Why this guide exists
Most online courses are built on a simple assumption: if we give people enough good content, they'll learn. So we record more videos, write more lessons, add more modules — and watch completion rates fall and conversations stay empty.
Sutra is built on the opposite assumption. Less content, more conversation. People are transformed less by what they consume and more by what they articulate, and by seeing their own experience mirrored in someone else's words. A conversational course is the format that puts that belief into practice.
In a conversational course, your content and your prompts are woven together into a single, paced journey. A participant watches or reads something small, is invited to respond before they can move on, and then — once they've shared — gets to see what everyone else said and explore the collective insight with an AI guide called Cobolt. Then the next piece of content unlocks, and the rhythm repeats.
This guide explains exactly how that format works, how to build one, and how to design it well. It draws on three live courses you can study as you go:
Unlearning: For Change Agents (with Charles Eisenstein) — a four-session course built as a hub of conversational pages.
Friction and Fuel — a single, deep conversational page on behavior change, with eleven reflective prompts.
Action Intelligence (the Fast Forward program) — a video-driven course of seven prompts that moves participants from reflection into committed action.
We'll return to these throughout.
1 · What it actually is
The core loop
A conversational course is a page (or a series of pages) where content and prompts alternate, and where progression is gated on participation. The participant moves through a repeating loop:
Encounter content. A short video, a paragraph, an image — one idea at a time.
Respond to a prompt. A response box appears. They write their reflection and post it.
See the group. Once they've posted, other people's responses become visible.
Explore insights. They can open Cobolt to surface highlights, themes, and connections between their response and others'.
Move on. The next piece of content unlocks, and the loop begins again.
This loop is the whole engine. Everything in this guide is about designing it well — choosing what content to put in front of each prompt, writing prompts worth answering, and deciding how the responses are shared and synthesized.
What makes it different
A standard Sutra space can hold content, a discussion area, resources, and members. A conversational course folds the discussion into the content rather than parking it in a separate tab. The difference matters:
Content first, discussion "if you have time" — Reflection is the spine of the experience
Discussion lives in a separate area most people never open — The prompt is right there, in the flow, impossible to skip past
You finish alone — You finish having seen and been seen by the group
The teacher is the only voice — Every participant becomes a source of insight
AI summarizes for you — AI (Cobolt) helps you see the other people more clearly
The design philosophy underneath this is important: the AI is never the point. Cobolt is a lens that helps participants see each other, not an authority that replaces them. Every insight it surfaces should lead back to a real person's words. Keep that principle in mind — it shapes nearly every design decision below.
2 · The anatomy
Four building blocks make the loop work. As a facilitator you control all four.
1. Content blocks
The ordinary stuff of any Sutra page: video, text, images, headings, quotes, embeds. In a conversational course these are deliberately small — a single video and a few sentences, not a wall of material. Each content block sets up exactly one prompt.
In Action Intelligence, the pattern is almost mechanical and very effective: one video, one short paragraph of instruction, one response box. Seven times. In Friction and Fuel, each prompt is a dense reflective paragraph that distills a concept from the video series into a single question. Both work because the content does one job: prepare the participant to respond.
2. The prompt
The interactive prompt is the heart of the format. Mechanically it's a response box you drop into the page wherever you want people to stop and share. When you add one, Sutra automatically creates a linked discussion space behind it to collect everyone's responses — you don't manage that; it just happens.
Each response box has a handful of settings that change the experience completely:
Require a response before continuing (gating). When on — which is the default, and what all three example courses use — the participant cannot see or reach the content below until they've responded. This is what turns a page into a paced journey.
Show everyone's responses. When on (again the default in all three examples), participants see each other's contributions after they post. This is what makes it social rather than a private journal.
Allow people to skip. Off by default. When on, a participant can move past a prompt without responding — useful for optional or sensitive prompts.
Show AI highlights and summary (enables Cobolt). When on, the "Explore insights" experience becomes available on that prompt. All three example courses have this on for every prompt.
Post privately / privacy check. Lets a participant choose to share a response only with the facilitator rather than the whole group — valuable for vulnerable material.
A note on the prompt text: the response box itself carries a gentle default label ("How do you feel?"). In practice, experienced facilitators write the actual question as a line of text or a heading immediately above the box, and let the box be the place to answer. That's why, in all three example courses, the real prompt lives in the paragraph just before each box. Write your question in the content; let the box be the invitation to respond.
3. Gating and pacing
Gating is what distinguishes a conversational course from a page with comments. Because each required prompt hides everything beneath it until the participant responds, the experience becomes a series of deliberate steps. People can't skim to the end; they arrive at each idea having already done the work of the one before.
You can also insert pacing between steps — a pause that holds the next section closed until a set amount of time has passed (for example, "the next section unlocks tomorrow"). This is how you turn a single page into a multi-day drip without splitting it into separate lessons. Use it when you want reflection to settle before the next prompt, or to keep a cohort roughly in sync.
When a participant reaches the bottom of an unlocked section, they'll often see a simple instruction like "Please mark this section complete to unlock the next section" — the explicit hand-off that moves them forward.
4. Cobolt
Once a participant has responded to a prompt that has AI insights enabled, they see an "Explore insights" card (marked with a icon) reading "Explore insights from the conversation." Opening it launches Cobolt, Sutra's AI guide, in a side panel.
Cobolt is not a generic chatbot bolted onto the page. For each prompt it works specifically with the responses people have given, and it does three things:
Highlights. It surfaces a curated set of the most resonant, highest-quality contributions from the group — attributed to the people who wrote them — so the best thinking rises to the top instead of being buried in a long list.
Connections (bridging). Because the participant has already responded, Cobolt can connect their contribution to others': "Like your post, this one catches the reflex of aligning before thinking — Natalie's, in a neighborhood discussion." It's the cross-pollination that almost never happens in a normal discussion thread.
Summary and themes. It synthesizes what the group is collectively saying — the patterns, tensions, and recurring ideas — always grounded in and pointing back to the actual responses.
The crucial design point: Cobolt rewards participation. A participant only gets the full richness of the insights after they've contributed, and the insights are personalized to what they said. The act of sharing is what unlocks the act of seeing. This is the opposite of doom-scrolling other people's answers — you earn the collective view by adding to it.
Today's boundary, worth knowing: Cobolt works per prompt. It surfaces insight from the responses to the prompt the participant just answered, not across every prompt in the course at once. So design each prompt to be rich enough to stand on its own as a moment of collective insight, rather than relying on the AI to weave the whole course together.
3 · The participant's experience
It helps to walk the journey exactly as a learner does. Here's a single loop in Action Intelligence:
The participant lands on the page and sees a welcome video. Below it, one line: "If you're taking this course, you likely have something meaningful that you're working on. Please take a few moments to share a few words about your vision below."
A response box sits under that line. Everything further down the page is hidden — they can't scroll ahead.
They type their vision and post. (If privacy is enabled, they can choose to post only to the facilitator.)
The page now reveals a "View responses" option and the "Explore insights" card. They can read what other participants wrote, and open Cobolt to see highlights and how their own vision connects to others'.
They click Next (or simply continue), and the second video and prompt unlock: "What's your relationship with 'not knowing'?…"
The loop repeats through all seven prompts, each one building on the last.
For a facilitator or manager, the gating is lifted — you see the entire page at once, all prompts and all responses, so you can review, moderate, and get a feel for the room without having to respond to everything yourself.
4 · How to build one
Step 1: Set up the page
Create a space (or open an existing one) and build the conversational experience on its content page — the main Tiptap document. You can make a single self-contained conversational page, or a hub that links out to several (more on structure in section 6).
Step 2: Lay down the teach → prompt → respond rhythm
Build the page as a sequence of small units. Each unit is:
A piece of content — a short video, a paragraph, an image. One idea.
The question — written as a sentence or heading directly beneath the content.
A response box — inserted right after the question.
Then repeat. Resist the urge to stack three videos before the first prompt; the format works because reflection comes early and often.
Step 3: Add and configure each response box
When you insert a response box, set it up intentionally:
Require a response before continuing — leave on for a guided journey; turn off only where you genuinely want an open, non-blocking moment.
Show everyone's responses — leave on to make it social. Turn off only if you want private reflection with no peer visibility.
Show AI highlights and summary — turn on to give participants the Cobolt "Explore insights" experience. (This is the switch that makes collective insight possible — keep it on unless you have a specific reason not to.)
Allow skip — turn on for optional or emotionally heavy prompts where forcing a response would be wrong.
Privacy / post privately — enable on prompts that touch vulnerable territory, so people can choose who sees their answer.
Sutra automatically creates the linked discussion space behind each box — no extra setup.
Step 4: Add pacing where you want it to breathe
If you're running a cohort or want reflection to settle, insert a timed pause between sections so the next prompt unlocks after a delay (e.g., the next day). For a fully self-paced course, leave pacing off and let people move at their own speed — the response gating alone still keeps the journey ordered.
Step 5: Decide on the overall structure
Single page (like Friction and Fuel and Action Intelligence): all prompts live on one scrolling, gated page. Best for a focused journey of up to ~10–12 prompts.
Hub of pages (like Unlearning): a top-level page links out to separate session pages, each its own conversational experience. Best for multi-week programs with distinct sessions, plus side spaces for live-call recordings, a bulletin board, events, and so on.
Step 6: Test it as a participant
Preview the course logged in as a regular member (not a manager), so the gating is active. Walk the loop: respond, view others, open Cobolt, move on. You'll feel immediately whether the rhythm is right or whether a section is too heavy before its prompt.
5 · Writing great prompts
This is where conversational courses are won or lost. The platform handles the mechanics; the quality of the experience lives in your prompts. A great prompt makes people want to write, produces responses worth reading, and gives Cobolt rich material to work with.
Seven principles
Invite experience, not opinion. "What do you think about failure?" produces essays. "Can you think of a recent situation where you tried something and felt like you failed?" produces stories. Stories connect people; opinions don't. (Action Intelligence uses exactly this move.)
Make it personal and first-person. The strongest prompts ask people to look at their own life: "think about a personal habit, belief, or aspect of your life that you've been maintaining…" (Friction and Fuel). A participant can always answer a question about their own experience.
Ground it in the content just delivered. A prompt should be the natural next step after the video or text above it: "Reflecting on the concept of friction in behavioral change, think about a specific area in your … life where you experience unnecessary friction…" The content teaches the lens; the prompt asks them to look through it.
One idea per prompt. It's tempting to ask three things at once. Resist it — or if you must scaffold ("identify a behavior… then consider its current state… then…"), keep it to one coherent thread. Multi-part prompts are fine when the parts build a single reflection; they fail when they're three unrelated questions stapled together.
Ask for specifics. "Identify two specific areas," "quantify the gap," "a recent situation." Specificity makes responses concrete, comparable, and far more useful when the group reads each other and when Cobolt surfaces highlights.
Lower the stakes to get the first word. The opening prompt should be the easiest one in the whole course. Action Intelligence opens with "share a few words about your vision" — almost impossible to get wrong. Once someone has posted once, they'll keep going.
Make it safe to be honest. When a prompt asks for something vulnerable, say so kindly, normalize it, and consider enabling private posting or skip. Action Intelligence softens a hard ask with reassurance: "Remember, there's nothing wrong with having assumptions."
Prompt patterns you can reuse
The recall: "Can you think of a recent time when…?"
The mirror: "Reflecting on [concept from the content], where in your life do you notice…?"
The reframe: "How might you reframe [X] as [Y]?"
The surface: "What assumptions are you holding about…?"
The commitment: "Pick one [thing] you're willing to commit to in the next 30 days. What is it, and what will you test?"
The integration / closing: "What are some key insights or questions you're taking away?"
Anti-patterns to avoid
The quiz. Prompts with a "right answer" kill conversation. If there's a correct response, it isn't a conversational prompt — it's a check-for-understanding, and it belongs elsewhere.
The essay trap. Prompts so broad ("Reflect on leadership") that people either freeze or write something generic.
The pile-up. Five questions in one box. People answer the first and ignore the rest.
The disconnected prompt. A question that doesn't follow from the content above it, so people don't know what frame to answer in.
Too many prompts, too fast. If every paragraph demands a response, the course feels like a form. Pace your prompts; not every idea needs one.
6 · Designing the arc
Individual good prompts aren't enough. The magic of a conversational course is that the prompts form a journey — each one building on the last, so by the end participants have travelled somewhere they couldn't have reached in one step. Because the experience is gated, you have something most courses don't: a guarantee that people arrive at prompt 5 having genuinely done prompts 1 through 4.
A reliable shape:
Open — an easy, personal, low-stakes prompt that gets the first word out and surfaces where people are starting from. ("Share your vision." "How do you feel about not knowing?")
Deepen — prompts that turn the lens on the participant's real experience and current patterns. ("Where do you experience unnecessary friction?" "What assumptions are you holding?")
Apply — prompts that ask them to do something with the insight. ("Come up with two or three prototype ideas." "What small adjustments could you make?")
Commit — a prompt that asks for a concrete next step they'll actually take. ("Pick a prototype you'll commit to in the next 30 days.")
Integrate — a closing prompt that gathers the takeaways and points forward. ("What are you taking away? How will you keep going?")
You can see this arc clearly in the examples:
Action Intelligence runs the full open → deepen → apply → commit → integrate arc across seven prompts, moving deliberately from "share your vision" to "commit to a prototype and a plan to capture feedback." It's a course designed to end in action, and the prompt sequence enforces that.
Friction and Fuel uses its eleven prompts to walk through a complete behavior-change framework, each prompt mapping the participant's own life onto the next concept, and closing with a continuous-improvement reflection.
Unlearning spreads the arc across four session pages, each a self-contained conversational experience that opens, deepens, and closes — with the final prompt of each session explicitly gathering "key insights or questions you're taking away."
When you sequence prompts, ask: what has the participant already shared by the time they reach this one, and how does this prompt build on it? That question is the difference between a list of good prompts and a real journey.
7 · Getting the most from Cobolt
Cobolt turns a pile of responses into a felt sense of the group. To make it sing:
Keep "show everyone's responses" and "show AI highlights" on. Collective insight is impossible if responses are private and the AI layer is off. The default-on settings in all three example courses are the default for a reason.
Write prompts that produce comparable, specific responses. Cobolt's highlights and connections are only as good as the material. Specific, experience-based prompts (section 5) give it real signal; vague prompts give it mush.
Let participation do the work. Remember that the richest insights appear after a participant has contributed and are personalized to what they said. So encourage people to share first and explore second — that's the intended rhythm, and it's also the more transformational one.
Frame the AI honestly with your participants. Tell them what Cobolt is: a way to see what the group is discovering together, not a grader or an oracle. The goal is for people to feel more connected to each other, not mediated by a machine. When in doubt, point them back to the original voices.
Seed early prompts if you're launching to a small group. Cobolt and the "see the group" moment are richer with more responses. If your cohort is small or just starting, a few thoughtful early responses from you or co-facilitators give the collective layer something to work with.
8 · Pacing & delivery
The same conversational page can run three ways:
Fully self-paced. No timed pauses. Response gating still keeps everyone's journey ordered; people move as fast as they reflect. Best for evergreen courses and individual enrollment.
Paced / drip. Timed pauses unlock sections over days or weeks. Keeps a cohort loosely in sync and gives reflection time to settle. Best for live programs.
Cohort with live calls. The conversational page carries the asynchronous reflection; separate spaces (events, call recordings, a bulletin board) carry the synchronous community — exactly the structure Unlearning uses around its session pages.
Choose based on whether shared timing matters. If the value comes from a group moving together, pace it. If the value is the reflection itself, let it run open.
9 · Checklist
Content
Is each piece of content small — one idea before each prompt?
Does every prompt follow naturally from the content right above it?
Prompts
Does each prompt ask for experience, in the first person?
Is the opening prompt the easiest one in the course?
Does each prompt ask for something specific?
Is it one coherent question (not three stapled together)?
Have you removed anything that has a "right answer"?
The arc
Do the prompts build — open, deepen, apply, commit, integrate?
Does the course end somewhere (a commitment, an integration), not just stop?
Settings
Is "require a response" on for the prompts that should gate?
Is "show everyone's responses" on, so it's social?
Is "show AI highlights" on, so Cobolt works?
Have you enabled privacy or skip on sensitive prompts?
Did you add pacing if you're running a cohort?
Experience
Did you walk the whole thing as a regular member, with gating active?
Does the rhythm feel like a conversation, not a form?
10 · Common pitfalls
Too much content before the first prompt. People disengage before they ever participate. Get them sharing fast.
Turning off "show all responses" out of caution. You lose the entire social and collective-insight payoff. Prefer privacy/skip on individual sensitive prompts instead of going dark on everyone.
Forgetting to enable AI highlights. The prompt still works, but participants miss the "Explore insights" experience that makes the group visible.
Prompts that don't build. A course of ten good-but-unrelated prompts is ten separate moments, not a journey. Sequence them.
Reviewing as a manager and assuming members see the same thing. Managers bypass gating. Always test as a member.
Relying on the AI to connect the whole course. Cobolt works per prompt today. Design each prompt to be its own rich moment.
A worked example
Here's a five-prompt skeleton for a conversational course on any topic. Replace the content; keep the shape.
[Short welcome video] → "Before we begin: in a sentence or two, what drew you here, and what are you hoping to work on?" → response box (require response, show all, AI on).
[Teaching video on the core concept] → "Where in your own life or work do you notice [this concept] showing up? Pick one specific example." → response box.
[Short video deepening the concept] → "Think of a recent moment when [the pattern] played out for you. What happened, and what did it cost or give you?" → response box.
[Video introducing the practice / tool] → "Choose one small experiment you could run in the next two weeks. What exactly will you try?" → response box.
[Closing video] → "What's one insight you're taking from this, and one question you're still sitting with?" → response box. (Add: "Mark this section complete to finish.")
Open easy, deepen into real experience, apply, commit, integrate. Five prompts, each building on the last, every one of them gated so people arrive having done the work — and every one of them giving the group, and Cobolt, something real to reflect back.
The format is simple. The craft is in the prompts and the arc. Build the loop, write questions worth answering, sequence them into a journey — and let the participants, not the platform, be the source of the insight.
