Updated Jul 2026
How AI writing uses your brand
If you’re going to let software draft posts that clients will read, you deserve to know exactly what it’s working from and what happens to the words you give it. This page explains both.
What the AI draws on
Every draft starts from two things: your brand profile and the description you type into the Create Post dialog.
The brand profile carries what you configured on the Brands page. Your brand’s name. The Tone and Style you picked. The Topics you want posts to cover and the ones you told it to avoid. Your Sign-off and brand hashtags. The AI reads all of it, every time it writes, which is why a brand set to Authoritative and Educational produces noticeably different text than one set to Casual and Storytelling.
The dialog description is the other half. When you write “Remind clients enrollment ends December 7,” that sentence becomes the subject the AI writes about, filtered through the voice your profile defines. A thin profile gives you generic drafts. A specific one gives you drafts that sound like you.
Why each platform gets its own version
One request produces one version per platform you selected, not one post copied everywhere. The request the AI receives includes the name of each target platform, so it writes for that platform specifically. The LinkedIn version and the X version answer the same brief in different words.
That separation matters after generation too. Each version lands in your queue as its own pending post, so you can edit, approve, or reject the LinkedIn version without touching the others.
What counts against your allowance
Every piece of post text the AI writes counts as one AI text generation for the month. That includes the version-per-platform behavior above: selecting three platforms means three generations, because three separate posts got written.
Regenerate counts as 4 per click. It’s not a surcharge. Each click genuinely writes four complete alternatives, and each one is a full generation whether or not you keep it. The same monthly meter also feeds your AI cost this month figure. Both reset when your billing period does. See Understand your plan and usage for how the meters work.
What the AI cannot do
It writes. That’s the whole job.
It can’t publish anything. Every AI draft enters the queue as pending, exactly where a post you typed yourself would land, and a pending post never publishes on its own. It can’t approve anything either. Approval is yours, and if you’ve turned on compliance review for the brand, the AI’s output faces the same gate as human writing: a finding blocks approval until you fix the text or acknowledge it. The review assists your supervision of the content. It doesn’t certify compliance, so read every draft before you approve it.
Your earlier text survives regeneration too. The original AI version is kept separately from whatever you save over it, and History can take you back.
Where your words go
Drafts are sent to OpenAI for generation and for compliance review. The request carries what the model needs to write: the post content and your edits, your brand name and voice configuration, the topic or context you supplied, your sign-off and branding settings, and the target platform names.
What never goes: end-user names, email addresses, or user identifiers. Your captured contacts and consent records are not part of any AI request. Your content isn’t used to train models, either ours or, under OpenAI’s API policy, theirs by default.
Everything the app itself stores lives in the United States, on Google Cloud, in a single region. The full picture, including every vendor that touches your data, is on the Data & security page, and where that page and this one differ, trust that one.