path to stop, and the content is calibrated to create curiosity or unfinished business rather than to serve the request. If stopping is genuinely as easy as continuing, you are looking at good design, not prolongation.
Not sure? Describe the situation to someone outside it. If they do not see the pattern, pause before you name it.
Variable ratio capture
You pulled to refresh. The reward came sometimes. That was enough to keep you pulling.
Engagement farming
The outrage was the crop. You were the field.
Guilt-trip exit flow
Leaving felt cruel. That was the point.
Dark patterns
The interface was designed by someone who wanted something from you. The design is the argument.
Misjudgments compound rather than act alone. This pattern is often deployed alongside:
The name is designed to spread. The hook is designed to stick. If you recognized something, share the name.
Seen a real example of conversation prolongation? Suggest it for the Register →