> For the complete documentation index, see [llms.txt](https://help.usejimo.com/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://help.usejimo.com/docs/integrations/mcp/what-to-ask.md). # What to ask? Talk about your project the way you would to a colleague. Use the names you see in Jimo: an experience's title, a segment name, an end user's email. The assistant picks the right tool. ## Get Insights | Ask | What you get | | ------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------- | | "How are we doing this month? What is expiring soon?" | A project overview: users, what is live, drafts, top performers, running experiments, activation. | | "Where do people drop off in the Onboarding Tour?" | The per-step funnel with the biggest drop-off point. | | "Is anything broken this week?" | Steps whose target element could not be found or was hidden, usually after a UI change. | | "What has seen, and where did she get stuck?" | One end user's story before a customer call. | | "Summarize this week's NPS comments." | The survey and NPS response feed, summarized. | | "What did the agent fail to answer?" | The Jimo Agent's needs-attention feed: unanswered questions, disliked answers, abandoned skill runs. | ## Operate | Ask | What happens | | ------------------------------------------------------------------------- | --------------------------------------------------------------------------------------------- | | "Draft a changelog post from these release notes, then show me the copy." | Creates a draft, invisible to end users, editable in the dashboard. | | "Publish it to staging first." | Shows a preview and asks you to confirm before anything changes. | | "Pause the Pricing banner right now." | Same rule: preview first, then your confirmation. | | "Launch the churn survey to Trial Users, ending Friday." | Sets the audience and schedule, then sets it live, with a live audience count in the preview. | | "Translate the Pricing modal into French and German." | Shows the languages it would write, then runs Jimo's smart translate once you confirm. | {% hint style="info" %} Every change is previewed before it happens, and nothing goes live without your confirmation. See [Permissions & Safety](/docs/integrations/mcp/permissions-and-safety.md). {% endhint %} ## Build and edit experiences With **Manage Experiences**, the assistant can draft every experience kind, wire experiences together, and edit them step by step, including once they are live.
From a draft to a live experience: draft, read back, edit, try it in your app, target, publish, watch, then fix and improve
| Ask | What happens | | -------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------- | | "Draft an NPS survey. Below 7, open the Talk to us modal." | Drafts the survey with its branching logic. A score that matches the rule launches the modal. | | "Start from the gallery template for a feature tour." | Lists the templates, drafts from the one you pick, and asks for the steps that need an element of your app. | | "Build a tour of the inbox: point at Search, then at Compose." | Drafts the tour with each step pinned to the element you captured in your app. See [Pinning steps to your app](#pinning-steps-to-your-app). | | "Make a Get started checklist. The first task opens the product tour." | Drafts the checklist, with a button that launches the tour and a task marked done once the tour is completed. | | "Add a task to the Get started checklist." | Inserts the task and leaves the others as they are. On a published checklist, it previews the change first. | | "Show the Pricing modal only on /billing, 5 seconds after the page loads, on desktop." | Sets where and when it shows. The preview says whether more or fewer people would see it. | | "Open the tour when users click Upgrade." | Sets an element-click trigger on the element you captured. | | "Apply our Dark theme to the Pricing modal and move the banner to the bottom." | Applies a project theme and sets the position. Applying a theme replaces styling made in the builder. | | "Tag every onboarding experience Onboarding." | Adds the tag, and creates it if it does not exist yet. | | "Set up an A/B test on the Pricing modal headline." | Creates variant B as a copy, edits its headline, sets the traffic split, and starts the test once the modal is live. | | "Send me a link to try the draft in our app." | Returns a signed preview link that opens the draft in your product, without the dashboard. |
Display rules: where an experience shows (pages, domains, devices), when it starts (page load, exit intent, tracked event, element click, On shared URL, On manual trigger, No automatic trigger), and how often it repeats
{% hint style="info" %} **An experience launched by another one** (a checklist task, a survey answer, a tour button) should have **No automatic trigger**: it then shows only when launched. **On manual trigger** is a different option. {% endhint %} ### Pinning steps to your app Tooltips, hotspots, cursors, and hints are pinned to an element of your product. The assistant never writes a CSS selector: Jimo writes it from an element you point at. 1. Open your product on a page where the Jimo snippet runs. 2. In the browser console, run `jimo.push(["get", "anchor:capture", [element]])` on the element, then copy what it returns. 3. Paste that text to the assistant with your request.
Pinning a step: capture the element in your app with anchor:capture, paste it to the assistant, and Jimo writes the selector
If the capture matches more than one element, capture something more specific, or add a `data-jimo-id` attribute to the element. When the assistant cannot use a browser, it can fall back on an element from a screen the Jimo Agent has already scanned. ### Reviewing a draft Before a draft is saved, the assistant reports warnings, such as a button that launches an experience end users cannot see yet. For tours, it turns a tooltip on a button or a link into a cursor: the element, the text and any step name you set stay, the tooltip title and Next button go, since a cursor has neither. It may also tighten the copy of a step. It always says what it changed. To keep a tooltip, ask it to switch the step back after drafting. ## Teach the Jimo Agent | Ask | What happens | | ----------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | "Teach the agent that exports are under Settings → Data." | The custom answer goes live for end users within minutes after you confirm. | | "Recrawl the docs site." | Refreshes an existing knowledge source now instead of waiting for the schedule. | | "Import the three onboarding docs from my Google Drive into the agent's knowledge." | Your assistant reads them with its own Drive tools and hands the text to Jimo. Jimo never gets access to Drive. | | "Train the agent on help.acme.com, only the /docs pages." | Previews the pages found in the sitemap, up to 500, and any existing source that already reads them, before anything trains. | | "Train the agent on our Zendesk help center." | Uses the Zendesk help center already connected in the dashboard, all articles or a selection. | A website source's settings cannot be edited in place. Add the new one, wait until training finishes, then delete the old one. ## Set Up and Run the Jimo Agent With **Build and run my Jimo Agent**, the assistant can take the whole Agent section: map your product's screens, train the agent on your help content, write the skills users need, test it the way users ask, put the launcher live, and review what happens once people use it.
It calls the same services as the dashboard, and a test question runs the real agent, so the answer matches what the widget would say. | Ask | What happens | | ------------------------------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | "Set up the Jimo agent for app.acme.com. Our docs are at help.acme.com. Users mostly need help with projects and invitations." | The assistant checks what is missing, maps the screens, previews the help center before training on it, writes skills and leaves them off, and tests a real question. It activates skills and adds them as launcher suggestions only after you agree. The launcher goes live only when you say so. | | "Teach the agent the project members screen I have open." | Writes a rule that covers every project, not only the URL that is open, and checks it against example URLs before saving. Once you confirm, it gives you a link to capture that screen's screenshot. | | "Write a skill that walks users through inviting a teammate." | Writes the skill, inactive, tests a couple of phrasings, and activates it once you agree. It can add a suggestion on the page that starts it. | | "How did the agent do this month? Fix what's worth fixing." | Reads usage and escalations against the previous period, then the disliked and escalated conversations, proposes a fix, and checks it with a test question. | {% hint style="info" %} Skills are created inactive, and putting the launcher live is its own confirmation. It is refused when your plan does not include the agent. See [Permissions & Safety](/docs/integrations/mcp/permissions-and-safety.md). {% endhint %} ### Screens Each screen the agent knows is a page in [Knowledge > Interface](/docs/experiences/agent/knowledge/interface.md). The assistant can add one in two ways. **From a description (Read all).** The assistant writes the page from a name, a description in your users' words, and URL rules. Once you confirm, it gives you a screenshot link: open it on that screen, with the Jimo Chrome extension or the browser's share-tab prompt. Until the screenshot is saved, setup still lists the page as missing one. **From a scan.** For a screen whose elements matter, the assistant hands you a scan link instead. It opens the page builder with **Scan and curate** or **Read all** already chosen, and you still confirm before anything is saved. * The Jimo snippet has to be on that site, or you use the Chrome extension. * The URL is filled in only for a site the project already uses. Otherwise you type it. * A scan saves a rule for that exact URL. The assistant then widens it to cover every record of the screen. {% hint style="warning" %} A Read-all scan of a page that already has a curated element list replaces that list. The scan link warns you before saving converts a page to the other mode. {% endhint %} ### Skills and Suggestions [Skills](/docs/experiences/agent/skills.md) are **explain**, **guidance** (walk the user through it), or **action** (do it for them). Instructions name on-screen items by their labels, and `{{email}}` stands for a value the agent asks the user for. Skill names are unique in the project, and changes to an active skill reach users within about 30 seconds. Suggestions are the chips above the launcher. A page has at most four active ones, a suggestion can start a skill, and `{{attribute|fallback}}` in its text is filled in for each user. ### URL Rules Every rule is checked against example URLs before it is saved. What it is matched on depends on where it is used: | Rule for | Matched on | | ------------------ | ------------------------------------------------------------------------------------------ | | A page | The path and query when the value starts with `/`, the full URL when it contains a domain. | | Launcher targeting | The path, query, and `#` fragment, never the domain. Limit domains with the domain filter. | | A website filter | Each page's path. | A rule written with a domain never matches launcher targeting or a website filter: write `/docs`, not `help.acme.com/docs`. A regex is written on the path, without `.` (use `[^/]+`, and `\x2e` for a literal dot). When a page is created from a URL, record ids (numbers, UUIDs, and 24-character hex ids) become a wildcard and slugs stay as they are. A screen that routes in its `#` fragment, such as `#/projects/123`, is written on the full URL instead, for example contains `app.acme.com/#/projects/`. Page names are unique in the project, and a new page that would load where another page already does is refused unless you say both should apply. ## Ready-Made Workflows Jimo also offers six prompts your assistant can list and run. They chain the right questions together and ask you for anything missing. | Prompt | What it does | | ---------------------- | --------------------------------------------------------------------------------------------------------------------------------- | | `build_tour` | Builds a tour with you: drafts the steps, pins each one to your app, and gives you a link to try it before anything is published. | | `release_announcement` | Draft, read back, revise, and publish an announcement, then teach the agent about the release. | | `weekly_digest` | Overview, top performers, activation, feedback, and agent gaps, written up as a digest. | | `cs_account_review` | Everything Jimo knows about one account or cohort, before a call. | | `set_up_agent` | Sets up the Jimo Agent from your product URL, help center URL, and what users need help with. | | `review_agent` | Reviews how the agent did over a number of days you choose, and proposes fixes. | ## Next Steps * [Tools Reference](/docs/integrations/mcp/tools-reference.md): the full list of what the assistant can do, grouped by permission. * [Permissions & Safety](/docs/integrations/mcp/permissions-and-safety.md): how previews and confirmations keep changes safe.