A play produces poor or inconsistent output

When a play generates outreach that misses, the cause is almost always in the play’s instructions, and the fix is a small change you make in the play editor. This page lists the symptoms reps report most often, what causes each, and where to fix it. For how to write a play well from the start, see Write effective play instructions.

The output reads like generic, templated copy

Cause: too much of the message is left for the agent to write, and too little is your own fixed copy, so the output drifts toward a bland average.

Fix: hardcode the lines that carry your message (your opener, your logic, your call to action) as fixed text, and let the agent personalize only the spots that must change per prospect. See Decide what to fix, and what to let the agent write.

The output is inconsistent from one contact to the next

Cause: the agent-written parts have no length guidance, so each run comes out a different shape and size.

Fix: give every agent-written spot an explicit length, stated at the start of the instruction. See Give every agent-written spot a length.

The message reads disjointed, like stitched-together fragments

Cause: there is no connective tissue between parts, so the opener and the next line don’t refer to each other.

Fix: tell each part after the opener to start by connecting the idea above it to why it is relevant to this prospect. See Connect each part to the last.

Prospects are not replying

Cause: the call to action asks the prospect to confirm interest (“is this relevant?”) instead of offering them something.

Fix: rewrite the call to action to offer something specific and useful for replying, and in a sequence escalate that offer step by step. See Write calls to action that earn a reply.

The copy states things that are not true about the prospect

Cause: a line that depends on research has no fallback, so when the research is thin the agent fills the gap with a guess.

Fix: add a fallback line to every research-dependent spot and tell the agent not to invent specifics. See Always give a fallback. A play that genuinely has nothing to say returns a Not applicable status with a Reasoning note rather than inventing one, see Generate plays for a contact.

The message shows labels like “Opener” or “Insight” above the paragraphs

Cause: section headings were placed inside a single message, and the agent treats them as content and prints them.

Fix: write a single message as one flowing note with no headings inside it, and tell the agent not to add any labels to the output. See Keep one message as one message.

Cold emails are landing in spam or junk

Cause: links in a cold email raise spam-filter risk sharply, especially for senior finance, operations, and executive contacts.

Fix: remove links from cold emails, state the benefit in plain text, and offer to share more if the prospect replies. See Keep links out of cold emails.

A play comes back empty or Not applicable

Cause: the contact has not been qualified and researched yet, a condition on the play excluded them, or the research was too thin and the play correctly declined rather than guess.

Fix: qualify and research the contact first so the agent has context to write from, check any conditions on the play, and make sure research-dependent spots have fallbacks. See Run a Qualification agent, Run a Research agent, and Generate plays for a contact.

A play that hit a technical problem while generating does not come back empty: it comes back with a Failed status and a technical-error reason, so you can tell “nothing to say about this contact” apart from “this one did not run”. Re-run a failed play, and if it fails again, raise it with support rather than editing the instructions - a failure of this kind is not an instruction problem.

A step in a sequence came back empty

Cause: the run finished but did not produce copy for that one step. Every step your play defines keeps its own position in the output, so a step with nothing in it shows as an empty step at its own number rather than disappearing and pulling the later steps up. If you see an empty Step 2 in a four-step play, the play still has four steps and the second one did not come out.

Fix: regenerate the play for that contact, see Generate plays for a contact. A one-off gap usually fills on the next run. If the same step keeps coming back empty, the instruction for that step is the place to look: give it an explicit length, and a fallback for anything it draws from research, so it always has something it is allowed to write. See Give every agent-written spot a length and Always give a fallback.

Do not read an empty step as the play being shorter than you built it. Count the steps against the play itself before you send: the numbering is the play’s, not a renumbering of whatever came back.

The message references a conversation that never happened

Cause: almost always the opposite of what it looks like. The agent writes prior interactions only from what is logged in your connected CRM, and only when the play asks for it, so a wrong reference usually traces to what synced in - an interaction logged against the wrong contact, or a note whose content does not match what actually happened.

Fix: open the contact and read the interaction history behind the claim, see the contact detail page. If the underlying record is wrong, correct it in your CRM and let it sync. If a play should not be reaching for history at all, remove the instruction that asks for prior interactions, see Say when the play should use prior interactions.

Write effective play instructions - the full craft, and the fix for most of the issues above.
Play Copywriting agent - what a play is and what it draws on.
Generate plays for a contact - running and regenerating plays on a contact.
Review agent outputs - where generated plays land and how to check them.