> For the complete documentation index, see [llms.txt](https://learn.withessential.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://learn.withessential.com/administration/using-the-essential-assistant/asking-the-assistant-to-do-things-actions-you-confirm.md).

# Asking the Assistant to do things: actions you confirm

Beyond looking things up, you can ask the Essential Assistant to *do* things — create a task, schedule a visit, draft and send a quote, add a contact, update a customer's notes, update a recurring service or record a service completion, kick off a workflow. The Assistant never acts on its own: it proposes what it wants to do, you confirm it, and only then does it run.

## How it works

1. Ask in plain language, for example "create a follow-up task for Bob to check the panel next week" or "send the quote to Bob and add a note on the customer".
2. The Assistant replies with an **action card** in the chat listing each action it wants to take: what it is, a one-line summary, and the values it will use — plus a before-and-after when the action edits an existing record.
3. Review the card, then click **Confirm** to run it or **Reject** to discard it.
4. Once it finishes, the Assistant follows up in the chat with what happened.

When an action creates a quote or invoice, its completed card includes a link to open the new document. The link stays in the conversation when you reopen it. For a duplicated quote, it opens the new copy.

Nothing on your records changes until you click Confirm.

## What you can ask it to do

The Assistant only offers actions your role already allows, so what you see may be a subset of the list below. It can also run one of your business's published automations for a customer when Workflows is turned on, and change your own user settings.

Wherever custom fields are involved, the Assistant works from your business's real list. Ask it to set one and it matches the field you name against the fields that actually exist — their types and, for dropdown fields, the exact options — rather than guessing. When it creates a customer, property, contact, asset group, or quote, it also knows which custom fields your business marks as required: it asks you for those values up front and fills them in once the record exists, or tells you plainly when that field can only be filled in on the record's own page. An empty custom field never blocks creating the record — but the Assistant won't leave a required one blank without saying so.

### Customers, contacts, properties and tasks

Create a customer, property, or contact; update their details and notes, and add a contact's customer- or property-level role; archive any of them. Create a task in a task folder, or rewrite its notes — moving an existing task to a different status is offered only from a bulk review, so for one or a few tasks it will point you at the task itself instead. Post a comment on a customer's timeline, add a customer custom field, or set its value. Send a one-off email or text message you compose to a customer's contacts.

It can also move a property onto a different customer, carrying its subscriptions, quotes, and invoices across. This is the most consequential action in the list: if the property was the customer's last one, that customer is deleted, and any of its contacts that don't move across are deleted with it. The Assistant asks you to confirm the deletion and say which contacts should follow before it offers the move — it won't decide either of those for you.

For billing, it can give a property its own billing profile — payment terms, autopay, tax exemption, and billing address — instead of inheriting the customer's, change that profile, or remove it so the property goes back to inheriting.

While you're working a task, it can save a property's access details (gate code, lockbox, who to ask for) and log that on the task at the same time. It can also record a completed step on a task's timeline, which is a note that something happened — it doesn't move the task along or notify anyone.

On the timeline, it can rewrite or delete a comment — but only one you posted yourself; a colleague's comment has to be changed from the timeline. Deleting can't be undone from the comment itself, but it doesn't erase anything: the timeline keeps a "Comment deleted" entry that still shows the original text, edit history, and attachments, so ask it to edit the wording instead when that's what you mean.

Finally, it can add, tick off, and delete to-dos on your own private list. Those are yours alone — no customer sees them, and they don't put work on anybody else. Because nothing can look the list up, this only works for a to-do created earlier in the same conversation; for anything else, it points you at your to-do list instead.

### Scheduling, visits and assets

Schedule a visit at a date and time, move an existing visit, or add an estimated visit to the backlog, and update a visit's technician or scheduling instructions. Delete a visit that hasn't happened yet.

Add an asset group to a property, rename it, or make it inactive so it stops being serviced; set an asset group's custom fields. Delete an asset group when you really mean to remove it and everything under it.

It can also set and correct the attribute values recorded on a group's assets — fix a serial number that was typed wrong, fill in a value a unit is missing, or clean up formatting across the whole group ("the panel IDs should all start with FP — uppercase the lowercase ones"). The Assistant reads the group's stored values first, so the change lands on the exact record you mean, and each change appears on the confirmation card in plain language — which unit, and what it will say. Adding a brand-new asset to a group is still done from the group's page, where its type is picked from the group's configured types.

Create or update an asset group's recurring service schedule — its service, cadence, visit type, or last service date — record a service completion for a single asset, delete a schedule, or switch every schedule on a group between automated outreach, a custom task, and neither.

A batch of visits created together — its window, who can self-book it, and cancelling what's still open — is edited from the batch itself in the scheduling area; nothing in chat can look up a batch's id, so the Assistant points you there rather than asking you to supply one.

### Quotes, invoices and jobs

Create a draft quote or a draft invoice with line items, send either one to a contact, and void a quote or an invoice. Editing the contents of a draft quote or invoice is a separate flow — see [Editing a quote or invoice through the Assistant](/administration/using-the-essential-assistant/editing-a-quote-or-invoice-through-the-assistant.md).

For a quote that has expired or is about to, it can set a fresh validity window, which is how an unsigned quote is brought back to life. The new window counts from today, not from the quote's original date.

On invoices, it can finalize one — locking it against further edits and syncing it to your accounting system. That can't be undone, so the card says so before you confirm. Once an invoice is paid, it can email the payment receipt to contacts on the account or to an address you give it. It can't create a contact as part of sending: the recipient has to exist already, or be a plain email address.

Quotes, invoices and jobs can be linked so the work and the paperwork stay connected. The Assistant can set which jobs a quote or an invoice is linked to, or attach one job to several quotes or invoices at once. Linking never changes a line item or a price — it only records the relationship, and every job has to be on the same property as the document.

On a job, it can set custom fields, choose the stock location it draws from, and adjust the parts recorded against it. Parts move by the size of the change rather than by a new total, so two people editing the same job don't overwrite each other. It can't change a job's manual added revenue — that feeds your margin figures and stays in the app. It can also delete a job, which is permanent; it only offers that when you've clearly asked for that particular job to be deleted, and the job's saved details remain visible afterward on its property's timeline.

Finally, it can email a work acknowledgment to a contact or address. It can't correct one's answers from chat — there's no way to read what's already on an acknowledgment, so it points you at the acknowledgment itself for that, and signatures can never be changed this way.

### Pricing, items and inventory

Set or clear a price override for a pricebook item — for a whole customer, or for one of their properties — and apply a set of overrides in one go when you're reworking a customer's pricing. A batch is all-or-nothing: if one line can't be applied, none of them are, so you never end up with half a pricing profile.

On the pricebook itself, it can add a service or physical item, change an existing item's price, name, SKU, type, tax setting or cost, and remove an item you no longer sell. Two things to know before you ask: pricebook changes are business-wide and apply to future quotes and invoices, not to documents that already copied the old price; and removing an item takes it out of every picker but leaves it on the quotes, invoices and purchase orders that already reference it. A removed item's SKU stays reserved, and adding an item with that SKU again brings the original back.

It can also create a reusable discount, either a flat amount or a percentage. That adds the discount to the picker for future use — it doesn't discount any particular quote or invoice.

For stocked items it can rename an item or change its SKU, and set or clear the reorder level that flags it as running low. Changing a reorder level can start an automation, just as a stock movement can.

For stock itself, it can move quantities between your inventory locations, take stock out of inventory, or bring stock in, and it can correct an item's on-hand count at one or more locations after a stock take. Tell it either the correction ("we're three short") or what you counted ("there are 40 on the shelf") — for a count it works out the difference against the live figure at the moment you confirm, so a number quoted earlier in the conversation can't throw it off. Locations are matched by name; if it doesn't recognise the name, it tells you which locations exist rather than guessing.

Finally, it can raise a purchase order to one of your suppliers, change an order's notes, expected delivery date or supplier, close it (the order becomes **Void**), and record what arrived. A few things to be clear about here. Creating a purchase order records what you ordered and sends nothing to the supplier — actually placing the order is still on you. Suppliers are matched by name from your existing list, and it never creates one. Receiving adds stock at your default inventory location, so if a delivery went somewhere else, ask it to transfer the stock afterwards. Receiving is cumulative and the Assistant can't take it back: recording the same delivery twice doubles it, and the fix is a stock adjustment. Closing an order is final for receiving — a voided order can't be received against, including anything still outstanding on it.

Custom price overrides and discounts need neither inventory nor a setup permission. Adding, changing, or removing a pricebook item needs the Set up Items permission. The inventory module itself is only needed for moving or adjusting stock, purchase orders, and renaming a stocked item, changing its SKU, or setting its reorder level. The Assistant only offers what your role already allows.

### Inspection reports

Start a new inspection report on a visit or a property, choosing which form it uses. The report is created empty and waits for whoever is doing the inspection. If the form supports carrying the previous report's answers forward, you can ask for that too.

It can also duplicate an existing report into a new draft, copying the answers, deficiencies and signatures across. That is usually the right move when a report needs redoing, because the original stays intact as the record of what was filed.

The Assistant can also fill in and correct what a report records — the same edits the person filling it in makes. Answer a question on the report or on one of its assets, set a pass or a fail, add an asset to a list or duplicate one several times, skip or remove an asset, reorder them, pass or skip a whole selection in one go, add rows to a repeatable list section, raise or edit a deficiency, and record a service completion date. Ask in plain language — "mark extinguisher 3 as failed, the gauge is low" or "pass everything on floor 2" — and it comes back as an action to confirm like any other.

Because a value on a report means little out of context, every one of those edits shows a **before and after** on the card, and a report that has already been submitted gets its correction paired with a re-submit. Both are covered in [Editing an inspection report through the Assistant](/administration/using-the-essential-assistant/editing-an-inspection-report-through-the-assistant.md), along with putting a photo you've attached onto a report.

Once a report is finished, the Assistant can submit it. Submitting is what turns a draft into the customer's record: it writes the assets back to the asset group, saves the deficiencies and closes out any this visit cleared, records the service completions, and renders the PDF. It can't be undone from the Assistant. Submitting does **not** email anything — if you want the report sent, ask for that as well and you'll get a second action for it, so filing a report and mailing it to a customer are always two separate confirmations. Emailing a submitted report to a contact needs inspection report emails turned on for your business.

Deleting a report is the one action here marked Highest risk. It hides the report and cannot be undone. For a submitted report it does not reverse service completions, recurring-service history, deficiencies, its follow-up task, job files, emails, timeline events, or compliance filings. Because a submitted report may be the customer's compliance record, the Assistant only offers deletion when you have clearly asked for that particular report to be deleted, never as tidying up. Submitted deletion also requires the **Delete Submitted Inspection Report** permission.

On deficiencies, it can mark one resolved, reopen one that was closed too early, and record that an item was retested without changing whether it is still open. It can also clear a deficiency, which takes it off the open work list without claiming it was fixed, and restore it later. Clearing and resolving mean different things and the Assistant uses the one that matches what you said. Both only work on deficiencies with advanced tracking switched on.

Two boundaries are worth knowing. The Assistant never captures a signature: signature boxes stay with the person signing, on the report itself. And it doesn't change the report **form** — adding, renaming, reordering or removing the questions on a template is report setup, handled by Essential through the **Talk to a human** tab. Ask it to add a question and it will point you there rather than editing the form under you.

## Confirming only part of a batch

When the Assistant proposes several actions, each row has its own checkbox and they all start checked. Uncheck anything you don't want and confirm the rest.

If dropping an action would defeat what you asked for — approving "draft the quote" but not "send it", for example — the Assistant says so on the card. Some of those warnings are advisory and you can go ahead; when the remaining actions genuinely wouldn't do what you asked, the Confirm button is blocked. Check the action back on, or ask the Assistant for a different plan.

For a large set of record changes, the chat shows a compact **Bulk changes** card instead of hundreds of action rows — but only when every action type in the set supports a bulk review; actions like sending invoices or quotes, creating tasks, or rescheduling visits always use the ordinary confirmation card instead, however many there are. Choose **Review changes** to open the full review page. Each table shows the action it will run, its total and selected counts, and paginated record details. All rows start selected; you can clear a whole table or choose individual rows on any page before confirming. Your selections stay in place as you page or sort, and Essential warns before you leave an unconfirmed selection behind.

After confirmation, the review page follows the batch while it runs. The selection column becomes an outcome column, and the table and counts show which rows completed, failed, were skipped, or were not selected. **Back to conversation** returns to the chat that created the review.

## Risk levels

The most consequential actions stand out at a glance with a small warning triangle on their row — yellow for a high-risk action, red for the highest-risk ones — while routine actions like adding a note or creating a task carry no marker at all. A handful of deletions (a timeline comment, one of your own to-dos, a pricebook item, or a property's own billing profile) also carry no triangle, since the confirmation card itself is considered enough of a safeguard for those. Hover over the triangle to see the risk level spelled out. The marker is there to help you review; it doesn't change what you're allowed to confirm.

## Emails and texts

A one-off email or text you asked it to compose shows the finished message on the card — recipients, subject, and body — exactly as it will go out. Sending a quote or invoice shows the recipients and body but no subject line, and sending a submitted report shows who it goes to with a note that the body uses your saved report email template rather than model-written text. A payment receipt, a work acknowledgment email, and an automation's own sends don't show a preview at all. Read whatever preview is shown before you confirm; there is no recall once something is sent.

## Before and after on a record edit

Inspection-report edits change values that mean nothing read on their own — a pass or fail on a question, a reading, a service completion date. For those the card shows what each change does: which question or asset it lands on, what it says now, and what it will say once you confirm. Other kinds of edits don't carry this before-and-after treatment.

Anything the change takes away is called out in red as cleared or removed. Those lines are never folded away: when an edit touches more than a handful of things, the rest collapse behind **Show N more changes**, but every removal stays on the card.

The preview is worked out against the record as it stands when the Assistant proposes, so what you see is what runs. If the record moves in between — someone in the field saving the same report while you read the card — the action stops rather than running against something you never saw, and tells you so. Ask again and the Assistant re-reads the record and proposes afresh.

## Actions run as you

Confirmed actions run under your name, with your permissions, exactly as if you had done them in the app yourself. They appear in timelines and history attributed to you, and anything that normally follows from that action — automations, notifications, outreach — happens too.

That also means the Assistant can only do what your role already lets you do. If you don't have permission for something, it won't offer it.

## While actions are running

After you confirm, the card shows progress for each action. You can close the panel or move to another page — the actions keep running on Essential's servers, and the card shows the outcome next time you open the conversation.

Actions run in the order listed. If one fails, the Assistant stops there rather than pressing on: everything before it stays done, the failed action shows why, and the rest are marked skipped. Ask the Assistant to try again once you've fixed the cause.

> **Where to find it:** Open the Essential Assistant panel from anywhere in the app and ask it to do something. An ordinary action card appears in the chat with Confirm and Reject buttons; a large record set appears as a Bulk changes card with a Review changes link.

## Related guides

* [Using the Essential Assistant](/administration/using-the-essential-assistant.md)
* [What the Essential Assistant can and can't do](/administration/using-the-essential-assistant/what-the-essential-assistant-can-and-cant-do.md)
* [Editing a quote or invoice through the Assistant](/administration/using-the-essential-assistant/editing-a-quote-or-invoice-through-the-assistant.md)
* [Attaching files to the Assistant](/administration/using-the-essential-assistant/attaching-files-to-the-assistant.md)
* [Roles and permissions: how access works](/administration/basics/roles-and-permissions-how-access-works.md)


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://learn.withessential.com/administration/using-the-essential-assistant/asking-the-assistant-to-do-things-actions-you-confirm.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.