Generating content

The generate screen asks one question about the output — what are you writing? — and it shapes the writing itself, not just the layout. A LinkedIn post is not a shortened blog; it is written differently from the first sentence.

The four post types each set their own length, so the word and paragraph counts under Advanced do not apply to them — the screen says so when you open it. The focus keyphrases still do. There is no separate platform question: the type you choose is the platform it is written for.

One run produces one format. To get the same story in a second format, run the same brief again with that format chosen.

Alongside the brief you can paste source material: notes, a report, a transcript, an existing article. It is treated as evidence to draw facts from, never as a draft to rewrite and never as instructions. Claims it cannot support from that material or from your verified brand data are left out rather than guessed at.

Source material works with every format. The case it was built for is a blog: a comms officer pastes in a 30-page cabinet report, sets the angle to “what this means for residents whose bin collection changes”, and gets a resident-facing article that holds to what the report actually says.

Pick a saved voice, or write one-off custom instructions for this run alone. You cannot do both — a piece has exactly one source of style, which is what makes it possible to say later why a piece sounded the way it did.

Describe what you want in ordinary language. There is no syntax, no keywords, and nothing to be gained from writing it as an instruction to a machine.

Three settings are available and all are optional. Left alone, the writer chooses for itself, which is usually right.

Values are re-checked server-side, so an implausible number is clamped rather than producing a strange result.

The run is queued and worked through a chain of stages: drafting in your voice, judging how well the draft matches it, shaping it for the chosen format, and a final pass that removes the patterns which make text read as machine-written. You do not choose or configure any of this.

Repurpose re-authors a finished piece into a different format. It reads the piece as it stands — including your edits — rather than the original brief, so corrections you made carry across instead of being made again.

It creates a new piece and a new run; the original is untouched. This is the intended way to get one story onto several channels, and it produces better results than running the same brief with a different format selected, because the second version is written from something you have already approved.

Improve takes text you paste in and re-writes it in one of your captured voices. Where generating starts from an idea, improving starts from someone else's draft — an agency submission, a colleague's first attempt, something inherited.

Like repurposing, it produces a new piece and leaves the original alone.

Failures happen and are recorded plainly — the reason sits against the run in Runs. A failed run produces no piece, which is why piece counts and run counts differ.

Retrying the same brief is reasonable; most failures are transient. Open the failed run from Runs and choose Try again to re-submit it as it was. A brief that fails repeatedly is usually asking for something the pipeline cannot produce, and rewording it is more productive than retrying it. Common causes and their messages are in Troubleshooting.

Generating consumes credits, as do repurposing, improving and previewing a voice. Reading, editing and exporting do not. The full list of metered actions — including image generation and sending to a destination — is in Billing and credits.