Skip to main content

AI Prompting

If your site has the AI features enabled, you can describe a piece of content in plain language and have an AI assistant build the whole content item for you — with the same structure as one built by hand. Several WinnerWare modules ship content schemas (recipes) that teach the assistant the exact shape of every custom part, so an AI-generated item lands correctly in the admin editor.

This page collects small, copy-ready prompts for the content types those schemas cover, plus prompts for querying survey and Agent Trainer report data.

Prerequisites​

  • Enable the CrestApps AI features and set up an assistant that can author content (see AI Chat).
  • To have the assistant send an email on your behalf, turn on the Send Emails capability first (see AI Capabilities).
  • Enable CrestApps.OrchardCore.Recipes. The content schemas register automatically once this feature is on — there is nothing extra to configure.
  • Enable the module that owns the content type you want to build (Surveys, Agent Trainer, Agreements, AI Chat, or Knowledges).

How to prompt​

Open your content-authoring AI assistant and describe the item. Keep the prompt short and concrete — always name the content type, then list the fields you want set. Give reusable elements clear, distinct titles so any references between them resolve unambiguously. Build one content item at a time, and always review the result before publishing.

Build a survey​

The Surveys module lets you describe a survey in plain language and have the assistant build the whole structure — pages, questions, controls, visibility rules, and control groups. The single prompt below builds a complete Survey content item, including conditional logic:

Create a Survey titled VPN Readiness Assessment that gathers what we need to set up a VPN for the interviewee. Use clear, distinct titles for every control so the visibility rules resolve correctly.

  • Page 1 – Work location: a predefined-list control "Where do you work?" with options Remote, Hybrid, On-site.
  • Page 2 – Home network: show this page only when "Where do you work?" is Remote or Hybrid. Add a numeric control "Home internet download speed (Mbps)"; a predefined-list control "Router type" (ISP-provided, Own router, Mesh system, Not sure) shown only when "Home internet download speed (Mbps)" is greater than 0; and a text control "Which mesh system do you use?" shown only when "Router type" is "Mesh system".
  • Page 3 – Devices & experience: a predefined-list control "Which devices will connect?" (Windows, macOS, iOS, Android, Linux); a numeric control "Devices connected at the same time"; a predefined-list control "Have you used a VPN before?" (Yes, No); and a text control "Which VPN did you use?" shown only when "Have you used a VPN before?" is Yes.
  • Page 4 – Access needs: a text control "Internal apps or sites you must reach through the VPN", shown only when "Where do you work?" is Remote or Hybrid.

This one prompt exercises multiple control types (list, numeric, text) and all three levels of visibility — page-level, control-level cascades, follow-up reveals, and an OR condition. See Surveys for the full survey structure, visibility model, and reporting options.

Build an Interaction Story​

The Agent Trainer module builds Interaction Story content items — guided training scenarios the AI role-plays with a trainee. Name the story, give the AI its scenario context and opening line, and describe each step:

Create an Interaction Story titled Frustrated VPN Customer for agent training.

  • Context: you are a remote employee whose VPN keeps dropping every few minutes; you are frustrated but cooperative. Stay in character.
  • Opening statement: "Hi, my VPN keeps disconnecting and I can't get any work done."
  • Default interruption phrase: "Are you still there? I really need this fixed."
  • Step 1 (AI): wait for the agent to acknowledge the problem and ask a clarifying question. Validate with semantic assertion that the agent showed empathy.
  • Step 2 (AI): describe that it drops about every five minutes on Wi-Fi. Score the agent out of 10 on whether they identified the network as the likely cause.

How it maps: the story-level fields fill the Interaction Story part (context, opening statement, interruption phrase), each step becomes a story step with an action type of AI or Predefined, an optional validation strategy (None, Classifications, Semantic Similarity, or Semantic Assertion), and an optional scoring rubric. See AI Agent Trainer.

Query Agent Trainer report data​

After agents run a simulation, you can ask the assistant to read the complete transcript data for one Interaction Story. It retrieves the available agent, score, messages, steps, interruptions, and coaching summary for every completed simulation in the date range, then analyzes that raw data for you. You do not need to choose a report type.

Start by asking the assistant for the exact story name:

Discover: What interaction stories can I run a report for?

Then request the report with the selected story name and an explicit date range:

Pull a report: Run the full interaction story report for De-escalation from 2026-06-29 through 2026-07-29. Summarize the agents' performance, common problems, scores, interruptions, and coaching feedback.

Ask a focused follow-up when you need an answer from the same data:

Find coaching themes: For De-escalation from 2026-06-29 through 2026-07-29, which steps caused the most interruptions, what behaviors were most often missed, and what coaching themes appear in the summaries?

If the report may be large, ask for the count before requesting the full records:

Check the size: For De-escalation from 2026-06-29 through 2026-07-29, tell me how many completed simulator transcripts exist. Do not return the full records yet.

Use the exact Interaction Story name returned by the discovery prompt and provide dates when possible. The report includes completed, non-dry-run simulations created in that date range. See AI Agent Trainer for creating and running stories.

Send an email​

If the assistant's Send Emails capability is turned on, you can ask it to send an email right from the conversation — no workflow needed. Give it the three things it needs, in this simple format:

Send to: jane@yahoo.com Subject: good times Body: this is the template the user needs to use to send the email through the AI prompting.

That's the whole template — a Send to address, a Subject, and a Body. Fill in your own address and message and send it as-is, or wrap it in a sentence such as "Please send this email:" followed by the three lines above. You can also add a Cc or Bcc line the same way if you need to copy other people.

Before this works, two things need to be turned on:

  1. Send Emails must be checked on the Capabilities tab of the AI Profile or Chat Interaction you're talking to (under the Communications group). See AI Capabilities for exactly where that checkbox lives and who can turn it on.
  2. Email delivery must be configured for the site under Settings -> Email, so WinnerWare has somewhere to actually send the message from.

If you ask the assistant to send an email and nothing happens, check both of those first — a missing capability or missing email setup is the most common cause.

Build a workflow email​

The Workflows module lets the assistant create the email and notification steps inside a workflow. Name the workflow, identify the users who receive the message, and provide the subject and body. Use the users' IDs when you create an Email Specific Users task.

Create a Workflow named VPN Readiness Reminder. Add an Email Specific Users task that sends an email to user IDs user-001,user-002. Use the subject Complete the VPN readiness survey and this HTML body: <p>Please complete the VPN Readiness Assessment by Friday.</p>.

How it maps: the workflow receives an Email Specific Users task with the selected user IDs, subject, and HTML body. Choose the workflow trigger and schedule in the workflow editor, then test it with real users before enabling it. This creates an email workflow; it does not yet schedule or email Survey report data.

Build an Agreement​

The Agreements module builds Agreement content items users must accept. Name the agreement, set its version, and say when acceptance is required:

Create an Agreement titled Acceptable Use Policy with the body text I provide. Set it to version 1, require users to accept it at sign-in, and make it effective today.

How it maps: version, "required at sign-in", and the effective date fill the Agreement metadata part. Increment the version later when the terms change so re-acceptance is required. See Agreements.

Build an AI Chat Search Form widget​

The AI Chat module builds an AI Chat Search Form widget — a search box on a page that an AI profile answers. Name the headline and the profile that should respond:

Add an AI Chat Search Form widget with the headline "Ask our Knowledge Base" that is answered by the AI profile named Support Assistant, and show the top 5 suggested keywords.

How it maps: the headline, the answering AI profile (by name), and the suggested-keyword count fill the AI Chat Search Form part. See AI Chat.

Build a Knowledge Base Article​

The Knowledges module builds Knowledge Base Article content items. Every article must be assigned a category before it can be published, so include the category in your prompt:

Create a Knowledge Base Article titled How to Connect to the VPN with the body I provide, and file it under the Networking category. Optionally add a tutorial link to our setup video.

How it maps: the title and body fill the article, the category fills the required classification part (a term in the "Knowledge Base Article Categories" taxonomy — add the term first if it doesn't exist yet), and an optional tutorial link fills the tutorial part. See AI Knowledges.

Query survey report data​

Once surveys have responses, you can ask the assistant to pull the Interview Data report and reason over it. The assistant lists surveys and fetches the raw per-respondent answers, then does any totals or averages itself — you do not need to describe columns or formats:

Discover: What surveys can I run a report on?

Pull a report: Pull the Interview Data report for the VPN Readiness Assessment survey from 2025-01-01 to 2025-12-31 and summarize the responses.

Ask for numbers: For the VPN Readiness Assessment survey last month, how many people responded and what's the average home internet download speed? Break it down by work location.

The assistant receives one row per respondent with each answer's text and, where applicable, a numeric value — then computes counts, averages, and groupings from that raw data. The assistant uses the built-in List surveys and Query interview data report tools. They are not on the Capabilities tab; any assistant whose deployment has Tool calling turned on can use them. These tools return every completed answer in the date range, so publish such assistants only to people who are allowed to see survey answers. See Reporting for how the survey reports are built.