> ## Documentation Index > Fetch the complete documentation index at: https://docs.callprep.app/llms.txt > Use this file to discover all available pages before exploring further. # Build your first Autopilot > Step by step through the builder — who it contacts, where it reaches out, how it sounds, what it sends, and when it stops. Go to **Dashboard → Autopilot** and either **Start from a template** (see [Templates](/autopilot/templates)) or **Build from scratch**. This page walks through the second one, because that is the version where you see every decision. *** ## 1. Name Name it after **who it targets**, not what it sends — `Agencies — Poland`, not `Email sequence 2`. In three months this list will have eight rows and you will be reading it in a hurry. *** ## 2. Who should it contact? The filter reads your **HubSpot contact properties**: pick a property, an operator, and a value. | Operator | Use it for | | ---------------------------------------- | ----------------------------------------------- | | **is** / **is not** | Exact match. Several values means *any of them* | | **contains** / **doesn't contain** | Free-text properties | | **>**, **≥**, **\<**, **≤**, **between** | Numbers — a score, a headcount, a value | | **is known** / **is unknown** | Whether the property is filled in at all | Add up to three rules and choose whether **all** must be true (AND) or **any** (OR). **An Autopilot with no rules matches every lead.** The list marks it *Catch-all* and greys out everything below it — nothing further down will ever see a lead. Deliberately, that is a useful bottom-of-the-list Autopilot. By accident, it takes over your whole account. Properties that get filled in during onboarding are often empty at the moment the contact is created. That is expected and handled: the lead waits in the queue and is re-checked as its data arrives. See [Settings](/autopilot/settings#queue-wait-time). *** ## 3. Where can it reach out? Toggle the channels this Autopilot may use. Only channels you have actually connected can send. Under the channels is a line worth reading before you commit: **how many of this Autopilot's leads are reachable that way**. A LinkedIn step is useless to a lead with no LinkedIn profile, and a WhatsApp step is useless without a phone number in your CRM. This is where you find that out, rather than two weeks later. WhatsApp and Telegram only message leads who already engaged, unless you switch that off per channel. A sequence made **only** of those steps will wait indefinitely for a warm signal it cannot itself produce — pair them with email or LinkedIn. *** ## 4. How should it sound, and what is it for? **Tone** changes the voice and the greeting. It never changes what a message asks for. | Tone | Reads like | | ------------ | ---------------------------------------------------------------- | | **Friendly** | Warm and casual, like writing to someone you know a little | | **Direct** | The shortest version, no softeners | | **Formal** | Polite and professional, no contractions — for enterprise buyers | | **Founder** | Personal and honest, typed quickly by a busy founder | **Goal** is what every message works towards, and some goals need something from you before the Autopilot can be switched on: | Goal | Needs | | --------------------- | ------------------------------------------------------------------------------------- | | **Engage with leads** | Nothing — messages simply invite a reply | | **Meetings** | Your calendar link (Calendly, Cal.com, anything) | | **Self serve** | Your pricing page | | **Qualify** | One to three qualifying questions, in the order you want them asked — one per message | A goal without its input cannot be enabled. This is deliberate: a "book a meeting" sequence with no link is a series of messages asking for something the reader cannot do. **How did these leads come to you?** — one sentence, in your words (*"they registered for a trial"*, *"they downloaded the pricing guide"*). It becomes the **opening line of the first message**, which is most of the difference between a cold email and an obvious one. *** ## 5. The sequence Each touch is a **day**, a **channel**, and an **angle**. * **Day** — 1 is the day the lead is enrolled. Later touches are business days from there. * **Channel and kind** — email; LinkedIn *connection request* or *message*; WhatsApp; Telegram. * **Angle** — your instruction for what this message should be about. Not the final text: the message is written per lead from the angle plus that lead's research. * **Subject angle** (email) — only on the first email when threading is on, since the rest reply inside the same thread. * **Uses research** — which signals to build on: their post, their title, the company, the fit with your product. Leave it alone for a sensible default. ### Show example **Show example** renders a real message for a real lead of yours, through the same code path that sends, with your signature underneath. If you would not send that one, change the angle and look again. ### What the angle can and cannot do Your angle **outranks** the house writing rules — it is your campaign. Four things it cannot override, because they are not style preferences: * the length a channel physically allows * **inventing facts** — including a performance number that is not in your product description or in the research * the rule that a reshared post is not the lead's own words * the sender-identity block on commercial email Quote a sentence in the angle and it will be reproduced nearly word for word in **every** message from that step. That is a good way to pin an opening you like — and a good way to send four hundred people the same sentence. The builder warns you when a step does this. ### A note on LinkedIn steps LinkedIn works in two moves: a **connection request**, then a **message** once it is accepted. A message step waits for the acceptance on its own and gives up after a week rather than hanging forever. A second connection request in the same sequence can never do anything — LinkedIn allows one — and the builder says so. ### ✨ Build with AI **Build with AI** drafts the whole sequence from your product, channels and filter. Treat it as a first draft: read every angle, run **Show example** on at least the first two touches, and cut anything you would not send yourself. *** ## 6. When should it stop? | Condition | What it means | | ----------------------------------------------------------------- | ----------------------------------------------------------------------------- | | **They reply — on any channel** | Email, LinkedIn, WhatsApp or Telegram. Leave this on. | | **You log a connected call, or they book a meeting (in HubSpot)** | Someone on your side already picked this lead up. | | **LinkedIn invite accepted (you're connected)** | A choice, not a default — see below. | | **A CRM property changes** | Nominate a property and a value. Leave the value empty to stop on any change. | **An accepted connection request is a connection, not an answer.** Turn this on if connecting *is* the goal of the sequence — the acceptance is then counted as a win. Leave it off if you plan to say something afterwards, or the sequence will end exactly when it becomes useful. Full detail, including what deliberately does **not** stop a sequence, is on [What stops a sequence](/autopilot/stop-conditions). *** ## 7. Save and switch on Save from the bar at the bottom, then switch the Autopilot on in the list. From that moment, a **new** contact matching the filter is researched and enrolled within about a minute. Existing contacts are not swept up — the Autopilot acts on contacts created from now on. After a day or two, check: * the **counters on the row** — leads, active, responded * the **delivery chip**, which appears only when leads are sitting uncontacted * **[Activity](/autopilot/activity)** for individual leads and skip reasons *** ## Frequently asked questions Three or four over two weeks is a sequence. Twelve is a campaign nobody finished reading. Note that the last email is written to ask for nothing — it exists to close politely, so cutting it makes the sequence end mid-pitch. Yes, and the changes apply to messages that have not been sent yet. Leads already enrolled keep their place in the sequence. Deleting a step from the middle can leave already-scheduled messages without a matching angle — they fall back to a generic one rather than sending something written for a different step. It stops routing new leads and leads in flight are stopped, but everything it did stays: the leads, the messages, the analytics. Nothing is deleted, and a lead that already replied keeps its win. LinkedIn accepts one connection request per person. A second one in the same sequence can only ever be skipped, so the builder defaults a new LinkedIn touch to *message* once a request already exists. Yes. Channels belong to your account, not to one Autopilot, and the daily limits are per mailbox or per seat — shared across everything you run.