BrandSphere AI System Manual

BrandSphere AI is a multi-tenant AI Brand Content Operating System. It turns a photograph, a sentence, a voice note or a web address into a fully branded, factually grounded, approval-governed social post — and publishes it everywhere your organisation is connected.

It is designed for the multi-company, multi-brand, diplomatically sensitive executive: someone who may operate a dozen subsidiary companies, seventeen digital platforms and several institutional offices at once. That profile is the design target, not an edge case.

An ordinary scheduler stores your text. BrandSphere AI stores your identity — your brands, logos, executives and standing notices — and writes from your own approved material rather than from a blank prompt. It never lets a model draw your logo or your Arabic text, it asks what to do with every post instead of assuming, and it reports each channel's outcome independently rather than saying “it failed”.

This manual documents the full product as specified in PRD v1.2. Because the product is built in phases, each module is labelled with what you can genuinely do today.

**Brand Lock.** No AI image model reliably draws a specific corporate logo, and none reliably renders exact Arabic, Urdu or Hindi text. Rather than accept a mangled logo, BrandSphere AI splits the job: the AI generates or selects only the background, and everything that must be exact — your logo, an executive's name, a title, an email address, a website, a phone number, a legal notice, a price — is placed from your stored files and verified data at exact coordinates using real fonts. This is the architecture, not a fallback.

**The Disposition Gate.** Every time the system finishes a post it stops and asks one unmissable question: your post is ready, what would you like to do? Publish now, save as a draft, schedule it, or send it to Autopilot. Each option shows its consequences first, including which channels are affected and whether approval will be required. It can never be switched off. Every competitor either publishes by default or drafts by default; both are wrong for an executive who sometimes wants immediate presence and sometimes wants a week's planning.

**Protocol Guard.** For diplomatic, government and institutional posts, the system will not assert an office it cannot verify. It offers the title you supplied, a verified title with its source and date, or neutral wording that avoids the title — but it never invents one. It also checks name spellings, applies correct honorifics, orders dignitaries by precedence, and flags any wording implying an agreement for explicit human confirmation. For Embassy, Government, Chamber and Association organisations it is always on and cannot be disabled, on every plan including the cheapest.

Your account sits inside an organisation. If you run several companies you switch between them from one sign-in rather than logging out and back in, and everything you do is scoped to whichever organisation you are currently acting for.

Setup is a guided wizard of eight steps, and it is deliberately a value ladder rather than a form gauntlet: every step can be skipped, everything is saved as you go, and you can resume on another device. A completeness score shows what will most improve the quality of what the system writes for you.

This is what makes the product more than a text box. Your brand stops being a folder of files and becomes structured data: logos in every variant, colours, approved fonts for both Latin and Arabic, tone of voice, approved and banned terminology, hashtags and standard calls to action.

Because one person often holds several offices, identities are stored separately from people. A post signed as “General Secretary” of one body carries that body's signature, hashtags, website and tone; the same person posting as a group CEO carries the company's instead.

There are six ways to start a post, and all six converge on the same pipeline: gather your brand and identity context, generate, validate, compose the branded image, build per-platform versions, show you the Trust Panel, then ask what to do with it.

**Quick Create** takes a sentence, a photo or a name and asks at most three questions. **Guided Create** offers a structured brief whose fields follow the category you chose. **Voice Create** lets you speak it, then correct the transcript. **URL Product Post** researches a page you paste. **Event Post** turns up to twenty event photographs into a branded collage with a caption naming the people in it. **Manual Editor** lets you write it yourself, with AI assistance available but never mandatory.

Everything written is checked before you see it: banned words, factual grounding, platform limits, duplicates and safety. A Trust Panel on every post shows what was sourced, what is unverified, what was removed and why approval may be required.

If you manage twenty or more web properties, this is what stops you pasting the wrong link. Each website, landing page and product is saved once with its own description, default call to action, hashtags, logo and associated social channels.

Give the system an address and it reads the page and produces a structured, sourced summary you can build posts from. Prices are extracted verbatim with their source and never inferred; anything it cannot establish is recorded as unknown rather than guessed. It re-reads later and shows you what changed.

Nothing sensitive goes out without the right person seeing it. A post moves through fifteen clearly named states from Idea to Archived, and once approved the version is locked — any later edit visibly voids that approval and sends it back round, with an explicit warning rather than a silent reset.

You can approve copy and visuals separately, or approve some platforms and hold others. Requesting changes and rejecting both require a written reason, which is recorded permanently against the post.

The calendar shows your whole plan in six views. Dragging a post reschedules it, snapping to your configured slots and refusing to land inside a blackout period. Best-time suggestions come from your own published performance, and say so honestly when there is not yet enough data.

Platforms differ enormously in what they permit, so every channel is honestly labelled with how it publishes. Where a platform offers no publishing interface at all, the system prepares everything and hands you a two-tap assisted path rather than pretending it cannot help.

Each channel gets its own independent job. Five channels means five independent outcomes, so “partially published” is a proper state rather than an error. Retries are automatic and scoped only to the channels that failed — a successful post is never published twice.

Autopilot keeps content flowing without you starting it, but only inside rules you set. Every action is the consequence of a rule a human configured, is logged, and can be stopped within sixty seconds at organisation, brand, platform or campaign level. High-risk content — political, diplomatic, legal, medical, financial — can never be auto-published under any configuration.

Analytics drill down from organisation to brand to campaign to post to channel. Its defining principle is honesty: where a platform genuinely does not supply a metric, it says so rather than showing a zero that looks like failure.

Nine plans run from a free trial to white-label reselling, with a separate Embassy and Institutional plan. Limits are metered rather than silently enforced — you are prompted to upgrade rather than blocked mid-task.

Every piece of data in the platform carries the identity of the organisation that owns it, and every query is scoped by it. This is enforced in the database layer and verified by an automated test suite that blocks any change violating it.

Files are never served from a public storage link. All media passes through an authenticated proxy that verifies your organisation owns the file before returning a single byte.

Every state-changing action writes an audit entry recording who did what, when, and why. Platform support staff hold no permission to read your content: access requires an explicit, time-boxed, logged break-glass grant.

Photographs are personal data under PDPL and biometrics are sensitive, so the product performs no automatic facial recognition of any kind. Faces are detected only for cropping geometry, never identified.