How to Build a Pathway with the AI Assistant

In this article:

• Overview
• How to start a Pathway build
• The questions it asks
• Building on a document or a Knowledge article
• The outline card
• Editing the outline before you approve it
• Approving the outline
• Confirming the draft, and what gets created
• Opening the draft in the Pathway editor
• Undoing a draft Pathway
• Things to keep in mind
• FAQs
• Who can do this

Overview

The AI Assistant can build a Pathway with you. It interviews you about what the Pathway is for and who it is for, puts an outline up on a card you can edit, and then, once you have approved that outline and confirmed the action, creates the Pathway as an unpublished draft.

There are three separate moments in that, and it is worth knowing them apart. The interview is conversation. The outline is a proposal: nothing exists on your account while you are reading it. Approving the outline is you telling the Assistant that the structure is right, and it is the only thing that lets the Assistant build anything at all. Confirming on the confirmation card is what actually creates the records. For the general mechanics of confirming and undoing, see How the AI Assistant Makes Changes.

What you get at the end is a draft. It is created unpublished, so no learner can see it, and it is not assigned to anybody. Publishing is a separate step and never one the Assistant takes.

How to start a Pathway build

Open the Assistant with Ask AI at the top of any screen, or ⌘ + K (Mac) / Ctrl + K (Windows), or go to the AI Assistant home page. See How to Use the AI Assistant for the panel itself and The AI Assistant Home Page for the full page.

Then do any one of these:

• Just ask. "Build me a 30 day onboarding pathway for new sales reps" is enough to start.
• Click the Build a new pathway with me chip on the AI Assistant home page.
• If you are working through the guided setup, click Build your first pathway in its next steps list. It is described there as "A few questions, then a structure, then a draft. About 10 minutes, right here in chat." See The Guided Setup Walkthrough.

Tell it as much as you already know in your first message. Anything you say up front is a question it will not need to ask.

The questions it asks

The Assistant works out what kind of Pathway you are asking for, then asks only what it still needs. It has a budget of at most five questions for the whole interview, not five per round, and it asks fewer whenever you have already told it enough.

The questions arrive on a question card, one at a time, with option tiles where the answers are a short list. There is always an Other row so you can write your own answer instead. Skip a question and it uses a sensible default and tells you which one it used.

It recognises five purposes, and the purpose is what shapes the questions and the starting shape of the outline:

It will not ask you for a fact you have already given it, in that message or earlier in the conversation, and it will not write the outline out as prose in its reply. The card is where you read it, because prose is something you cannot edit.

Note: If you ask how to build a Pathway yourself, that is a how-to question and you get the steps from the help centre instead of an outline. Ask it to build one and you get the interview.

Building on a document or a Knowledge article

An outline can be grounded in something you already have, which usually beats describing it from memory.

Attach a document. Use the + button in the composer, or drag the file onto it, and the Assistant reads it before it replies. No extra step is needed from you.

• Readable types are .pdf and .docx for a Pathway build. Spreadsheets (.xlsx and .csv) are read as competency matrices instead, which is the Roadmap route.
• 10 MB maximum, and one file per message. Attach two and you will see "One file per message. I've taken the first one."
• Slides are not supported. Attach one and it tells you so: "I can only read Excel (.xlsx), CSV, PDF and Word (.docx) files. Slides aren't supported."
• Long documents are read from the beginning up to a limit of 20,000 characters, and this is stated rather than glossed over. The card says "Grounded in the beginning of your file, not the whole document" along with how much was read, and the Assistant tells you in its reply that it only has the beginning and asks whether the part that matters comes later.
• If the file cannot be read at all, because it is a scan, a set of images, damaged or password-protected, you are told which of those it looks like and offered the interview instead.

Point it at one of your own Knowledge articles. Say "base it on our joining instructions article" and the Assistant looks the article up and reads it. If more than one article could be the one you mean, it asks which. There is no picker for this, you just name the article. An article with no content yet cannot be built from, and the Assistant says so and offers to work from your answers instead.

One source at a time. If more than one is in play in the same reply, the first one is the one that counts.

Either way, the Grounded in ... note on the message is written by the platform and not by the Assistant, so it is a statement you can rely on. The Assistant is not allowed to claim a document as the basis of an outline on its own account.

A grounding source stays in play for the next five replies, and for as long as a build is actually running, so you do not have to re-attach the file to get from the interview to the outline.

The outline card

When the Assistant has enough, it puts the whole outline up on one card. The card is badged Structure only, and that badge is literal: no Pathway, no section, no step, no Form and no draft exists because of it.

The header shows the working title and the counts, for example 4 sections · 14 steps. The counts are derived from what is on the card, so they correct themselves as you edit.

Inside, each section lists its steps in the order a learner meets them. Every step carries a type chip:

• content a course.
• form a quiz, survey or sign-off.
• event something attended.
• text a written page.
• knowledge one of your own Knowledge articles.
• section a heading inside the section.

Where the Assistant wrote a note about why a step is there, a small question mark sits beside the title and the note is in its tooltip.

If your organisation has AI guidelines set at Settings → Features, a line above the card reads Applied your organisation's AI guidelines. That line is drawn by the platform from the setting itself rather than claimed by the Assistant, so it is true whenever it appears. Guidelines are standing rules the AI follows whenever it generates content for you, so they shape the wording of every outline and every assessment it writes. See Managing AI Features.

Editing the outline before you approve it

Everything you change here is local to the card until you approve it. Nothing is saved on a keystroke, because there is no such thing as a half-approved outline.

1. Rename a section or a step. Click the pencil on the row, type, and press Enter. Esc cancels.
2. Remove a section or a step. Click the × on the row.
3. Reorder the sections. Drag a section by the grip handle on its heading. Steps are not draggable, so move a step by removing it and adding it where you want it.
4. Add a section. Click Add a section at the bottom. A new section starts empty.
5. Add a step. Click Add a step inside a section. It is added as a text step, and you then rename it with the pencil. Add at least one step to any section you have added: an empty section is dropped when the outline is approved.
6. Attach a record you already have. On a content, event, form or knowledge step, click Attach from library and choose the record. A content step searches your library as you type; events, Forms and Knowledge use a filterable list. The attached row then shows a green tick with the record's name, or From your library for one the Assistant attached itself. Click the × on it to detach.
7. Ask for a new record instead. Click Create new on the same four types and the row is labelled New, created when you confirm. Nothing exists yet either way: the new record is written when you confirm the draft, which is the step after approving the outline.
8. Create a course from a document. On a content step, Create from a document opens a file picker for a PDF or Word (.docx) file, and the file starts uploading straight away while you carry on editing. Approve is held back until that upload finishes, with the reason beside it: "Waiting on the document, retry or remove it if the upload failed." Only one step per outline can be a course made from a document.

A step can attach a record or ask for a new one, never both, and choosing one clears the other.

Good to know: A form step with neither an attachment nor a create on it is not unfinished. That is the case where the platform writes the assessment for you, and it is the usual answer. Attach one of your own assessments when you have a quiz or sign-off you already use, and ask for a new blank one only when you want to write the questions yourself.

A content, event or knowledge step with nothing on it is fine too. It becomes a placeholder page in the finished draft, waiting for you to attach the real record in the Pathway editor.

If the shape is wrong rather than the wording, click Ask for changes. That sends nothing at all: it just puts your cursor in the composer so you can say what is wrong in your own words, and the Assistant proposes a new outline.

Approving the outline

Click Approve structure & draft content.

This is the consent step, and the Assistant genuinely cannot build without it. Ask it to skip ahead and it will tell you: "There is no approved outline for me to build. Ask me for a pathway outline first, then approve it on the card, I only build what you have approved."

Two things follow from that:

• One approval, one draft. An approval is spent the moment a draft is prepared from it. Ask for a second draft off the same outline and it declines and points you at the first one. If you want another Pathway, or a different shape, ask for a new outline and approve that.
• What you approved is what gets built. The Assistant sends no outline of its own with the build; the platform builds the version that came back from your card, including every rename, removal and attachment you made.

The card locks once you have approved it, and it stays locked when you reopen the conversation later, so an approval cannot be spent twice from the transcript.

Confirming the draft, and what gets created

Approving hands you a confirmation card headed Create draft pathway, summarising the action as "Create as an unpublished draft pathway." Under it the card lists the consequences in plain terms, which is where you see how much of the outline is already real:

• how many steps use records you already have, so those steps are ready to go;
• how many new records will be created in your library when you confirm, which do not exist yet;
• how many assessments are written when you confirm, so their questions are not in the preview;
• and, always, "It is created unpublished, so no learner can see it. Publishing is a separate step, and never one I take."

Click Confirm and the build runs. Confirm within the hour: an unconfirmed preview expires, because the data behind it may have moved on.

While it works you see it happen a step at a time on a progress bar, for example "Created the draft pathway Sales Onboarding, now adding its steps", then "Step 4 of 14 added". Writing an assessment's questions is the slowest part and the line says so while you wait on it. This is the longest-running thing the Assistant does, so leave the panel open until the receipt appears. That receipt is where the undo lives.

When it finishes, here is what is on your account:

• The Pathway, unpublished and not public, with its sections and steps in the order you approved. You are its owner.
• Steps that attached an existing record, pointing at that real course, Event, Form or Knowledge article. Nothing is copied or duplicated, and the step takes its title, description and image from the record itself, so the Pathway reads the way your Library does.
• Assessments the platform wrote, created in your Forms library as real quizzes with real questions, each named after its step's title.
• New records you asked for in your library: a blank Event, a blank Knowledge article or a blank assessment, each named after its step's own title and each carrying a short note telling you to write it and then delete the note. A blank Event is given placeholder dates rather than a real schedule, and a blank Knowledge article is created as a draft.
• A course made from your document, where a content step asked for one.
• Placeholder pages for any content, Event or Knowledge step that had nothing attached to it, each saying what to attach.

The receipt afterwards says "The draft pathway has been created. It is unpublished, so nobody can see it yet." If some of the outline could not be added, you get a partial receipt instead, naming what was skipped and telling you to open the Pathway in the editor to see what is there. It also flags the honest edge cases: a record the outline pointed at that has since left your library is built as if nothing had been chosen for it, and a document that is no longer available means its step is created as a placeholder page.

Note: If AI assessment writing is switched off for your account, an outline containing assessments cannot be built. The Assistant tells you so and offers an outline with no assessments in it instead. An Admin can turn it back on under Settings → Features.

Opening the draft in the Pathway editor

The receipt carries an Open in pathway editor link straight to the new Pathway. That is where you finish it: fill in the placeholder pages, review the assessments the platform wrote, add covers and descriptions, and publish it when you are happy.

The draft also just sits on your Pathways page like any other unpublished Pathway, so you can come back to it later without the conversation.

Undoing a draft Pathway

The receipt's undo button reads Discard draft, because nothing was ever visible to a learner. It deletes the Pathway again, along with the records that were created for it, and it is available for 7 days after the build. The receipt counts the days down for you.

Two things it will not do:

• If the Pathway has already been assigned to somebody, it refuses, because deleting it would take their learning with it: " has already been assigned to somebody, so deleting it would take their learning with it. It has been left in place, delete it from the Pathways page if that is what you want."
• If one of the created records cannot be removed, or an assessment written for the Pathway is now in use elsewhere, that record is left in your library and the undo says which kind it was.

See How the AI Assistant Makes Changes for the full undo rules, and The AI Assistant Activity Log for the record of what was built and by whom.

Things to keep in mind

• Nothing is created until you confirm. The outline creates nothing, and approving the outline creates nothing either.
• What you get is an unpublished draft, not assigned to anybody. The Assistant cannot publish a Pathway and cannot assign one as part of a build. Assigning is a separate action, covered in How to Assign Learning with the AI Assistant.
• The interview budget is five questions for the whole build.
• An outline can hold at most 20 sections, and at most 30 steps in a section. These are there to catch an over-eager proposal rather than to restrain you, since they sit well above any realistic Pathway.
• The whole outline also has to fit a size budget of 32 KB. If it does not, the Assistant refuses it before you ever see it: "That outline is too large for me to put to you. The limit is 32 KB. Ask me for a smaller one: fewer sections, fewer competencies, or shorter notes."
• Attachments are 10 MB maximum, one file per message, and readable as .pdf, .docx, .xlsx or .csv. A new course can only be made from a .pdf or a .docx.
• Searching your library from inside the Assistant is limited to Owners and Admins. A Manager sees "Searching the library from here is available to administrators only." on the content picker, so a Manager attaches an existing course by opening the draft in the Pathway editor afterwards instead.
• Adding a section on the card without adding a step to it means that section is dropped when the outline is approved.
• A new Event, Knowledge article or blank assessment created for the draft is genuinely blank. Each one carries a note asking you to write it.
• AI guidelines at Settings → Features shape everything the Assistant writes for you, including the outline's wording and the questions in any assessment it generates. They are worth setting before you build.
• AI can make mistakes. Read the outline properly before you approve it, and read the draft before you publish it.

FAQs

Does approving the outline create the Pathway?
No. Approving tells the Assistant the structure is right. The Pathway is created after that, when you confirm the Create draft pathway card. Those are two deliberate clicks.

Is the Pathway published or assigned to anyone?
Neither. It is created unpublished, so no learner can see it, and nothing is assigned. You publish it yourself from the Pathway editor.

Can I change the outline after the Assistant has proposed it?
Yes, and you are meant to. Rename and remove sections and steps, drag the sections into a different order, add your own, and attach records from your library. For anything structural that the card cannot do, click Ask for changes and say what you want in the composer.

Can I reorder the steps inside a section?
Not by dragging. Sections drag; steps do not. Remove the step and add it back where you want it, or ask for a revised outline.

What happens to a step that has nothing attached?
A content, Event or Knowledge step becomes a placeholder page telling you what to attach, which you do in the Pathway editor. A Form step with nothing on it is different: that is the assessment the platform writes for you.

Where do the assessment questions come from?
They are written when you confirm, not when you approve, so they are not in the preview. Your organisation's AI guidelines are applied to them.

Can I build a Pathway from a PowerPoint?
No. Slides are not supported. Save the material as a PDF or a Word document and attach that.

Can I attach more than one document?
One file per message. If you attach several, the Assistant takes the first and tells you it has done so.

Can I use one of our own Knowledge articles as the basis?
Yes. Name the article and the Assistant reads it, and the card states what it read.

Can I get rid of a draft it built?
Yes. Use Discard draft on the receipt within 7 days, which deletes the Pathway and the records created for it. After that, or if the Pathway has already been assigned to somebody, delete it from the Pathways page instead.

Can I build two Pathways from the same outline?
No. One approval builds one draft. Ask for a new outline and approve that.

Who can do this

Owners and Admins can build a Pathway with the AI Assistant, and so can SuperPath Admins. Managers can too, but only when your organisation's Who can use the AI Assistant setting at Settings → Features is set to Admins and managers, and with one real limitation: searching your library from inside a conversation is Owner and Admin only, so a Manager attaches existing courses in the Pathway editor after the draft is built.

People Managers can use the AI Assistant on the Admins and managers setting, but they cannot create Pathways at all in SuperPath, so the Pathway builder is not available to them. Content Managers, Instructors, Employees and Restricted users cannot use the AI Assistant, whichever setting is chosen, even though Content Managers and Instructors can create Pathways by hand.

Turning the AI Assistant on, choosing who can use it, setting your AI guidelines and switching AI assessment writing on or off are all done in Settings → Features by an Owner or Admin. See Managing AI Features.

Updated on: 11/09/2026