AI drafting writes a first version of a post from a short brief, including different wording for each channel. You then edit it, and a practice administrator approves it as normal.

It uses your organisation's credits — the same balance AI Insights uses, so topping up or spending in one place affects the other.

Writing a draft with AI

1. Open a draft (or select New draft)
2. Next to Content, select Draft with AI
3. Describe what the post should say — for example: *"flu clinics run every Saturday in October, walk-in, no appointment needed"*
4. Select Generate

The generated title and copy appear in the composer for you to edit. If you've already ticked channels, each one gets its own version sized for that platform — X within 280 characters, longer copy for Facebook, LinkedIn and the newsletter.

[SCREENSHOT: Composer with the AI brief box open]

Generating on a draft that already has text replaces it. Copy anything you want to keep first.

What it costs

Each generation deducts credits based on how much text goes in and comes out — typically a few credits per draft. Longer briefs and more channels cost more.

After each generation you're told exactly what was used and what's left: *"Draft written — 3 credits used, 2,410 left."*

Credits are shared across your organisation and with AI Insights, so a colleague generating a draft draws on the same balance.

If you run out, generation stops with a message and no content is produced. A practice administrator can top up under AI Insights. Nothing is charged for a failed generation.

What the AI will not do

Two limits are built in:

• It never writes about an identifiable patient. If a brief asks for content about a named or identifiable individual, it refuses and tells you why. Public communications should never contain patient information — this is a backstop, not a substitute for care when writing the brief.
• It never publishes. It produces a draft. Approval and scheduling remain human actions, and only an administrator can approve.

Please read what it writes

AI drafts are a starting point, and they can be wrong. Check before approving:

• Facts — dates, times, clinic names, phone numbers, eligibility. The model only knows what your brief told it, so anything you didn't state may be invented.
• Tone — that it sounds like your practice.
• Clinical wording — it is instructed not to give medical advice, but you are responsible for what your practice publishes.
• Accessibility — plain English, no unexplained jargon.

The approval step exists precisely so a second person reads it. Don't approve your own AI draft without reading it properly.

Getting better results

• Be specific. "Flu clinics, Saturdays in October, 9am–1pm, walk-in, over-65s and at-risk groups" beats "flu clinic post".
• Say the audience if it matters — parents, older patients, carers.
• Tick your channels first, so variants are written for the right lengths.
• Pick the campaign first if the draft belongs to one; its name gives the model useful context.
• Generate, then edit. Treat it as a first draft, never a finished post.

Related articles

• Getting started with the Communications Hub
• How approvals and scheduling work
• Getting started with AI Insights
• Insight credits: what they are and how they are used