> For the complete documentation index, see [llms.txt](https://herd-security.gitbook.io/herd-security-docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://herd-security.gitbook.io/herd-security-docs/trainings/getting-started-with-trainings/create-training-with-ai.md). # Create Training with AI This page provides guidance on generating trainings in minutes using Herd’s AI ## **Generate Training with HerdAI** AI-Generated Trainings make it fast and easy to deliver learning content without starting from scratch. By simply providing a topic or reference material, Herd’s AI builds a structured training for you in minutes. AI-Generated Trainings are ideal when you need to: * Quickly create **onboarding modules** for new hires. * Provide **fast refreshers** on security or compliance topics. * Share **updates** on new policies, procedures, or emerging risks. #### Starting an AI-Generated Training Creating a training with HerdAI is a **conversational** experience—you describe what you want in chat and HerdAI builds a draft you can refine. There's no structured form to fill out. * On the left-hand sidebar, click **HerdAI**, or * From the **Trainings Page**, click **New Training** and choose **Create with HerdAI** (this opens the HerdAI chat with a starter prompt ready to go).
## Building Your Content
**Step 1:** Describe the training you want, or upload a reference document for HerdAI to build around. The more detail you give, the closer the first draft will be. Helpful things to mention: 1. **What topic** should the training cover? (e.g., phishing awareness, password security, data privacy, compliance, etc.) 2. **How many steps** would you like? (default is around 5–7 steps) 3. **Any quiz questions?** Should some steps be quiz questions to test knowledge? 4. **Any specific style or tone?** (e.g., formal, casual, scenario-based) **Step 2:** Send your message. HerdAI generates a complete training preview right in the chat. Keep chatting to refine it—ask HerdAI to add steps, change the tone, or adjust the title and tags, and the preview updates in place. ## **Training Content** Your **HerdAI Generated Trainings** can be found under your company's **Training tab** within the **Trainings Page.** **Step 1:** Once your training has been generated and published in HerdAI, click **View Training** to view and manage the content. When you're on a training’s details page, you're able to make changes to the training title, description, tags, content and structure. ### Training Controls
The training page header carries **Preview** and **Duplicate** buttons, and a **More actions** menu with **Create similar training**, **Export SCORM**, **Close training**, **Archive training**, and **Delete training**. (See [Navigating the Training Page](/herd-security-docs/trainings/getting-started-with-trainings/navigating-training-page.md#the-training-page) for the full header.) * **Preview** Click Preview to see how the steps will appear when delivered. The preview renders your training inside a realistic **Slack-style window** — or a **Teams-style window** for Teams-only organizations — with sample values filled into variables, so you see exactly what members will see. * **Duplicate**\ Make an editable copy of this training in your library. * **Create similar training**\ Spin up a new HerdAI draft that reuses the structure and style of this training as a starting point. * **Export SCORM**\ SCORM stands for **Sharable Content Object Reference Mode.** Clicking the **Export SCORM** button in Herd will allow you to take a training you've built in Herd and **export it as a SCORM package** so you can use it in other LMS platforms (like Cornerstone, Docebo, Moodle, TalentLMS, etc.) * **Delete training**\ This option soft deletes a training. Use it if you no longer need a custom or cloned training. ## Step Editor
**Step 1: Review and edit** the generated content or quiz questions to fit your team’s voice and priorities. **Step 2:** To **Reorder** training steps, click on the 6 dots next to each step and drag it to the desired location. **Step 3:** To add more content, scroll to the bottom of the created training and press **Add Step.** **Step 4:** To tailor the content to your organization's goals, click into any of the drafted steps and customize the content your users will see. * Here’s what you can do in the step editor: * **Message Body & Preview**\ Type your content into the **Message Body** box.\ To preview how the steps will look like, press the **Preview** button above the training shows how the message will appear in **Slack or Microsoft Teams**. * **Formatting Tips**\ When you click into the step you wish to edit, you’ll find the following formatting shortcuts above: * **Bold** * *Italics* * Underline (kept in Slack, Microsoft Teams and the web player) * ~~Strikethrough~~ * [Links](https://example.com) * Click **“+ Add Button”** to insert response options or quiz prompts. * These buttons help guide users to the next step or capture answers. * **Limit: 30 characters** to ensure proper display in Slack. * Use the **Correct Answer** field to select the correct response for quiz steps. * **Minimum time** (optional): tick the **Minimum time** checkbox at the right-hand end of the step's toolbar row and enter a number of **seconds** (it starts at 60; 1–86,400, so up to one day). Learners can't continue past the step until that much time has passed since it was shown, and the time spent is recorded for compliance reporting. Untick the box to remove the minimum. Learners see the wait in minutes and seconds, for example *after 45 sec* or *1 min 30 sec of 2 min required*, and the training CSV export reports the required and actual time in seconds. In Slack and Teams, the step arrives with a note saying when its buttons will appear, and the buttons are held back until the time is up: * **Slack**: the buttons appear **on the same message**, and the note changes to *You can continue now.* Editing a message sends no notification, so a learner who stepped away isn't pinged; reminders still go out as usual. Steps with an uploaded file, audio or video, and phishing or deepfake voice steps, get their buttons in a new message instead. * **Teams**: the buttons arrive in a new message. If the step is sent again after its buttons have appeared (for example when a learner switches back to the training), it arrives with its buttons. If you shorten or remove a step's minimum while learners are waiting on it, anyone whose new time is up gets their buttons at once, and the rest wait only until the new time. {% hint style="info" %} If you do not align the **Correct Answer** with the corresponding button, users will be marked incorrect even if they selected the right choice. This is essential for **accurate scoring**—at least one correct answer must be marked. {% endhint %} **Step 5: Attach Images or Videos** * Use the step's **Attach Content** menu to enrich your content. It holds **Upload**, **From library** (your [Media Library](/herd-security-docs/trainings/getting-started-with-trainings/media-library.md)), **Generate AI video**, **Generate AI image**, **Generate AI audio** (text-to-speech), and **Generate multiple** (several types in one pass). While AI generation is unavailable, only the AI items are disabled, with the reason shown above them. * A generated video is named after its training (for example *Phishing Basics.mp4*), and that's the file name learners see in Slack and Teams. To change it, click the name under the video's thumbnail in the step editor and enter a new one; the `.mp4` extension is kept. Videos generated before this change keep their old name (such as *talking-head-….mp4*) until you rename them. #### AI video generation **Attach Content → Generate AI video** opens the **Generate AI video** page (the Media Library's **Generate media → AI video** opens the same one). Under **Style** you choose between two kinds of video: * **Animated scene** — AI-generated b-roll with a voiceover. You can edit the narration script, pick the **Voiceover** provider and voice (or leave your organization's [default narration voice](/herd-security-docs/settings/settings/general-and-approvals.md#ai-narration-voice)), add closed captions, and optionally supply a **Starting image**. * **Talking head** — an avatar reads the script to camera. Requires HeyGen to be configured for your workspace. For a talking head, the **Avatar and voice** section has an **Avatar** picker and a **Voice** picker. Both are searchable: start typing (**Search avatars…** / **Search voices…**) and every word you enter narrows the list across name, style, gender, and language, so "calm", "Liam", or "liam german" all work. Quick filters sit beneath each box — **Gender** and **Style** for avatars; **Gender**, **Tone**, and **Language** for voices — and the **Workspace default** row is always offered. Each avatar has a **Play preview** control that shows it moving, and each voice a play control that plays a short **sample**, so you can check both before you commit. Once you've rendered a talking-head video for a step, the dialog reopens with the avatar and voice you used last time for that step preselected (in the Media Library, your own last choice), so you don't pick them again for every render. Under **Finishing**, **Add our logo** places your organization's logo in the corner of the finished video, using the logo saved for [completion awards](/herd-security-docs/trainings/getting-started-with-trainings/completion-awards.md). While a video renders, Herd reports real progress right in the chat — *"Writing the narration script…"*, *"Rendering scene 2 of 5…"*, *"Stitching audio, video, and captions…"* — and the finished video card includes a collapsible **Transcript** of the narration and a **Download video** link.
{% hint style="info" %} **Want a real person from your organization on camera?** Admins can build a **custom avatar and voice** of a colleague from one clip, with that person's consent, via **Create custom avatar and voice** in the talking-head **Avatar and voice** section. Once ready it appears under **Your organization** in the Avatar and Voice pickers. See [Custom Avatars and Voices](/herd-security-docs/trainings/getting-started-with-trainings/custom-likenesses.md). {% endhint %} AI image generation follows a content policy: a prompt asking for a real person's likeness, graphic violence, or similar disallowed content is refused with a clear message rather than generating an image. Rephrase the prompt and try again. Security-awareness imagery (phishing, scams, breaches, hackers) is fine. **Step 6:** Your edits **save automatically**—a **Saving…** / **Saved** indicator at the top of the editor confirms when changes are stored. There's no separate Save button to press. After finalizing, you can preview the AI-Generated Training and then choose to [**send it immediately** ](/herd-security-docs/trainings/getting-started-with-trainings/sending-trainings.md)or [**queue it for future release**](/herd-security-docs/trainings/getting-started-with-trainings/queue-trainings.md). *** ## Ask HerdAI for a Training in Chat You don't have to start from the builder—you can simply **describe the training you want to HerdAI** and get a ready-to-review draft back in the conversation. **Step 1:** Open **HerdAI** from the left-hand sidebar, or use the **HerdAI** widget on your **Dashboard**. **Step 2:** Type what you want, then send. For example: * *"Build a training on phishing awareness."* * *"Create a 10-step training on password security with 3 quiz questions."* If you'd rather be guided, just say so—HerdAI will ask what topic to cover, how many steps it should have, and anything else it needs before generating a draft. **Step 3:** HerdAI generates a **training preview** right in the chat. The preview card shows the **title, description, tags,** and every step rendered the way it will appear in **Slack or Microsoft Teams**, with quiz steps showing their answer options and the correct answer marked. **Step 4:** Keep chatting to refine the draft. You can ask HerdAI to regenerate it ("make it funnier"), add steps ("add another step"), change the title or tags, or remove a step—and the preview updates in place. **Step 5:** When you're happy with it, press **Publish** on the preview card (or tell HerdAI to publish it). Once it's published, click **View Training** to open it on the **Trainings** page, where you can edit, assign, and send it like any other training. {% hint style="info" %} HerdAI will only publish a training **after** you've reviewed the preview and confirmed. It won't generate and publish in a single step. {% endhint %} ### Build on an Existing Training From a training's page, choose **Create similar training** in the More actions menu to hand HerdAI the training as a **style reference** so a new draft matches its structure, tone, and depth. When you start a chat this way, a banner shows which training you're building on, and you can clear it at any time. ### Format Your Own Content As-Is By default HerdAI **transforms** what you give it: it restructures, rewrites, and writes the quiz questions. If you already have a finished training and want it kept verbatim, ask for it to be formatted as-is (*"format this as-is, don't change anything"*) or open the model picker in the chat composer and untick **Transform**, which switches the mode to **Format as-is**. In this mode HerdAI only splits your content into steps at your headings and maps your questions and answers one to one; nothing is rewritten, reordered, added, or removed. If your content can't be preserved faithfully, HerdAI tells you what it couldn't keep instead of quietly changing it. The choice applies to the current chat, so next week's one-line brief isn't formatted verbatim by mistake. When you paste formatted content (from Word, Outlook, Google Docs or a web page) into the HerdAI chat box or the **Herd AI** panel in the training editor, its web and email links come with it. HerdAI keeps each link's exact address, and in **Format as-is** a training that drops, shortens or changes a link doesn't pass. ### Pacing a Step on a Typed Reply Every step that has no buttons gets a **Continue** button at delivery. If you'd rather a particular step hold until the learner types a reply, tell HerdAI which one (*"the second step should wait for the learner to reply"*) and it turns on that step's **Wait for the learner to reply before continuing** setting. Quiz steps can't use it, since they already wait for an answer. See [Creating Trainings](/herd-security-docs/trainings/getting-started-with-trainings/creating-trainings.md#building-your-content) for what the setting does. *** ## Grounded in Your Own Documents AI-generated trainings aren't generic: generation automatically grounds itself in **your organization's** [**Policies**](/herd-security-docs/policies/policies.md) — the policies, runbooks, and internal docs you've added there — plus your [Organizational Context](/herd-security-docs/settings/settings/general-and-approvals.md#your-organizational-context). Relevant excerpts are retrieved by topic behind the scenes, so a training about data handling reflects *your* data-handling policy, not a textbook one. You can also ground a specific draft directly: attach a document to your HerdAI message, or paste a URL into the topic and Herd reads it as source material. {% hint style="info" %} The richer your Policies corpus and Organizational Context, the more your generated trainings sound like your organization. Grounding never blocks generation — if nothing relevant is found, you simply get an ungrounded draft. {% endhint %} *** ## Story-Based Trainings Ask HerdAI for a **story-based** (narrative) training — *"tell it as a story"*, *"make it first-person"* — and it writes the training as a first-person account of a real-shaped attack, taken apart lever by lever: what the target saw, what they felt, and which psychological levers the attacker pulled. Story-based trainings carry a green **Story-based** badge on the training page and in chat previews. How raw those accounts are allowed to get is an organization-wide dial: **Story-based training tone**, in the training settings drawer (gear on the Trainings page, under **Creation & feedback**): * **Reserved** — measured and clinical; stakes come across through facts rather than emotion. * **Balanced (recommended)** — real emotional stakes in restrained language. * **Bold** — keeps the uncomfortable edges the real attack had, and says plainly how cynical it was. At every level the content explains the tactics from the outside — it never uses them against the learner. *** ## Letting Herd AI Plan Your Library Beyond building one training at a time, HerdAI can work out **which trainings your organization is missing**. Ask it — *"figure out what trainings would be good for us"*, *"what are we missing?"* — and it reviews your tech stack, policies, live compliance campaigns, threat feed, and existing library, then proposes a prioritized list with its reasoning. Pick the ones you want and HerdAI drafts each one, showing a preview card per training. Nothing goes live until you confirm: press **Publish** on a card, or **Publish all N** to publish the whole batch. The drafts also appear under the **Drafts** tab on the Trainings page. Nothing is ever assigned to anyone automatically, and if your organization requires review, the new trainings wait in **Reviews** instead. You can also put this on autopilot: the **Nightly training generation** toggle (training settings drawer → **Herd AI automation**, off by default) has Herd AI run the same review each night and create up to **3** missing trainings for your review. It pauses while your organization already holds **10** unused Herd-written trainings (not assigned to anyone, a group, a track, or a compliance campaign, and not archived), and below that it only creates what fits under 10. Herd-written trainings wait quietly in **Reviews**: approvers aren't messaged about them, in the daily review digest or when one is submitted, so those messages are kept for trainings a person wrote. *** ## Questions Mid-Training Every training runs as one guided, step-by-step conversation. (The separate free-form conversational training format has been retired, and any trainings built that way were converted to steps.) Learners still get a conversation when they want one: in Slack or Teams they can reply in the training thread with a question at any point, and HerdAI answers it in the context of the current step, then steers them back to the lesson. If a learner keeps asking questions on the same step, after the fifth one Herd re-sends the step with its buttons so they can pick up where they left off.