Fill your content calendar with the Content Agent

The Content Agent writes content for you. It drafts every piece with SEO/AEO research, revises it until it reaches your quality score, puts it in Review, and messages your reviewers in Slack. It never publishes on its own.

There are two ways to use it:

• Saved Background Genie Agents (Content Engine → Background Genie Agents): reusable profiles you assign directly to ideas, drafts, published pieces, or calendar days. Covered in Assign saved agents.
• The Content Agent workflow task: runs on a Timer and picks its own ideas. Covered in the walkthrough below.

Assign Saved Agents

Create an Agent

Open Content Engine → Background Genie Agents and click New agent, then choose Use a suggested agent or Build from scratch. Suggested agents (Long-Form Blog & Article Writer, LinkedIn & Social Post Specialist, Customer Case Study Writer, Competitive Battlecard & Comparison Agent, Email & Newsletter Copywriter, Whitepaper & Sales Enablement Author) can be created as-is, customized first, or tuned with Genie; ones you've already set up are hidden. Customizing or building from scratch opens the agent's own setup page, where Build with Genie can fill it out for you. Click any agent card to open its detail page, edit it, see recent runs, or archive it.

Enter a name and description (shown when teammates pick an agent to assign), then fill in the three sections below. Save anytime with Create agent / Save changes.

1. What it writes: the primary asset types the agent is responsible for and Require customer call evidence (only pick ideas backed by call transcripts, and write in your customers' own language: their terminology and the way they describe pain and value, not just pasted quotes). Audience, funnel stage, and angle come from the idea, draft, or calendar item the agent is assigned to; put any standing audience or "never write about" guidance in the Agent System Prompt.
2. How it writes: the Agent System Prompt (see Agent system prompt). Click Ask Genie for help and Genie asks a few questions about audience, voice, and structure, writes the prompt for you, then offers to fill out the rest of the agent. Below it are the agent's tools, summarized in one line with an expandable list, any custom tools you attach, and an optional Model & skills section to override the AI model and pin or exclude skills.
3. When it finishes: the quality score & review loop toggle (keep it on for long-form content; turn it off for single-pass social posts and emails). When it's on you also set the quality bar and tool budget (minimum score, max review rounds, max tool calls per round); see Quality score and review loop. Below that, Notify on completion has separate Slack and Email toggles for choosing who hears about drafts that land in Review. The Slack toggle only appears when Slack is connected; otherwise the form says so, and admins with integration access get a Connect Slack link to Settings → Integrations.

If you try to assign an agent before creating one, the assign modal links directly to the agent creation page. Anyone with Content Engine write access can create, edit, and assign agents.

Agent System Prompt

The system prompt is the standing direction the agent follows on every piece: its role, voice, audience, structure, and what to emphasize or avoid. It sits on top of your content skills, so keep brand-wide rules in skills and put what's specific to this agent in the prompt. For example: "You are a senior B2B writer for RevOps leaders. Open with a concrete operational problem, use short paragraphs, cite customer call evidence, and end with one actionable takeaway."

Agent Tools

While drafting and revising, the writer can call these tools:

• Content brief (generate_content_brief): builds an evidence-backed outline before long-form drafts.
• SEO / AEO research (research_seo_context): pulls live search and answer-engine context for the topic.
• Call transcript search (search_content_transcripts, fetch_content_transcript): finds customer quotes and proof points.
• CRM records (fetch_content_records_by_id): reads linked accounts, contacts, opportunities, and activities.
• Related content (find_related_content_assets): finds existing library pieces for internal links and consistency.
• Skills lookup (search_skills, read_skill): reads organization skills for product and positioning facts.
• Image generation (generate_image): creates header and OG images with alt text.
• Canvas editing (edit_canvas, search_canvas): writes the draft and makes targeted revisions.

Custom Tools

In Prompt & tools, click Add tool under Custom tools to attach workflow tools you've built in Automation (Workflows → Tools), such as a pricing lookup or an internal knowledge search. The writer calls them like built-in tools and fills in their inputs. Up to 10 custom tools per agent. Attaching or removing custom tools requires Automation (workflows write) access; other teammates can see them but not change them.

The writer doesn't save, review, or score its own work. The runner saves every version, then runs the editorial review and content score when the quality loop is on. Max tool calls per round limits how many tool calls the writer can make in one drafting or revision turn.

Quality Score and Review Loop

When Enable quality score & review loop is on, the agent doesn't stop after its first draft. Each piece goes through this loop:

1. Draft: the writer researches and drafts the piece, using up to Max tool calls per round.
2. Editorial review: the draft is checked for must-fix issues: AI clichés, banned phrases from your skills, disallowed em dashes, unsupported claims, and weak structure.
3. Content score: the draft gets a 0–100 score that combines four dimensions: Quality (clarity, depth, structure), Uniqueness (original insight, not generic filler), Helpfulness (answers the reader's real question with concrete takeaways), and SEO/AEO readiness (search intent, headings, and answer-engine-friendly structure).
4. Revise: if the score is below Minimum quality score or the editorial review has must-fix findings, those specific findings go back to the writer for another round.

The loop stops as soon as a saved version meets the score and passes the editorial review, or when it reaches Max review rounds. Pieces that never reach the bar still land in Review, flagged Needs attention, so a person can finish them. The most a piece can use is max review rounds × max tool calls per round (4 × 30 = 120 tool calls by default).

Picking settings:

Higher scores and more rounds give better drafts but take longer and use more model calls. If pieces are often flagged Needs attention, add more context in the Agent System Prompt or skills before you lower the bar. With the loop off, the agent drafts each piece once (still limited by max tool calls) and moves it straight to Review.

Assign Work

When assigning a Background Genie Agent from any page, you can also configure optional derivative / companion pieces (for example 2 LinkedIn posts derived from a blog article). Each derivative format automatically routes to the saved Background Genie Agent whose primary asset types match that format (such as your LinkedIn Agent for Social Posts), or you can override the agent per format in the assign modal.

• Ideas: in the Content Idea Lab saved ideas or Content Library → Ideas, tick ideas and click Draft with agent (or open an idea and choose Draft with agent). Each idea becomes a Studio draft revised to your quality bar. Save suggested ideas first.
• Drafts: in Content Library → Drafts, tick drafts and click Finish with agent, or use Finish with agent in a row's menu. Drafts are improved in place.
• Published content: in Content Library → Published, choose Update with agent. The live piece is never edited directly. The agent creates a separate revision draft, and the published row shows Revision pending. When you approve or publish the revision, its content is applied to the original, which stays published.
• Calendar days: on the Content Calendar, click the bot icon in a day's header, or shift-click several days and click Assign agent to days. Choose how many days before the date the agent starts (default 2) and whether it should create content for empty days. The agent finishes the items planned for that day and, if enabled, creates a new piece for an empty day. Days already inside the lead time start right away. To run on a single calendar item now, open it and click Finish with agent.

Monitor Runs

The Activity tab on the Agents page has three views:

• Running: live progress, the current review round and score, and Stop. Stopping finishes the current step and keeps anything already drafted.
• Scheduled: upcoming calendar assignments with their start date. Cancel any you no longer need.
• Completed: finished runs with each piece's score. Retry re-runs only the pieces that missed the bar. Dismiss hides a run from the list (its history and content are kept). Select several runs to dismiss them together, and turn on Show dismissed to see hidden runs again. Each run can be retried once, and only while its agent is active.

Limits that keep runs reliable: one assignment can target up to 20 ideas, pieces, or calendar items (and up to 90 calendar days). A run stops starting new pieces after about 90 minutes and finishes as Needs attention with what it has drafted. If a run is interrupted (for example by a platform update), it's marked Failed so you can retry it, instead of re-drafting pieces twice.

Every piece links to where it lives: Studio for drafts, the Library for published pieces, the Idea Lab for ideas, and the calendar day for calendar work. Agent badges on calendar cards, Library rows, and saved ideas link back to the run that worked on them, and the Agents sidebar item shows a count while runs are in progress.

Run the Agent From a Workflow

The Content Agent workflow task picks ideas, writes them, and schedules them on the Content Calendar for review. Optionally choose a saved agent in the task to reuse that agent's rules and quality bar.

What You'll Build

A workflow that, every Monday:

1. Reads the content theme for the coming week.
2. Drafts two articles plus a LinkedIn post for each.
3. Revises each piece until it scores at least 75 and passes the editorial review.
4. Schedules the pieces on next week's calendar in Review.
5. DMs your content lead in Slack with scores and Studio links.

Estimated time: 10 minutes.

Concepts used: Workflows, Triggers, and the Content Agent task.

Prerequisites

You'll need:

• Content Engine on your plan.
• Content skills that describe your voice and rules (Content Engine → Settings; see Write content skills). If you have none, the GTM Engine defaults are used.
• Optional: a month or week theme on the Content Calendar, if you want the agent to follow it.
• Optional: Slack connected, if you want reviewer notifications.

1. Add the Task

Create a workflow, add a task, and choose AI & Agents → Content Agent.

2. Choose Where Ideas Come From

• Content themes follows the theme on the Content Calendar for the schedule window. Most specific active theme uses a day theme if there is one, then a week theme, then the month theme.
• Call intelligence looks for topics and objections customers raise a lot that you have little content on.
• Instructions only generates ideas from the instructions you write.

Instructions apply to every piece in any mode, for example "Write for RevOps leaders evaluating forecasting tools. Lead with a concrete example." You can insert workflow variables with @.

3. Set What to Write

• Primary asset types: the main pieces, such as Article or FAQ. Multiple types rotate across the run.
• Pieces per run: how many primary pieces to write (up to 10).
• Require call evidence: only keep ideas backed by real customer call evidence, and write in customers' own words and framing.
• Related content: derivative pieces created from each primary piece and linked to it, such as LinkedIn posts from an article. Each derivative format automatically uses the saved Background Genie Agent specialized for that format, or you can select a specific agent per format.

4. Set the Quality Bar

• Enable quality score & review loop (optional, enabled by default): best for long-form content like articles, blogs, and whitepapers. You can turn this off for short-form formats like LinkedIn/social posts so the agent drafts each piece in a single pass and moves it directly to Review.
• Minimum quality score (default 75): each piece is scored on quality, uniqueness, helpfulness, and SEO/AEO readiness. It must reach this score and pass the editorial review (no AI-slop must-fix findings, banned phrases, or disallowed em dashes).
• Max review rounds (default 4, 1–10): how many times the draft is scored. Between rounds the agent revises using the specific findings.
• Max tool calls per round (default 30, 5–100): caps how many tool calls (SEO research, call transcript search, canvas edits, OG image generation, related content lookups) the writer can make inside a single drafting or revision turn.

Pieces that don't reach the bar are still scheduled, marked Needs attention on the calendar so a person can finish them.

5. Set the Schedule Window

• Start in (days): how many days from the run date the window starts. Use 1 to start tomorrow.
• Window (days): how many days the window covers, for example 7 for a week.
• Timezone: pieces are scheduled at 9:00 local time, spread across the least-busy days in the window.

6. Notify Reviewers

Select the teammates who should get a Slack DM or email when the run finishes. Optionally pick a Slack channel from the searchable dropdown (or enter a channel ID) to post a summary there too, add external email addresses, and pick a custom Slack bot if your workspace uses one.

7. Choose Skills

Content skills apply automatically based on each piece's asset type. Use Exclude content skills to ignore one for this task, or Also use these skills to pin extra guidance.

8. Add a Timer Trigger

Add a Timer trigger (for example every Monday at 8:00) so the agent fills the coming week automatically. You can also run the workflow manually at any time.

What Happens When It Runs

The workflow step returns right away with a run_id and status: queued, and the agent keeps working in the background. A run usually takes a few minutes per piece.

On the Content Calendar, agent-created pieces show an Agent badge with their final score. Open a piece to review it in Studio, then approve and schedule publishing as usual.

Troubleshooting

• No pieces were created: in Content themes mode, check that a theme covers the schedule window. With Require call evidence on, there may be no evidence for the theme yet; turn it off or switch to Call intelligence.
• Everything is marked Needs attention: lower the minimum score, raise max review rounds, or review your content skills for rules that conflict.
• No Slack message: make sure Slack is connected and each reviewer has their Slack account linked to their GTM Engine user.
• A scheduled day never ran: check the agent isn't archived (archived agents skip their assignments) and that the day had planned items or Create content for empty days was on.
• Update with agent didn't change the live page: that's expected. Approve or publish the revision draft (the Revision pending chip opens it) to apply it to the original.