Brand and knowledge

When MentionFlow writes content for you, or fact-checks what the engines say about you, it is only as accurate as what it knows about your brand. This is where you teach it.

What it does

Brand and knowledge holds two things: your brand identity, meaning its name and every alias engines might use for it, and a knowledge base of your real facts, meaning positioning, pricing, features, and the claims you never want made. That gives the content and fact-checking agents one source of truth. Fill it in well and every draft stays true to you, and every fact-check has something to check against.

The Brand hub's four tabs

The hub is one sidebar entry with four subpages. Each answers one question. /brand stays the front door and lands on Identity.

• Identity: who you are, and what the engines should call you. The brand's logo, its name, the domain you collect for, and the aliases every mention is matched against. A Site card sums up what the crawler saw on that domain: detected title, pages discovered, last crawled.
• Knowledge: what is true about you. The structured fact sheet and its Sources sub-page, both below. This is the ground truth every draft and fact-check reads.
• Voice: how to say it. The tone, audience, avoid, and sample-voice profile the content agent writes your drafts in, plus its generation defaults. It shapes how a draft reads, never what it claims.
• Matching rules: how answer text gets credited to you rather than a competitor. See Matching rules.

The logo

On the Identity tab you can set a brand logo by URL, or press Fetch from site. MentionFlow reads your homepage's declared icons (touch icon, social-preview image, site icon, favicon) and offers them as candidates. The server only ever fetches your own homepage's HTML to see which icons the site declares, and all image loading happens in your browser. With no logo saved, the mark falls back to the app's honest letter avatar, the same fallback every favicon uses. Never a stock or invented image.

How to use it

• The knowledge base is a structured fact sheet of eleven fields, including a short and a long product description. Each field has a character cap, and lists hold up to 40 items, entered one at a time or pasted in bulk as individual pills. The content agent treats the sheet as ground truth. It never invents facts, and never contradicts your "never claim" list.
• You can auto-generate a first draft of the fact sheet, descriptions included, from your own website. It is always shown for review before saving. A field your site doesn't support comes back empty, never invented.
• A separate knowledge library holds longer documents, on the Knowledge tab's Sources sub-page. These are chunked and embedded, then retrieved at draft time: the top 5 by similarity. When nothing matches, generation falls back to the fact sheet.

How it's computed

Brand context is always resolved through a scoped lookup. The active-brand cookie is attacker-controlled, so it is never trusted on its own. The lookup is scoped to your workspace, which guarantees you only ever see your own workspace's data.

Archive and delete (moved to Project settings)

A brand's lifecycle controls moved to Project settings, under Danger Zone (/settings, anchor #danger). Destructive actions now live with settings, and the Identity tab keeps a quiet pointer to them. They stay owner and admin only:

• Archive (soft): collection stops. The brand frees its project slot and its prompts' pooled quota right away. Engine slots are billed once at the workspace level, so there is nothing per-project to free. Every answer and metric is kept. The brand stays in the switcher and the API with an archived badge or flag, its history stays truthful wherever it renders, and it is left out of Portfolio blends by default, with an explicit show-archived toggle. Restore re-checks each of those quotas first and refuses with the numbers if they have been re-spent. Archiving is never a quota loophole.
• Delete (hard): typed confirmation of the brand's name, then a permanent cascade of the brand and everything under it: prompts, answers, mentions, citations, metrics, competitors, its collection config. Audit-logged. Cannot be undone.

Limits

The fact sheet is bounded by its per-field character caps. The knowledge library's embedding and retrieval need an OpenAI key. Without one, retrieval falls back to the fact sheet alone.

Related

• Content: the drafts and briefs this knowledge grounds.
• Matching rules: competitor aliases, custom patterns, and chart colors (your own brand's aliases are set here on the Brand hub).
• Actions: briefs that draw on the same knowledge base.
• Answers and receipts: the real answers fact-checking is measured against.