Skip to content

Posts drafted for you, published under your name

Canopus is an autonomous growth manager that drafts posts in batches of three, in LinkedIn, X and search formats, for a founder to edit and publish under their own name, and that never posts on social networks on their behalf.

Three drafts per batch, each in a different format, each written against your strategy, your product brief and what past work taught. You edit one, publish it where you publish, and type back what it did. Articles, comparison pages and FAQ blocks written from your audit findings are coming soon.

How is a batch written, and what do you do with it?

  1. Three drafts, three different formats

    A batch runs on its own. The producer is asked for at most three posts, and each one has to use a different format. Every draft carries a hook, a body, a call to action that is usually a question, and one sentence of rationale: why this post, for this audience, now. The hooks of the last sixty days are passed in so the same post is not proposed twice.

  2. You edit one and publish it yourself

    A copy button, and nothing else. There is no publish action, no scheduling, no connected social account. You make the last edit where you publish, because your name is on it. Declining a post is one click and is recorded, because a refusal is data.

  3. You type back what happened

    Impressions, clicks and any note worth keeping. The fields are optional, and a blank stays blank: "published but not reported" has to stay different from "published and got nothing", or every later average is dragged down by diligence about publishing and negligence about reporting.

What does this replace?

  • Opening a blank post, staring at it, and closing it twenty minutes later.

    Three drafts are waiting, each with a stated reason for existing.

    Replaces: The blank page
  • Pasting your product into a chatbot and getting a post about shipping fast.

    The writer is given your strategy, your brief and what past work taught.

    Replaces: A generic prompt
  • Picking a post shape by instinct, and writing the same one every time.

    Six LinkedIn formats and three X formats with named beats, and a different one required for each post.

    Replaces: Guessing the shape
  • Publishing for a year with no record of which posts were worth the hour.

    A published post you reported results for becomes an entry in the playbook.

    Replaces: No memory

What does the writer know, and why do you press publish?

The producer is given four things. The product brief, written by an agent from a crawl of your site and edited by you: the name, the one-liner, what the product does, how you sound, what you claim sets you apart, the proof you already publish. The adopted strategy: its positioning sentence and the segment the batch is for, with that segment's pains and objections, so the post speaks to the same person your campaigns do. The playbook, which is the conclusive lessons from finished campaigns and reported posts over the last ninety days. And the formats available on the channel, each with its beats spelled out. When a calendar slot asks for a post, the batch is a single post on the angle the slot names, in the one format it chose.

A format is not a template. It names the condition under which that shape is the right choice, then lists the beats in order, and the producer writes the copy itself. "Before and after" wants a specific day rather than "teams struggle", then one decision, then the after with a number, then what you would do differently. "The mistake I made" requires a real mistake, what it cost in money or months or a customer, and no product mention at all. "Teardown of something public" asks for three things the artefact does genuinely well before the one you would change, because the fairness is what earns the read. "The thing everyone says, that is wrong", "Building in public, with real numbers" and "How we do the thing we sell" round out the six LinkedIn formats. Three X formats sit in the same library: a five-post thread built so the first post stands alone, a single observation with no link, and a build log. The three posts in a batch must use three different formats: three posts in the same shape are one post sent three times.

The prompt is the substance, and its strongest instruction is a test the model applies to its own draft: if the post could have been written about a different company, it is not a post. Everything a model reaches for when asked for a LinkedIn post is instantly recognisable as generated, so the register the platform is drowning in is forbidden by name. No emoji, no "Here is what I learned", no single-sentence paragraphs stacked for rhythm, no "Agree?" at the end, no hook that withholds the point to force a click on "see more". Inventing a fact, a number, a customer or an anecdote is forbidden outright: where a post needs a story the producer does not have, it leaves a plain gap for you to fill rather than a plausible fiction. The hook has to survive alone, because everything after the first line is hidden behind "see more".

Canopus does not post on social networks on your behalf: your name is on the post, and the last edit before it goes out is yours. That is why the loop needs you to type the results back. Nothing here is connected to a platform, so impressions and clicks are hand-copied figures from a dashboard you looked at, which is worse data than an API would give and is the only data there is. The interface says so rather than presenting a typed number as though it were measured. A post you reported on is concluded by the same job that closes campaigns, with a higher impression floor and looser rates precisely because the figures are self-reported: under three hundred impressions nothing is judged, and a post seen and not clicked is inconclusive rather than a failure, since most of what this channel returns is replies and recognition that no click count records. A post that worked becomes a playbook entry, and that entry is read by every agent that writes anything afterwards, including the producer writing the next batch.

The producer takes the channel as an argument, and the table that stores posts is not specific to LinkedIn. Batches are written for LinkedIn today, and the same producer drafts the approach message and the brief for a creator when the calendar asks for one, for you to send; the X formats are written and available to the strategist when it scores channels, and it will write for X when the strategy says so. The search formats are the third family, and they are the work of the SEO and GEO content agent, on Growth and Scale: from your latest SEO and GEO audit, it proposes an editorial plan where each piece answers one finding, takes a format from the library (comparison page, alternatives page, how-to, definition with FAQ, teardown series) and targets a real search or a question a buyer would ask an AI assistant. You drop a piece or ask Canopus to write it; you edit the draft and publish it on your own site. Beside the plan sits a kit to paste, built from your validated brief and your scanned pages: an llms.txt, the structured data and the title and description of your home page. A rescan then compares your search and citability scores with the audit the plan was drawn from. The same rule holds across all three: Canopus drafts and measures, and you decide what goes out under your name.

Questions founders ask

See your own numbers first

Connect Stripe and your real revenue is on screen in five minutes, before an agent has said a word. Create your account, answer the questions, and your strategy gets written.