How AI works in Brand Control Center
Brand Control Center (BCC) uses AI across brand setup, video enrichment, clipping, group matching, moderation, and feed ranking, while people approve what publishes.
BCC uses AI at five fixed points in the content lifecycle: brand setup, video arrival, community matching, guideline checks, and placement ranking. Each point uses a different model with a different input, so AI works differently at each stage of the workflow.
The AI is governed: it speeds up operations without replacing human judgment. AI recommends, flags, and scores, and an admin or moderator approves every decision that affects what gets published.
The AI in BCC is part of Supervised Intelligence. For the whole product, see Supervised Intelligence: what it does and where to start.
AI in BCC has the most impact when you manage high-volume UGC or creator content, grow your social experience quickly, or run many placements and environments. It helps most where you need speed without giving up brand safety and trust.
Where AI works in Brand Control Center
| Module | What AI does | What you control |
|---|---|---|
| Build | Suggests categories, keywords, guidelines, a persona, and brand assets from your website, and sets up the structure, taxonomy, and brand rules later steps use | Every suggestion is editable before you save it |
| Curate | Transcribes, describes, and tags each video; cuts long videos into clips and scores them; matches videos to communities and groups; prioritizes and recommends | Which sources to import, which options to apply, which community a video lands in, and every approval |
| Moderate | Checks transcript and frames against platform, brand, and community guidelines | Your guideline text, and the approve or reject decision on every flag |
| Grow | Ranks videos into placements and feeds from page and viewer context; suggests product links | Which context signals a placement uses; which links to keep |
AI also works in Monetize, where it aligns content with context and optimizes performance. Learn more in How AI automates revenue in your social experience. Report shows the engagement and outcomes that follow.
Brand setup
When you enter your website during onboarding, BCC sends the URL to Genuin's data-science service. The service reads the site and returns suggestions for each setup step:
- Industry type and the consumer brands associated with you.
- Categories, topics, and keywords that describe your content. Learn more about brand categories and topics.
- Brand guidelines: a starter list of rules for moderation. Learn more about brand guidelines.
- Brand persona: text for the seven persona fields (Mission, Vision, Values, Policies, Procedures, Working conditions, and Code of conduct/behavioral expectations). Learn more about the brand persona.
- Social handles found on the site, for Social Import.
- Logo and brand elements, with a Regenerate option for the logo.
Each suggestion appears in the form for that step. Nothing is saved until you click Save, and you can replace any suggestion with your own text.
The persona and guidelines you save become inputs to moderation and ranking later. The more specific they are, the more specific the AI's checks are. For how Supervised Intelligence uses your website and guidelines as context, see Brand Context and Website Crawler.
The same service powers the GenAI chat in BCC. A routing agent sends each question to a specialist agent. There are agents for categories, topics, and keywords, for content strategy, and for a phased onboarding plan.
Other agents cover monetization, brand assets such as logos and color schemes, and questions about your reporting data. Answers can be wrong, and the chat says so under every reply ("gAI can make mistakes. Check important info."). Nothing it produces is applied to your brand until you enter it in a form.
Community images use the same generator. On a community's profile and banner fields, Generate AI Image creates an image from a text prompt instead of an upload.
Video enrichment
Every video that enters BCC goes through the same enrichment before anyone sees it in Curate. Videos arrive by Upload, Social Import, Clip It, or AI avatar generation. Enrichment runs three steps:
- Transcription: the audio becomes a time-coded transcript.
- Visual description: a vision model describes what's on screen, scene by scene, including objects, scenes, and products.
- Semantic tags: a language model extracts keywords, assigns IAB content categories, and produces an embedding (a numeric fingerprint of the video's meaning).
Enrichment also reads captions, metadata, overlays, and source and context signals. The result is a semantic understanding of each video before it reaches moderation or curation.
The transcript and description are what every later step reads. Moderation checks them against your guidelines, and community matching compares the embedding with the descriptions of your communities and groups. Placement ranking compares the embedding with a page's context.
A video with no speech and a static image gives these models little to work with. Its matches and scores are weaker than those of a video with clear narration.
Uploads accept files up to 2 GB and 300 seconds, in 9:16 aspect ratio.
Clip It
Clip It turns one long video into short clips. Paste a link to a video up to 60 minutes long, choose the communities and groups the clips can go to, and pick the options to apply:
| Option | What it does |
|---|---|
| Show Title on Video | Overlays a generated title that states the clip's key message |
| Show captions | Burns in captions from the transcript, in the Best fit style or a Custom style you define |
| Add background music | Adds a music track chosen to match the clip |
| Add description | Writes the post description from the transcript |
| Smart crop | Reframes landscape footage to vertical by following faces and gaze, and cuts to a new framing when the number of faces or the gaze direction changes |
| Add clickbait thumbnail | Generates a thumbnail with a hook line |
| Merge videos into clips | Joins adjacent segments into one clip when they form a single thought |
The clipper first splits the source into chapters from the transcript, then builds clips of up to 120 seconds. A language model then scores each clip's transcript on four rubrics:
- Hook: how strongly the opening earns attention.
- Flow: how smoothly the clip moves from one point to the next.
- Value: what the viewer gains, and whether the clip delivers on its opening.
- Trend: how well the content aligns with current cultural or platform trends.
Each rubric is a score out of 100, and Viral is their weighted average. These are predictive signals: they estimate engagement likelihood, retention potential, and virality probability. The community classifier adds a relevance score for each community or group.
The scores appear on every clip in Curate. Use them to decide what to publish first, plan your placements, and focus on content that drives outcomes. They don't publish or rank anything on their own.
Learn more about Clip Generation from Long-form in Studio.
Community and group matching
When a video arrives with a set of candidate communities, the classifier compares the video's embedding with the name and description of each community and group. It assigns the video to one group.
A match counts only when the similarity is at least 0.6 on a 0–1 scale. A video that clears that bar for no group stays unassigned, for you to place by hand.
Among matches, meaning comes first. The classifier then applies small adjustments: a boost for groups that are new or have few videos, and a penalty for groups that are already full. Those adjustments never override a clearly better match.
For example, a skincare routine video goes to a Beauty community, in its Skincare Tips group. A long product demo is auto-clipped first, and its clips go to a Product Reviews group.
You choose the candidates when you import. On the Social Import screen, for example, you pick a community and its groups before you choose the accounts to import from.
If you allow it for an import, the classifier can also propose a new community or a new group. It does this for videos that fit none of the existing ones, and a proposed entity is a candidate, not a destination.
It appears in Curate labeled as curated by AI and is created only when you approve a video into it. A proposed community always comes with one default group.
Matching cuts manual tagging and keeps placement consistent with your rules. Learn more in Add and curate content in AI-generated communities and groups.
Moderation
Every video is checked three ways before it reaches the moderation queue. The checks run on the transcript and on the visual description, so they catch spoken, written, and shown content.
Together, the checks look for policy and guideline violations, unsafe visuals or language, copyright and IP risks, spam or misleading claims, and contextually sensitive content.
| Check | What it compares | Labels you see |
|---|---|---|
| Platform | Transcript and sampled frames against Genuin's fixed policy categories | Hate Speech, Threatening Language, Harassment, Sensitive Topics (self-harm, violence), Sexual Content, Inappropriate Visual Content (nudity), Depictions of Violence (weapons, blood, explosions), Substance Use (drugs, alcohol, smoking), Inappropriate Audio Content |
| Brand | Transcript and visual description against your brand guidelines, one rule at a time | The guideline that was violated |
| Community | Transcript against the guidelines of the community the video is posted to | The guideline that was violated |
A flag records the source (audio or visual), the reason, the model's confidence from 0.1 to 1.0, and the timestamps of the offending segment. Frames are extracted at those timestamps so a reviewer can jump straight to them.
Flagged videos appear in Moderate > Content Moderation > Queued. The Flagged by column shows AI Moderator Agent, Guideline shows the guideline type, and Reason shows the label. Comments go through the same platform check.
Each item moves through three states: Queued (awaiting review), Approved, and Rejected.
The model's decision is a flag, never a removal. A moderator approves or rejects every queued item, and an approved item publishes regardless of what the model found. No setting lets AI publish or reject on its own.
Guidelines are matched strictly: a video that mentions or references a rule is treated as violating it. We recommend specific prohibitions ("No mention of competitor retailers") over general values, because each rule is checked and reported individually, by its position in your list.
Learn more about Moderation Baselines and Queue and Community Guidelines and Appeals.
Curation
In Curate, AI speeds up editorial review without removing oversight. It prioritizes high-quality or high-potential content and recommends approving, rejecting, or moving each item.
It also shows Clip It's virality and engagement scores, which cuts review fatigue when volumes are large. Editors and admins can always approve, reject, or move content, and override any AI recommendation.
Placement and feed ranking
Two models decide where a video shows up on your site or app. They consider page context, such as a homepage, a product detail page, or a discovery page. They also consider the viewer's journey stage (Discovery, Pre-purchase, or Loyalty) and the relevance rules for your communities and groups.
Placement classification runs when a video is approved. For each placement style you've built, the model scores the video against every section of that style.
It uses four criteria: relevance to the section's purpose, fit with the consumer-journey stage, specificity, and engagement potential. It keeps the top 10 style-section matches, each with a confidence from 0 to 100. By default, every match is eligible, because the confidence threshold is 0.
Feed ranking runs each time a placement renders, as part of how Maya orders each feed. Hard filters scope the feed first: your brand, the selected communities and groups, and any posted-by rules. Boosts for the viewer and the page then order it:
| Signal | Boost |
|---|---|
| Context match (custom context, page context, previous page context) | 2.0 |
| Viewer location | 1.5 |
| Viewer language | 1.3 |
| Place | 1.2 |
| Viewer segment (gender, age) | 1.1 |
| Scheduled show time | 1.0 |
Pinned videos always come first. For a placement with sections, each section's title and description is embedded, and videos are sorted into their best-matching section at render time. With no boosts in play, the feed falls back to newest first.
You choose which context signals a placement sends under Grow > Placement > Style > Implement. The signals are Custom Context, Brand ID, Community & Group ID, User Interests, Posted By User, and Previous Page Context.
Two gatherers add page context. On-Page Context Gathering by Crawler reads the links and text of the page the placement renders on. Real-time context gathering reads the ads running on it. Turning a signal off removes its boost; it doesn't change the scoped set of videos.
The result is a contextual feed in each placement instead of a generic one. Learn more in Create a placement and map it to the shopper journey and Personalization and relevance: how Maya ranks every feed.
Product link suggestions
For brands with a website, AI proposes linkouts for a video. It extracts search terms from the video's description and transcript, searches your site for matching products, and validates each result against the video.
It returns the 3 most relevant links, each with a confidence shown as a percentage match. In production, only links at 80% or higher are suggested. You accept, replace, or remove each link, and suggestions never attach to a post on their own.
Learn more about how to manage video linkouts centrally.
AI video generation
A separate pipeline generates videos rather than cutting them: talking-avatar videos from a script, teaser videos, and news-to-video. It runs the same stages for every type: ingest, analyze, plan, assemble, narrate, compose, post, and publish.
Its posts go through the same enrichment, matching, and moderation as uploads. Generation is a paid capability. A subscription plan can include Generate and post videos with AI, charged per generation. The plan's activity view shows Minutes Generated with AI this month.
Engagement and outcomes in Report
Report shows how the content AI enriched, matched, and ranked performs: viewer engagement such as watch time and interactions, and performance by placement. The Real Time tab shows views, average watch time, and VCR as they come in.
Learn more about how to track content interactions in Report. The Relevance Dashboard shows how well each feed matched its viewer and what that did to engagement and revenue.
What stays in your control
AI in BCC is configurable, auditable, and transparent. Its recommendations are visible, explainable, and overridable at every stage. You keep control through:
- Brand and platform guidelines.
- Community-level relevance rules.
- Approval workflows.
- Activity logs and audit trails. Learn more about Audit Log, Two-Factor, and SSO for BCC.
For how Supervised Intelligence decides which changes wait for your approval, see Supervised Intelligence Autonomy and Approvals.
By type of business
- Retail media networks: AI improves ad relevance and shoppable video performance, and optimizes placements with shopper intent signals.
- Media brands and publishers: AI keeps editorial standards consistent at scale and turns audiences into active community members.
- Consumer brands: AI powers community-led storytelling and drives engagement and conversion from your owned channels.
Limitations
- Input quality bounds every model: scores, matches, and flags come from the transcript, the visual description, and metadata. Silent, music-only, or text-free videos get weaker results.
- One group per video: the classifier never places a video in two groups. Cross-post by hand.
- Clip length: Clip It clips are at most 120 seconds, and sources are up to 60 minutes.
- Uploads: up to 2 GB and 300 seconds, in 9:16.
- Product links: up to 3 per video, with an 80% minimum confidence in production.
- Strict policies reduce recommendations: strict moderation policies can lower the number of automatic recommendations.
- Plan-gated capabilities: some AI capabilities are tied to your plan. AI video generation is one of them.
- Human approval is always required: nothing publishes, and nothing is removed, without a person acting in Curate or Moderate.
