Why our drafts read like your brand wrote them: brand memory, style profiles, per-platform language, and image-2 image generation.

How our AI content generation works (world-class, hands-off)

TL;DR — Drafts sound like your brand because they draw on the same context an in-house team would use: brand memory, style profiles, per-platform language, and image generation via GPT-image-2. It's not "AI content" — it's your voice, produced faster.

Why most AI content sounds like AI content

The typical failure mode is a generic model, a bare prompt, and no context. The result is text that could belong to any brand — hedged, buzzword-heavy, and disconnected from anything specific.

We solve that with layers, not with a bigger model.

Layer 1 — Brand memory

Scraped at onboarding and refined as you use the app:

Visual identity — real CSS-extracted brand colours, logo, typography
Tone & voice — inferred from your existing content and confirmed on the founder call
Do-not-say list — words, claims, or comparisons you want us to avoid
Positioning — what you sell, to whom, and why
Recent context — new launches, campaigns, hires, changes

Every draft is generated with this brand memory in the prompt. Nothing generic gets past it.

Beyond brand memory, we build a style profile from 3–5 pieces of your existing content: cadence, sentence length, favourite openers, favourite closers, comparison structures, list formats. New drafts blend the style profile with a session fingerprint so successive posts don't feel copy-pasted from one template.

You can pick a global style ("editorial" is the universal default) or override per project.

Layer 3 — Per-platform language

A LinkedIn draft is not an Instagram draft with different padding. Each platform gets its own prompt shape:

LinkedIn — narrative hook, insight, credible detail, soft CTA
Instagram — visual-first caption, punchy first line, hashtags at the end
X — declarative first post, then supporting thread posts, no filler
Facebook — mid-length, community-tone, image required
Blog — long-form with TL;DR, headings, at least one table or chart, FAQ block, semantic HTML

You can also override the language per platform in Brand Hub (great for multi-market brands).

Layer 4 — Image generation

Every visual is generated through GPT-image-2, OpenAI's image model. We use it across every path: from-scratch drafts, reference edits, style regeneration, carousels, calendar, feed, and free reports. No Gemini image fallbacks — if image-2 rate-limits, we retry image-2 rather than substituting a lower-quality model.

Images are generated with:

Your brand colours and typography guidance
A mandatory "no text in image" negative prompt (blocks the classic AI-image garbled-text problem)
Composition guidance tuned to the target platform's aspect ratio
Optional reference images (your existing product shots) as style anchors

Layer 5 — Post-generation safety

Before a draft ever appears in your dashboard we run:

Length checks per platform
Buzzword suppression (revolutionise, unleash, game-changer — you know the list)
Auto-shortening for platforms with hard character limits, without changing meaning
Image-required checks — Instagram and Facebook drafts are blocked from publishing if the image failed to generate

What you keep control of

Everything. Every draft is editable in-line. Every image can be regenerated. Every scheduled post can be paused. Autopilot can be turned off at any moment. The system is designed to be trusted with the wheel, not to take it from you.