# AI for Marketing, Brand & Communications URL: https://justuju.org/role/marketing/ For your role AI makes copy, images, video and voice in minutes, so volume is no longer the hard part. The hard parts are staying on brand, owning what you publish and saying when it was made by a machine. ## How AI is changing this work - Production of variants for every channel, market and audience drops from weeks to hours. - Brand control moves into the tools: voice, style and approved claims are checked as content is made. - Rights and labels become daily questions: what was the model trained on, and must this be marked as AI-made? ## Key layers, most important first ### 1. Layer 9 · Applications & Interfaces URL: https://justuju.org/layer/applications-interfaces/ Why it matters: Marketing teams work in design, content and campaign tools. AI inside them is where briefs become finished assets. What to do here: - Brief to variants: Turn one approved concept into sized, localised versions for each channel. (see https://justuju.org/concept/embedded-ai/) - Design by conversation: Describe the layout or the edit in words, then adjust by hand. (see https://justuju.org/concept/chat-assistants/) - Pre-publish check: Run every asset through an automated check for voice, required wording and accessibility. (see https://justuju.org/concept/copilots/) Examples as of 2026-11-11: Adobe GenStudio for Performance Marketing (https://business.adobe.com/products/genstudio/performance-marketing.html); Canva AI (https://www.canva.com/canva-ai/); Jasper (https://www.jasper.ai/) ### 2. Layer 5 · Foundation Models URL: https://justuju.org/layer/foundation-models/ Why it matters: Image, video and voice quality come straight from the model, and so do the usage terms. Marketers now choose models the way they once chose photographers. What to do here: - Pick the model per job: Use different models for product shots, video and voice, and compare them on your own briefs. (see https://justuju.org/concept/specialised-models/) - Read the usage terms: Some models are cleared for commercial use and some are not, even inside the same app. (see https://justuju.org/concept/open-weights/) - Know its habits: Models drift toward the average look and tone. Strong references and examples pull them back. (see https://justuju.org/concept/pretraining/) Examples as of 2026-11-11: Adobe Firefly models (https://www.adobe.com/products/firefly/features/ai-video-generator.html); Veo 3.1 (Google) (https://ai.google.dev/gemini-api/docs/veo); ElevenLabs (voice) (https://elevenlabs.io/) ### 3. Layer 7 · Orchestration & Tooling URL: https://justuju.org/layer/orchestration-tooling/ Why it matters: A model does not know your brand. This layer hands it your voice, style rules and product facts on every request, and checks the output. What to do here: - Brand voice as context: Store voice, style guide and audience profiles once; apply them to every generation. (see https://justuju.org/concept/context-engineering/) - Approved facts only: Ground product claims in a maintained fact sheet so copy cannot invent features. (see https://justuju.org/concept/rag/) - Brand checks as tests: Score outputs against your guidelines automatically and send failures to a person. (see https://justuju.org/concept/guardrails-evals/) Examples as of 2026-11-11: Jasper Brand IQ (https://www.jasper.ai/brand-iq); Brand guidelines in Adobe GenStudio (https://experienceleague.adobe.com/en/docs/genstudio-for-performance-marketing/user-guide/guidelines/brands) ### 4. Layer 4 · Data URL: https://justuju.org/layer/data/ Why it matters: What a model was trained on decides whether its output is safe to publish. Provenance labels are becoming the way audiences and platforms tell what is real. What to do here: - Ask about training data: Prefer models whose maker states the training sources and stands behind commercial use. (see https://justuju.org/concept/data-provenance/) - Attach provenance: Add signed provenance data to published assets so their origin and edits can be checked. (see https://justuju.org/concept/data-provenance/) - Keep your own material yours: Check whether uploads, such as unreleased campaigns, may be used to train the supplier's models. (see https://justuju.org/concept/pretraining-data/) Examples as of 2026-11-11: C2PA Content Credentials (https://c2pa.org/faqs/); Adobe Firefly training-data statement (https://news.adobe.com/news/2025/02/firefly-web-app-commercially-safe) ## High-leverage use cases - Campaign variants: One approved concept becomes every size, channel and language, checked against brand rules. - First drafts: Blog posts, emails, social posts and press drafts in your voice, edited by a writer. - Product imagery and short video: Backgrounds, scenes and clips generated from real product shots. - Research and listening: Summaries of reviews, comments and competitor moves, with links to the source posts. - Localisation: Translation and adaptation per market, reviewed by a native speaker. ## Start here: the next 30 to 60 days - Week 1 to 2: Write your brand voice down in one page, with five good and five bad examples. Every tool needs it. - Week 3 to 4: Run one live campaign with AI-made variants beside human-made ones, and compare results. - Day 30 to 60: Set a publishing rule: which models are approved, who reviews, and when content is labelled as AI-made. - Day 30 to 60: Build a fact sheet of approved product claims and connect it to your writing tool. ## Risks and pitfalls - Rights in what you publish: Usage terms differ by model, even inside one app, and purely machine-made work may not be yours to protect. Check terms and keep a human author in the loop. - Labelling duties: In the EU, deepfakes must be labelled, as must AI-written text published to inform the public unless a person edited it and takes responsibility. Source: https://eur-lex.europa.eu/eli/reg/2024/1689/oj - Off-brand sameness: Unsteered models produce average-sounding copy. Without your voice and examples, everyone's content converges. - Invented claims: A model will happily write a feature or statistic that does not exist. Ground claims in approved facts. - Provenance is not proof: Content Credentials record origin and edits; they do not say whether content is true. Source: https://c2pa.org/faqs/ ## Sources - Adobe GenStudio for Performance Marketing (business.adobe.com): https://business.adobe.com/products/genstudio/performance-marketing.html - Canva AI (canva.com): https://www.canva.com/canva-ai/ - Jasper (jasper.ai): https://www.jasper.ai/ - Adobe Firefly models (adobe.com): https://www.adobe.com/products/firefly/features/ai-video-generator.html - Veo 3.1 (Google) (ai.google.dev): https://ai.google.dev/gemini-api/docs/veo - ElevenLabs (voice) (elevenlabs.io): https://elevenlabs.io/ - Jasper Brand IQ (jasper.ai): https://www.jasper.ai/brand-iq - Brand guidelines in Adobe GenStudio (experienceleague.adobe.com): https://experienceleague.adobe.com/en/docs/genstudio-for-performance-marketing/user-guide/guidelines/brands - C2PA Content Credentials (c2pa.org): https://c2pa.org/faqs/ - Adobe Firefly training-data statement (news.adobe.com): https://news.adobe.com/news/2025/02/firefly-web-app-commercially-safe - Regulation (EU) 2024/1689 (the AI Act): https://eur-lex.europa.eu/eli/reg/2024/1689/oj --- Justuju · https://justuju.org/ · Content CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Quote with a link to the page.