# Create and edit scenarios > Turn a repeatable conversation into a clear practice brief with meaningful scoring. Source: https://lira.thoughtly.com/docs/manage-training/create-scenarios Audience: manager, admin Last reviewed: 2026-10-22 Managers and admins can author scenarios. Managers can edit the scenarios they created; admins can manage scenarios across their workspace. ## Create a draft Open **Scenarios** and choose **Create scenario**. Give the scenario a specific name, identify the product, and describe what the learner will practice. A name such as “Handle a renewal objection” is easier to choose than a generic title such as “Sales practice.” The section navigation lets you move freely between the product, counterpart, conversation, goals, knowledge, call settings, and scoring. Checkmarks show which required sections are complete. On a phone, use the **Section** menu. You can save an incomplete draft and return to it later. ![Scenario settings with section navigation, product fields, and a Generate with AI action.](https://lira.thoughtly.com/docs/images/product/scenario-editor.webp) Move between setup sections while keeping the scenario context visible. ## Generate with AI Choose **Generate with AI** and describe the conversation you want to practice. Include the product facts, the counterpart, the situation, and the outcome you want the learner to achieve. You can also request a difficulty, call length, or custom skill. Include a public product website, such as `https://example.com`, or name a recognizable public product. When web research is configured, Lira uses Firecrawl to read relevant public pages before drafting. Public source links appear with the result so you can check the evidence. Your private workspace sources remain separate; full prompts and private source content are not sent to the web search service. If research is unavailable, Lira tells you and uses only the information already supplied. Select **Generate draft**. AI prepares the editable scenario details, call settings, skills, and certificate requirements. Your selected knowledge sources stay selected. You can close the dialog, change pages, or refresh: the job keeps running. The **AI activity** control in the workspace header shows its progress and lets you return to the result. A finished proposal does not save or publish the scenario. If you stayed on the same draft without editing, the result fills the draft for review. If you returned later or changed fields while AI was working, choose **Review draft**, compare the proposal, and apply it deliberately. **Undo** restores the draft from before you applied the proposal. Newer edits are never silently replaced. If generation fails, reopen it from **AI activity** and try again. Dismissing a finished update only hides it from the activity list; closing the generator does not cancel ongoing work. ![AI scenario generation dialog with an example coaching scenario prompt and Generate draft button.](https://lira.thoughtly.com/docs/images/product/generate-scenario.webp) Describe a specific conversation, then review and edit the generated draft before publishing. ![The workspace header showing a completed AI draft with a Review link in its activity panel.](https://lira.thoughtly.com/docs/images/product/ai-activity.webp) Return to a generated draft from any workspace page. ## Shape the conversation Build the brief around four questions: 1. Who is the counterpart, and what is their role? 2. What do they already know, and what is happening now? 3. What might they ask or object to? 4. What observable outcome should the learner reach? Use approved [knowledge sources](https://lira.thoughtly.com/docs/manage-training/knowledge-sources) to ground product information. If you use generated suggestions, review and edit them before publishing. Do not present an invented customer detail or product claim as fact. ## Configure and test Use the **Difficulty** slider to choose how challenging the counterpart should be. Each level explains the behavior to expect. - **Easy:** a cooperative customer who raises a mild objection and shares more after useful discovery. - **Medium:** a guarded customer who challenges vague answers and warms up when the learner listens. - **Hard:** a busy, skeptical customer who expects specific answers before offering more time. - **Expert:** an impatient customer who frequently interrupts weak or lengthy pitches and looks for an exit. The learner needs to acknowledge their time, earn permission, ask relevant questions, and listen. A useful response earns more attention; ignoring a final warning can end the browser call. Expert is designed to test earning the customer's attention, not just completing a script. A weak but relevant sales approach gets one brief opportunity to recover after a final warning. Repeating a generic pitch, asking only about the product category, or asking the customer to supply the relevance does not reset that warning. Dismissive or hostile replies to the initial relevance challenge, and inappropriate personal questions such as asking what the customer is wearing, end the conversation immediately without another chance. Concise, respectful permission and specific discovery earn time; ordinary pauses, filler, accents, or stuttering are not failures. At any difficulty, a clear request to hang up or stop the role-play ends the call instead of switching to a general assistant. A customer-ended call still appears in the learner's attempts with the available recording, transcript, and feedback, subject to the workspace's privacy settings. Set **Maximum conversation length** from 1 to 30 minutes. This is a time limit: the call ends automatically when it is reached. Existing assignments keep the limit from their assigned scenario version. Open **Customer voice** to browse 31 voices, including Coral, Marin, Cedar, and the expanded GPT-Live voice catalog. Search by name and play a sample before choosing **Use** to confirm. Scroll within the framed list to browse more voices; the confirmation buttons stay visible. Starting another sample stops the previous one. Cancel keeps the current voice. ![A searchable customer voice picker with 31 voices, including Coral, individual play buttons, and a confirmation action.](https://lira.thoughtly.com/docs/images/product/voice-picker.webp) Find a voice, hear a sample, then confirm your choice. Open the matching **Language** card to search languages and regional variants by name, native name, or language tag. Select an option, then choose **Use** to apply it. Cancel keeps your current language. For a tag outside the list, choose **Use a language tag** and enter a valid BCP 47 tag, such as `zh-Hant-TW`. ![Difficulty and maximum conversation length sliders above matching customer voice and language cards.](https://lira.thoughtly.com/docs/images/product/call-settings.webp) Set the challenge level and time limit, then choose a voice and language. ![A searchable language picker with native names, monochrome flags, and a selected regional language.](https://lira.thoughtly.com/docs/images/product/language-picker.png) Search by language or region, then confirm your choice. Use the author preview to check the selected voice and language, whether the counterpart behaves as intended, and whether the goal is achievable. Review the [scoring rubric and certificate requirements](https://lira.thoughtly.com/docs/manage-training/scoring-and-certificates) before publishing. An author test is for checking the setup; it does not earn a training certificate or become a learner submission. ## Save and publish Use **Save draft** while working. A draft is not part of the published learner collection. **Review & publish** shows any sections that still need attention and links directly to them. Select **Publish scenario** when the brief, sources, call setup, and requirements are ready. To edit an existing scenario, open its overview and select **Edit**. Use the settings navigation to move between sections, then save your changes. > **Existing assignments keep their standard** > > Saving changes to a published scenario creates a new published version. Existing assignments and attempts retain the version they started with, including its rubric and certificate requirements. > > Your editable scenario card shows the latest published difficulty and opens that version. If you also have an older personal assignment, its settings remain available through **Open your assigned version** on the scenario page. After publishing, use **Assign** on the scenario to [choose the people who should practice it](https://lira.thoughtly.com/docs/manage-training/assignments).