Setting up AI variables
How AI variables write each message in Vetta agents and sequences: the AI Message Settings page, the prompt, the channel, the Input sources including custom CRM columns, and skill files.
Updated October 25, 2026 · 8 min read
An AI variable is a piece of a message that the AI writes for each lead: a whole email body, a subject line, an opening sentence, a short phrase like the lead's target market. You insert it in a step like any other variable, as {{aiMessage.first_email_body}}, and every lead gets their own version.
Everything about AI variables lives in AI Message Settings, and it works the same in agents and in standalone sequences.
Where to find AI Message Settings
- In an agent: open the agent's Settings tab and pick AI Message Settings.
- In a sequence: open the sequence's settings (the gear in the sequence toolbar) and pick AI Message Settings.
- From a step: in any email, LinkedIn or WhatsApp step, the Manage AI message button in the editor's toolbar (shown as AI message on when the step already uses AI) opens the same page.
Agents · AI Message SettingsThe page has two parts:
- Generate AI message for this sequence: Generate sets up one AI variable for every step in this sequence (a subject and a body for each email step, a message for each LinkedIn or WhatsApp step) and inserts them into the empty steps. It is the fastest way to start.
- AI variables: every variable in the sequence, with its name, its prompt and how many input sources and skill files it uses. New AI variable adds one; Configure variable opens one.
Configure a variable
Configure AI variable has five parts:
- Variable name: letters, numbers and underscores, starting with a letter. The
{{aiMessage...}}tag under it is what you insert in a step. - Written for: the channel (Email, LinkedIn or WhatsApp). For email, tick Also write the subject line to have the variable write a subject too, with an optional Subject prompt. Leave the subject prompt empty and the subject is derived from the message; it uses the same inputs and skill files either way.
- Prompt: what the AI should write for this message, per lead. Be specific about length, structure and what to leave out. For example: "Write the first outreach email body in our voice: open on the signal that surfaced them, one line on what we do for finance teams like theirs, one proof point, and a soft ask for 20 minutes. 80 words at most, no subject line and no sign-off."
- Input: what the AI knows about the lead (below).
- Skill files: the business context documents that set the voice and the facts (below).
Save variable keeps your changes.
To use the variable, open a step in the sequence and insert it from the AI message variables. It shows up in the step as a chip, and each lead gets their own version when the message is written.
Setting up an AI message variable, a part of a message written fresh for each person. On the agent, open Settings, then AI Message Settings, and click Configure variable. Pick the channel, write the prompt, and review Input: basic details, signals and previous AI messages are always on, Hyper-personalisation can be turned off for a cheaper prompt, and Show more adds CRM fields. Add business context as skill files, save, insert it in a step, then Preview from Leads to see credits.
Transcript
An AI variable is a part of a message that's written fresh for each person. This is the agent's sequence view, a simpler way into a sequence on the canvas. On your agent, open Settings, then AI Message Settings. Click Configure variable on the one you want to change. Pick the channel it's written for, and write the prompt: what the AI should write for each lead. Under Input, basic details, signals and previous AI messages are always on. Signals only exist for leads found by Lead Scout. Hyper-personalisation is the one you can turn off, for a shorter, cheaper prompt. Show more lets you add single fields from the CRM, like the company's tech stack. Under Skill files, click Add business context and pick the documents that guide the writing: tone, structure, phrases to avoid. Then save the variable. To use it, open a step in the sequence, and insert it from the AI Message variables. It shows up as a chip. To check the result, go to Leads, tick a lead and click Preview. You'll see the estimated credits before anything is written, and nothing is sent: the drafts are yours to review. Each quarter, Alex asks Claude for an opener variable. Claude writes it for ten leads and flags two. Claude, ChatGPT or any assistant you connect to Vetta works on your real workspace, so test one variable, then give it a bigger job each week. In the agent's settings, check the opener, then review the messages yourself.
Input: what the AI knows about the lead
Open Input to see the sources sent with every lead:
| Source | What it sends | |
|---|---|---|
| Basic details | Who the lead is and where they work: name, job title, email, LinkedIn, city, and the company's name, industry, size, website and address | Always on |
| Hyper-personalisation | Enrichment notes, when the lead has them: work history, post history, reactions and comments, company posts and job search | On or off |
| Signals | Why this lead was picked up: the post, comment, hiring event or job change that surfaced them. Only leads found by Lead Scout carry signals; for imported or hand-added leads nothing is sent | Always on |
| Previous AI messages | What has already been written for this lead in this sequence, so a follow-up does not repeat the opener. Only variables that come before this one are included, since later ones have not been written yet | Always on |
Turn Hyper-personalisation off for a shorter, cheaper prompt that sticks to the basics.
Passing custom CRM columns
Click Show more to pick individual fields from the lead's record and everything related to it: Person, Company, Linked Opportunities and the other records related to the lead. Expand a section and switch on the columns you want, including your own custom fields, such as Hiring roles or Tech stack on the company. The person and company essentials are left out here, since Basic details already sends them.
This is how you hand the AI data that only your team has, for example a segment, a qualification note or a field filled by a workflow.
Skill files
Open Skill files to see the documents attached to the variable, then Add business context to pick more in Select business context files: tick the files and Confirm. Skill files are the business context documents of your workspace, so one file (your voice, your proof points, your pricing) can guide many variables. Editing them is covered in Tuning AI messages with skill files.
Tips
- One job per variable. A variable for the opener and another for the body are easier to tune than one that does everything.
- Put the rules in the prompt, the facts in the skill file. The prompt says what to write; the skill file says how you sound and what is true about your product.
- Preview before you send. Generate a preview for a few leads after every change: see Previewing and regenerating messages.
When the messages are written
AI variables are written for each lead just before their first message goes out, and then kept for the rest of the sequence. So a change to a prompt, an input or a skill file applies to every lead whose messages have not been written yet. Messages already written, including the ones you previewed, stay as they are until you regenerate them in the preview, and you can edit any step there that has not been sent yet.
Writing AI messages uses credits. If your credits run out, writing pauses and picks up again after a top-up or your monthly grant.
Common questions
Where do I find AI Message Settings?
In an agent's Settings tab, in a sequence's settings (the gear in the toolbar), or from the Manage AI message button in any email, LinkedIn or WhatsApp step. See Where to find AI Message Settings.
What's the fastest way to set up AI messages for a whole sequence?
Click Generate under Generate AI message for this sequence. It creates one AI variable for every step and inserts them into the empty steps. See Where to find AI Message Settings.
Can the AI write the subject line too?
Yes. In an email variable, tick Also write the subject line and add an optional Subject prompt; leave it empty and the subject is derived from the message. See Configure a variable.
How do I give the AI data from my own CRM fields?
Open Input, click Show more and switch on the columns you want from the person, the company, linked opportunities and related records, including custom fields. See Passing custom CRM columns.
Why doesn't the AI mention the lead's LinkedIn post?
Only leads found by Lead Scout carry signals. For imported or hand-added leads nothing is sent in the Signals input, so there is no post or comment to quote. See Input: what the AI knows about the lead.
I changed the prompt. Why did the messages stay the same?
AI messages are written just before each lead's first message and then kept, so a change reaches only leads whose messages have not been written yet. For the others, regenerate in the preview. See When the messages are written.
Do AI variables use credits, and what happens if I run out?
Yes. If your credits run out, writing pauses and picks up again after a top-up or your monthly grant. See When the messages are written.
How does the AI avoid repeating the opener in a follow-up?
The Previous AI messages input sends what has already been written for the lead in this sequence, so a follow-up knows what the opener said. Only variables that come before it are included. See Input: what the AI knows about the lead.
Key outcome
A prompt, the right inputs and a good skill file are what make an AI message read like it was written for one person. Set them up once per variable and every lead the agent or sequence reaches gets a message built from their own data.
Still stuck?
We reply the same business day, 9:00 to 17:00 CET, Monday to Friday.