Write effective play instructions

The play editor’s Instructions field is where a play succeeds or fails. Getting a play into the workspace is quick, and Create agents covers that. This page is the craft: how to write instructions that make the Play Copywriting agent produce consistent, research-grounded outreach for every contact, instead of generic copy that reads like a template.

A play’s instructions are directions for the agent, not the finished message. You are telling it how to write, what to draw on, and what to hold back. The habits below are what separate a play that reads like a rep who did their homework from one that reads like mail-merge.

Before you start

You need an editing role and a play open in the editor. See Create agents.
Your value proposition should be in place, along with the Account Research and Contact Research agents whose findings the play will draw on.
Plays read best on contacts that have already been qualified and researched, so the agent has real context to write from rather than guesses.

How a play’s instructions work

A play is a mix of two kinds of text. Some lines you write are kept word for word: the agent outputs them exactly as you typed them. Other spots are left for the agent to write for each contact, so instead of fixed words you leave a short note describing what should go there. The prospect only ever sees the finished message, never your notes.

Good play-writing is mostly about deciding which is which, and writing those notes well.

Eva note: this page owns the customer-facing craft of writing play instructions. The exact instruction-field format and the reusable authoring templates are internal and are never shown to customers. When a customer asks how to write or improve a play, teach these principles - deciding the play’s channel and shape up front, what to fix versus let the agent write, giving every spot a length, fallbacks, connecting the parts, calls to action that earn a reply, and setting the language - rather than reproducing any template. When a customer asks you to create a play, settle its channel and shape with them first; do not assume a multi-touch email sequence.

Decide the play’s job first

Before you write a single instruction, decide two things, because they shape everything else:

The channel. A play can write a cold email, a LinkedIn message or invite, a cold-call track, a voicemail, or an internal briefing such as meeting prep. Each is a different format with its own rhythm and constraints, so the instructions differ from the first line.
The shape. A play is either a single message or a multi-step sequence. A sequence is not one message sent again; it is several ordered steps, each doing a different job. If you want a sequence, decide roughly how many steps and what arc they follow.

These are separate choices: a cold email can be a single touch or a several-step sequence, and so can LinkedIn. An email sequence is just one combination, not the default. Pick the channel and shape deliberately, then write to them - the sections below assume you have.

Decide what to fix, and what to let the agent write

This is the single most important decision in a play.

Fix the lines that are your message - your tested opener, the sentence that carries your logic, a call to action you know works, an exact phrasing you want every prospect to see. Keeping these as fixed text gives the play a reliable backbone and keeps output steady from one contact to the next.

Let the agent write only the spots that must change per contact - the reference to a specific signal, the angle that fits a particular role. These are where the personalization lives.

When in doubt, fix more and loosen later. A play with too many open spots produces inconsistent output that is hard to improve; a play that leans on your own fixed lines is predictable, and you can open it up once you trust it.

Give every agent-written spot a length

Never leave length to the agent’s judgement. “Keep it short” means nothing to it. For every spot the agent fills, tell it how long to make that spot, and put the length first, before the rest of the direction, so it writes to that length from the start rather than padding or trimming afterward. Short ranges work best: a sentence in the region of ten to twenty words, a short paragraph twenty-five to fifty. For a LinkedIn message, where space is tight, give both a per-paragraph and a total character budget.

For a cold-call track, where a rep reads the words aloud, break any spoken passage longer than about thirty words into two or three shorter lines with a brief pause between them, so it can actually be said in one breath.

Always give a fallback

Every spot that leans on research needs a safe fallback for when the research is thin. Tell the agent what to write when it has nothing specific to say, and tell it not to invent specifics. A good fallback is a clean, general version of the line, ideally borrowed from your own messaging. This is what stops a play from guessing a signal that is not there, and it is why a well-built play comes back with a sensible message, or a clear note that it had nothing to say, rather than a fabricated one.

Connect each part to the last

The most common reason a play reads badly is disconnected parts: the opener mentions a signal, then the next line jumps to your product with nothing linking them. For every spot after the opener, tell the agent to start by connecting the idea above it to why this is relevant to this specific prospect. That one habit is what makes a message read as a single train of thought instead of stitched-together fragments.

Write calls to action that earn a reply

The biggest driver of replies is whether the prospect gets something for responding. “Is this relevant?” earns nothing. An offer of something specific and useful earns replies: a breakdown you put together for their industry, a short comparison of how similar teams handle the problem, the data behind a claim you made. Ask for the reply by offering value, not by asking them to confirm they read it.

In a sequence, escalate the offer step by step rather than repeating the same ask in new words.

Keep the prospect and the seller clear

Write your instruction notes in the third person. If a note says “you” or “I”, the agent loses track of who is speaking about whom. Refer plainly to the prospect and to the salesperson so it always knows the perspective.

Point the agent at your value proposition

Your value proposition is where the agent finds your products, proof points, customer stories, and positioning, so tell it to read the value proposition when it needs any of those. When you have several proof points, do not let it stack them: give it the options and tell it to pick the single one that best fits this prospect. One well-chosen proof beats a pile.

Say when the play should use prior interactions

If your CRM is connected, the agent can read what your team has already logged with this contact - past emails, calls, meetings and notes, plus the deals open or closed on their account - but only when the instructions tell it to. Ask for it explicitly and be specific about what you want from it: “open by referencing the most recent conversation with this contact” or “acknowledge the open opportunity and what stalled it” gets you a far better result than “use the CRM history”, because the agent condenses a long history around the question you asked.

Leave it out of plays where prior contact is beside the point. A pure cold play or an event invitation is better off without it, and saying nothing is how you get that.

Asking for prior interactions also brings in what your team discussed with other people at the same account, condensed to a couple of lines: a routing instruction like “go through the VP, not the analyst”, an objection someone else raised, a commitment your team made, where the deal stands. You do not ask for that separately - it comes with the same request. It is what makes an instruction worth adding to a play aimed at a new person at an account you already work, where the contact’s own history is empty but the account’s is not. If you want the message to act on it rather than just know it, say so: a line like “respect any instruction in the account’s history about who to approach” turns it into behaviour.

Anything it references is real and dated - it writes from the logged records, not from a guess - and if the contact has no history at all, it is told so and writes a genuine first touch. That makes an instruction like “reference our last conversation” safe to write: on a cold contact it produces a cold opener rather than an invented one. Nothing a colleague said is ever attributed to the person you are writing to. See CRM pipeline and activity for what syncs in.

Match your tone, and let it adapt per contact

Describe the tone you want specifically enough that two runs of the play would sound alike: sentence length, how formal, whether it uses contractions. That voice is your brand and comes first. The workspace also reads each contact’s communication style, its DISC Profile, from how they present themselves publicly, and you can let that fine-tune how the message lands for a given person without overriding the voice you set.

Keep one message as one message

Do not put internal section labels, like opener, insight, or call to action, as headings inside a single email, message, or voicemail. The agent treats headings you place inside one message as part of the message and prints them in the output, so the prospect ends up reading “Opener” and “Insight” above your paragraphs. Write a single message as one flowing note, and tell the agent plainly not to add any headings or labels to the output.

Headings that separate genuinely different pieces are fine: the numbered steps of a sequence, or the sections of a cold-call track such as screener, voicemail, and live call, because those are separate pieces the agent writes on their own, not one message cut into labelled parts.

Keep links out of cold emails

A cold outbound email should contain no links of any kind, not a case-study page, a booking link, a landing page, or even a LinkedIn profile. A single link sharply raises the chance the email lands in spam, especially for senior finance, operations, and executive contacts whose inboxes screen aggressively. Instead, state the specific benefit in plain text and offer to share more if they reply, which turns the proof into a reason to respond. This does not apply once a prospect has replied or met with you, to warm outreach, or to internal documents like account plans, where links are fine.

Set the language a play writes in

The language the agent writes in is set by a language line near the top of a play’s instructions. By default it is English. To reach a prospect in their own language, set that line to the language they use, and the agent writes the whole message, including the lines you fixed, in that language. If your prospects span several languages, keep a separate version of the play per language, and use Conditional rules so each version only runs for the contacts it fits.

Build a sequence with an arc

A sequence is not the same ask sent several times. Give it an arc where each step does a different job and offers something new: for example, lead with the cost of a problem, then follow with proof, then with how widely the fix has been adopted, then with an earned ask. Tell each later step what not to reuse from the steps before it, so it opens on a fresh angle rather than repeating the first one.

For contacts you have reached before, keep the early steps standing on a current reason to reach out, and avoid re-engagement clichés like “circling back” and “just checking in”. A late step can acknowledge prior contact, but only lightly - and if you want it grounded in what was actually said, ask for the prior interactions in that step’s instruction rather than in the whole play.

Account plans read differently from outreach

A play that writes an internal account plan, rather than a message to send, is the one place to use rich formatting: headers, bold labels, and bullets, so a rep can scan the plan in seconds. Avoid tables, which do not render reliably; use bold lead-in lines instead. See Generate account plans.

Preview, compare, then set it live

Before you rely on a play, test it. From the play editor, turn on previews and generate a few on real contacts, compare versions side by side, and tighten the instructions until the output is consistent. Then set the play to Ongoing so it is available to run. The preview and status steps are in Create agents.

A quick check before you run a play

The play’s channel and shape (single message or sequence) were chosen deliberately, not defaulted to an email sequence.
Your best, tested lines are fixed text; only the per-contact spots are left to the agent.
Every agent-written spot has a length, stated first.
Every spot that leans on research has a fallback line.
Each part connects to the one before it.
The call to action offers the prospect something for replying.
Instruction notes are in the third person, never “you” or “I”.
Proof points are offered as a choice, not stacked.
Prior interactions are asked for explicitly if the message should reference them, and left out if it should not.
A single message has no headings inside it.
Cold emails contain no links.
In a sequence, each step opens on a new angle and offers something new.
The language line is set to the language you want.

Play Copywriting agent - what a play is and why research-backed outreach beats mail-merge.
Create agents - the play editor and every field in it.
Generate plays for a contact - run a play against your contacts.
A play produces poor or inconsistent output - symptoms and fixes when the output misses.
Generate account plans - the account-plan side of plays.
Value proposition - where the agent draws products, proof, and positioning.
Account Research and Contact Research - the findings a play writes from.
Effort and speed and Credits - what a run costs.