| name | training-ai-foundations |
| description | Produces an AI literacy training guide for client teams who are entirely new to AI in marketing — covering the augmented intelligence model, the three AI types (Mechanical/Thinking/Feeling), what AI can and cannot do in the East African context, hands-on exploration of five free tools, and the human quality standard for editing AI output. Invoke when the user says "create an AI foundations training guide", "write an AI basics workshop for my team", "train my staff on AI for marketing", "AI literacy training for beginners", or needs a structured introductory training document for client employees with no prior AI experience. Primary sources: Anderson, D. (2022) AI in Digital Marketing Training Guide; Ltifi, M. (ed.) (2025) Advances in Digital Marketing in the Era of AI. |
AI Foundations for Marketing Teams — Training Guide
Use when
- Produces an AI literacy training guide for client teams who are entirely new to AI in marketing — covering the augmented intelligence model, the three AI types (Mechanical/Thinking/Feeling), what AI can and cannot do in the East African context, hands-on exploration of five free tools, and the human quality standard for editing AI output. Invoke when the user says "create an AI foundations training guide", "write an AI basics workshop for my team", "train my staff on AI for marketing", "AI literacy training for beginners", or needs a structured introductory training document for client employees with no prior AI experience. Primary sources: Anderson, D. (2022) AI in Digital Marketing Training Guide; Ltifi, M. (ed.) (2025) Advances in Digital Marketing in the Era of AI.
- Use this skill when it is the closest match to the requested deliverable or workflow.
Do not use when
- Do not use this skill for graphic design, video production, software development, or legal advice beyond the repository's stated scope.
- Do not use it when another skill in this repository is clearly more specific to the requested deliverable.
Workflow
- Collect the required inputs or source material before drafting, unless this skill explicitly generates the intake itself.
- Follow the section order and decision rules in this
SKILL.md; do not skip mandatory steps or required fields.
- Review the draft against the quality criteria, then deliver the final output in markdown unless the skill specifies another format.
Anti-Patterns
- Do not invent client facts, performance data, budgets, or approvals that were not provided or clearly inferred from evidence.
- Do not skip required inputs, mandatory sections, or quality checks just to make the output shorter.
- Do not drift into out-of-scope work such as code implementation, design production, or unsupported legal conclusions.
Outputs
- A structured markdown document, plan, playbook, or strategy ready for client-facing or internal use.
References
- Use the inline instructions in this skill now. If a
references/ directory is added later, treat its files as the deeper source material and keep this SKILL.md execution-focused.
How to Use This Skill
Collect the Required Input below. Then generate the full training guide across four modules, substituting all bracketed placeholders with the client's specific details. Output is a complete, facilitator-ready training document — not a slide deck. For a slide deck version, use deck-strategy-presentation as a model and build slides separately.
Required Input
Ask for the following before generating the training guide:
- Client business name — trading name of the business
- Industry — sector (e.g. FMCG, hospitality, professional services, healthcare)
- Country / city — default Uganda / East Africa
- Primary goal — what the client wants the team to achieve after training
- Team size and prior AI experience — number of participants; experience level: none / basic / intermediate
- Primary platforms used — which platforms the business is active on (e.g. Facebook, Instagram, WhatsApp)
- Training format — in-person / virtual / self-guided handout
- Time available — 2-hour express / half-day full / spread across 4 weekly sessions
Output: Complete Training Guide
Generate the following four modules in full. Use the client's name, industry, platforms, and city throughout. Write in plain English — no jargon. Tone: practical, encouraging, honest.
Training Overview
Programme: AI Foundations for Marketing Teams
Total Duration: Approximately 2.5 hours (150 minutes) — or adapt to time available
Audience: Marketing and communications staff with no prior AI experience
Format: [Insert training format]
Prepared for: [Client Business Name]
Industry: [Industry]
Primary Sources: Anderson, D. (2022) AI in Digital Marketing Training Guide (Self-published); Ltifi, M. (ed.) (2025) Advances in Digital Marketing in the Era of AI (CRC Press); Farri, O. and Rosani, M. (2025) Co-Intelligence: Working and Learning with AI; Nayebi, H. (2025) AI-First Marketing
The Core Frame: Augmented Intelligence, Not Artificial Intelligence
Open every training session with this frame (Ltifi, 2025). Do not skip it.
The correct mental model for AI in marketing is augmented intelligence — using AI to amplify human creativity, insight, and relationships, not to replace them.
Present this as three columns:
| Human | AI | Together |
|---|
| Strategy and direction | Research and drafting | Faster production |
| Brand voice and tone | Scheduling and publishing | Greater consistency |
| Cultural insight | Pattern recognition | More responsive output |
| Emotional intelligence | Volume generation | Human quality at scale |
| Relationship-building | Data summarisation | More time for the work that matters |
The junior assistant analogy (use this in the room): AI is like a junior assistant who never sleeps, has read everything on the internet, but has no taste, no judgement, and no cultural intelligence. You would not publish what a junior assistant wrote without reading it first. Apply exactly the same discipline to AI output.
This framing sits between two wrong narratives:
- The fear narrative: "AI will take our jobs." Wrong — AI cannot do what humans do best.
- The hype narrative: "AI can do everything." Wrong — raw AI output is mediocre without human editing.
Establish this frame clearly at the start. Return to it throughout the session.
Module 1: What Is AI? (30 minutes)
Learning Objective
Explain how AI language tools work in plain English — and correct the most common misconceptions before they take root.
1.1 How Large Language Models Work
Explain the following to participants. Use plain language throughout:
AI tools such as ChatGPT and Gemini are language prediction engines. They were trained on enormous amounts of text — books, websites, articles — and they learned the patterns of how words follow one another. When you type a question, the AI predicts the most probable next word, then the next, then the next, until a response is formed.
This means:
- AI does not know things — it predicts patterns
- AI does not understand your question — it matches it to patterns it has seen
- AI does not think — it calculates probability
- AI does not have opinions, feelings, or cultural awareness
Why does this matter? Because it explains why AI output is often plausible-sounding but wrong, generic, or culturally off.
Reference: Anderson (2022) — the GIGO principle (Garbage In, Garbage Out) applies directly. Vague instructions produce vague output; no context produces no-context output.
1.2 The Three Types of AI in Marketing
Explain the three AI types (Ltifi/Huang & Rust, cited in Ltifi, 2025) using marketing examples for each. All three exist in tools the team may already use.
1. Mechanical AI — automates repetitive tasks
What it does: schedules posts, sends automated replies to FAQs, generates basic reports, resizes images for different formats.
Marketing examples:
- Buffer or Hootsuite auto-scheduling posts at optimal times
- A WhatsApp chatbot that replies "Thank you, we will be in touch shortly" to every enquiry
- Canva's background remover or image resizing tool
EA reality: this is the most accessible AI type in East Africa. Free tools deliver functional Mechanical AI today.
2. Thinking AI — analyses data and generates recommendations
What it does: identifies patterns in large datasets, segments audiences, predicts which content will perform, surfaces insights from analytics.
Marketing examples:
- Facebook's Advantage+ automatically distributing budget to the best-performing ad sets
- Mailchimp recommending the best send time based on audience behaviour
- A tool suggesting which content topics generate the most engagement for your page
EA reality: Thinking AI is available through platforms most businesses already use (Meta, Mailchimp, Google). It works in the background — you may already be using it.
3. Feeling AI — personalises emotional tone and detects sentiment
What it does: adjusts tone to match the emotional register of a conversation, detects whether customer messages are positive, negative, or frustrated, responds with appropriate empathy.
Marketing examples:
- A chatbot that detects a complaint and switches from a cheerful tone to an apologetic one
- Sentiment analysis tools that flag negative brand mentions across social media
- Email personalisation that adjusts the warmth of a message based on customer history
EA reality: Feeling AI is emerging and less reliable in local language contexts. Luganda, Swahili, and regional dialects are not well served by current sentiment tools. Always apply human judgement to tone and customer relations in the EA market.
1.3 Common Misconceptions — Address These Directly
Do not assume participants have no misconceptions. Walk through each one:
"AI is only for big companies."
Wrong. Free tiers of ChatGPT, Gemini, and Canva are functional for any business. A market stall owner with an Android phone can use ChatGPT today.
"AI will produce perfect content."
Wrong. Raw AI output is mediocre. It is a first draft, not a finished post. The skill is in the editing, not the generation.
"AI knows my market."
Wrong. AI was trained on the global internet, which is predominantly English-language, Western, and urban. It does not know Kampala in 2026. It does not know what happened in your market last week. It does not understand local cultural tensions or community dynamics. This training data bias is not a setting that can be adjusted — it is the data the AI learned from. It affects not just text but imagery: AI tools default to Western-centric, gender-stereotyped representations of people. Any AI-generated content depicting East African people, places, or communities must be reviewed by a human with direct cultural knowledge before publication (Source: Ching & Mothi, 2025).
"AI is always right."
Wrong. AI hallucination is well-documented — AI invents facts, statistics, names, and sources with complete confidence. Any fact AI produces must be verified before publishing. Never publish an AI-generated statistic without checking the source.
Module 2: What AI Can and Cannot Do in Marketing (45 minutes)
Learning Objective
Give the team a clear, honest map of where AI adds value and where it fails — calibrated specifically to the East African market.
2.1 What AI Can Do Well
Present this as a working list. Add client-specific examples where possible:
- Generate first drafts of captions, blog posts, and emails at speed (requires human editing before publishing)
- Suggest content ideas from a brief or content pillar
- Schedule and publish content automatically at pre-set times
- Respond to common FAQ messages via automated chatbot replies
- Analyse large datasets and surface patterns (e.g. which posts drive the most link clicks)
- Translate content between languages — with limitations (see 2.2)
- Research topics, competitors, and industry trends quickly
- Resize and reformat content for different platforms
- Generate multiple variations of a caption or headline for comparison
- Summarise long documents, meeting notes, or reports
2.2 What AI Cannot Do
This section is as important as 2.1. Be specific:
- Read local cultural tensions. AI cannot tell you who is offended by what in Kampala in 2026. Cultural nuance, tribal sensitivities, political tone, and community dynamics require a human who lives in that context.
- Produce genuinely novel ideas. AI recombines what exists. It does not invent. Creative breakthroughs still come from humans.
- Build trust or relationships. Customers know when they are talking to a machine. Trust is built through authentic human interaction — responses from a real person, unscripted employee moments, genuine community engagement.
- Replace authentic human voice. An employee talking honestly to camera about why they love their work cannot be replicated by AI. This is a competitive advantage that should be protected.
- Make ethical judgements. AI will produce content that is offensive, inaccurate, or inappropriate if not supervised. Ethical responsibility remains with the humans who use it.
- Know what happened last week. AI tools have knowledge cutoffs. They do not know about last month's news, competitor launches, or local events. Always brief the AI with current context.
- Understand Luganda, Swahili, or regional dialects reliably. Current AI tools perform poorly in local EA languages. Translated output requires review by a fluent speaker before use.
- Guarantee factual accuracy. Always verify any fact, statistic, or claim AI produces. This is non-negotiable.
2.3 The Four-Phase Creative Process Model
Source: Ching & Mothi (2025). Use this model to map AI's role at each stage of the content creation process. The key teaching point: human oversight is most essential at Production and Realization — the phases where content goes live.
| Phase | What Happens | AI Role | Human Role |
|---|
| Ideation | Generating content concepts and campaign ideas | Generates volume — ideas, angles, topic lists, formats | Selects the best ideas; applies cultural and brand judgement |
| Development | Turning ideas into drafts and testing them | Provides iterative feedback; generates multiple draft variations | Directs the development; edits and approves drafts |
| Production | Creating final content assets for publication | Automates repetitive tasks: resizing, captioning, translating | Reviews every piece before it goes live — non-negotiable |
| Realization | Distributing content and personalising it to audiences | Personalises at scale; optimises timing and targeting | Monitors output; corrects bias; responds to cultural moments |
Why this matters for East Africa: At the Realization phase, AI personalisation tools may serve different content to different audience segments in ways that reflect training data bias. A human must monitor distribution and confirm that content is reaching the right audiences without discriminatory or culturally inaccurate targeting.
2.4 The Platform Table — AI Applications for East Africa
Use this table when applying AI to the client's specific platform mix. Discuss only the platforms the client uses:
| Platform | AI Use | EA Note |
|---|
| Facebook | AI captions, scheduling, boosted post targeting | Meta Advantage+ works with UGX budgets; available to small businesses |
| Instagram | AI captions, Reels scripts, hashtag research | Reels perform best at 30–60 seconds; AI scripts need local editing |
| WhatsApp | Automated responses via ManyChat, broadcast scheduling | Africa's Talking for SMS/WhatsApp integration; native AI not yet available |
| TikTok | AI script suggestions, auto-captions, trend research | Low data cost; zero-rated in some EA markets; trend cycles move fast |
| YouTube | AI transcription, chapter generation, description writing | YouTube Studio auto-captions useful for accessibility; review before publishing |
| Email | AI subject line testing, send-time optimisation | Mailchimp free tier includes basic AI suggestions; functional for small lists |
| LinkedIn | AI post drafting, profile optimisation | B2B context; EA professional register differs from Western LinkedIn norms |
Module 3: AI Tools — Hands-On Exploration (45 minutes)
Learning Objective
Walk participants through five AI tools they can use immediately on their Android phones, all on free tiers, on a 3G connection or better.
Before You Begin
Confirm the following before the hands-on section:
- Every participant has a smartphone (Android or iPhone)
- WiFi or mobile data is available (minimum 3G for ChatGPT/Gemini; WiFi recommended for Canva image generation)
- Participants have a Google account (required for Gemini)
- Participants have created a free ChatGPT account (or do this as the first activity)
Do not push paid plans at any point in this session. Every activity runs on free tiers.
Tool 1: ChatGPT (chat.openai.com)
What it does: Text generation, idea generation, drafting, editing, research, summarisation.
How to access: Browser or app; free account required. Works on 3G.
Hands-on activity:
Ask participants to type the following into ChatGPT — substituting their own product and business name:
"Write a Facebook caption for [Business Name] using the PAS framework (Problem → Agitate → Solution). The product is [best-selling product]. Target audience: [brief audience description]. Under 150 words. End with a clear call to action."
Then ask: "Now rewrite the same caption in our brand voice — [describe brand voice in 2–3 words]."
Key lesson: The first output will be generic. The second will be closer. The published version still needs a human editor to add one specific local detail and remove any AI-sounding phrases. Generation is the easy part; editing is the skill.
Tool 2: Google Gemini (gemini.google.com)
What it does: Similar to ChatGPT; integrates with Google Workspace (Docs, Sheets, Gmail). Useful for teams already using Google tools.
How to access: Browser or app; Google account required. Works on 3G.
Hands-on activity:
Ask participants to open a competitor's website or Facebook page. Then type into Gemini:
"Summarise this business in 3 sentences: [paste the competitor's About section or page description]. Then identify 3 things [Client Business Name] does differently or better."
Key lesson: Gemini can accelerate competitor research significantly. The output is a starting point — verify the differentiators against real knowledge of the market.
Tool 3: Canva AI (canva.com)
What it does: Magic Write for copy generation; text-to-image; background remover; Magic Eraser. Integrated into Canva's design environment.
How to access: Browser or app; free Canva account. Image generation is data-heavy — use WiFi.
Hands-on activity:
Open a new Canva social media post. Click Magic Write (the AI icon). Type:
"Write 3 headline options for a [platform] post promoting [product]. Each headline under 10 words. Tone: [brand tone]."
Copy the best headline into the design.
Key lesson: Canva AI captions require the same editing discipline as ChatGPT. Magic Write is a first-draft tool, not a final-copy tool. Background remover and Magic Eraser are the most immediately useful AI features for small businesses with limited photography budgets.
Tool 4: FeedHive (feedhive.com)
What it does: AI-powered social media scheduling, content recycling, performance prediction, and content variation suggestions.
How to access: Browser; free tier available. Works on 3G for scheduling; data use is moderate.
Hands-on activity:
Connect one social media account. Use FeedHive's AI suggestions as a starting point to plan one week of posts. Review the suggestions and replace any generic content with client-specific posts.
Key lesson: FeedHive's AI suggestions are a useful prompt — not a publishing queue. Treat them as content ideas to react to, not content to approve without review. The scheduling and recycling features deliver the most immediate value.
Tool 5: Otter.ai (otter.ai)
What it does: AI meeting transcription, action item extraction, conversation summarisation.
How to access: Browser or app; free tier allows 300 minutes of transcription per month. Works on 3G for recording; transcription uploads require data.
Hands-on activity:
Ask one participant to record a 2-minute spoken pitch for the business's best-selling product or service. Upload the recording to Otter.ai. Review the auto-transcript and highlight the 3 strongest selling points.
Key lesson: Otter.ai is particularly useful for extracting content from customer conversations, team meetings, and interviews. The selling points extracted from a spoken pitch can become caption copy, email content, or a FAQ answer — without writing from scratch.
EA-Specific Tool Notes
- All 5 tools function on Android — no iOS required
- Free tiers are sufficient for most small-business use cases; do not commit to paid plans before testing free tiers thoroughly
- WhatsApp automation requires a third-party tool: ManyChat (manychat.com) or Africa's Talking (africastalking.com). Native WhatsApp AI is not yet available
- Low-bandwidth tip: ChatGPT and Gemini work on 3G. Canva image generation is data-heavy — use WiFi. FeedHive is moderate. Otter.ai uploading requires a stable connection
- All 5 tools are available in Uganda without VPN
Module 4: The Human Quality Standard (30 minutes)
Learning Objective
Teach participants to recognise AI-generated text and apply a consistent editing process that brings output up to human quality before publishing.
4.1 The 5 Signs of AI-Generated Text
Train the team to spot these patterns in any content before it goes live:
- Excessive hedging — phrases like "it's important to note that", "it's worth mentioning", "of course", "certainly". These signal AI caution, not human confidence.
- No concrete examples — AI produces generalities. Human writing anchors ideas to specific people, places, prices, and dates. If a paragraph could apply to any business anywhere, it needs a specific detail added.
- Overly smooth transitions — "Furthermore", "Moreover", "In conclusion", "In today's fast-paced world". These are AI connective tissue. Remove them.
- Generic openings — AI almost always opens with the broadest possible statement about the topic. Human writers start in the middle of the action.
- No local specificity — AI does not know your neighbourhood, your customers' names, or the Kampala matatu that passes your shop. If the content could have been written for any city in any country, it has not been localised.
4.2 The 3-Step Edit Process
Apply this to every piece of AI-generated content before publishing:
Step 1: Read aloud.
Does it sound like us? Would a real person say this? If it feels stiff, formal, or generic when spoken, it needs editing. Trust your ear.
Step 2: Add one local or specific detail.
Insert one thing that is true of your business, your customer, or your location. A street name. A customer type. A local reference. A real price. This grounds the content in reality and removes the generic feel.
Step 3: Remove one cliché or AI-sounding phrase.
Find the weakest sentence — the one that could appear in any AI output for any business. Delete or rewrite it.
Three steps. This takes under 5 minutes per post and transforms the output.
4.3 The "Would I Say This?" Test
Before publishing any AI-generated content, ask the team to apply this test:
If you would not say this in a meeting, do not post it.
This catches: overly formal language, culturally inappropriate framing, claims the business cannot support, and tone that does not match the brand's real voice.
4.4 Banned Vocabulary
Train the team to remove these words from all AI-generated content. Their presence signals AI authorship to any reader who has spent time with these tools:
Banned: delve, tapestry, leverage (as a verb), foster, robust, seamless, synergy, game-changing, cutting-edge, innovative solution, dive into, in today's fast-paced world, it's important to note, certainly, of course.
Why this matters: These words appear in AI output because they appeared frequently in the training data. They are statistically common in corporate writing. They are not how real people talk, and they are not how authentic brands communicate.
Reference: For the full quality control process including a complete humanisation checklist, see the ai-content-humaniser skill.
4.5 The Approval Rule
Establish this as a non-negotiable standard:
No AI-generated content goes live without a human reading it in full.
Scheduling AI to post content it generated, without a human review step, is the fastest way to publish something embarrassing or incorrect. The efficiency gain of AI generation is real. The efficiency shortcut of skipping review is not worth the risk.
Workshop Exercises Summary
Exercise 1 — The Misconception Round (Module 1, 10 minutes)
Each participant writes down one thing they believed about AI before this session. Share and discuss which beliefs the training has changed. Facilitator records on a whiteboard or shared screen.
Exercise 2 — ChatGPT Caption (Module 3, 15 minutes)
Each participant writes a ChatGPT prompt for their own business using the PAS framework. Run the prompt, review the output, and apply the 3-step edit from Module 4. Share the before and after with the group.
Exercise 3 — Spot the AI (Module 4, 10 minutes)
Facilitator presents three short pieces of social media copy. Participants identify which one is unedited AI output and explain which of the 5 signs gave it away.
Related Skills
training-ai-prompt-writing — next-level training on the Alpha-Beta-Gamma-Delta-Epsilon prompt structure and copywriting frameworks; deliver this session after AI Foundationsai-content-humaniser — full quality control process, editing checklist, and banned vocabulary reference for AI-generated contentbrand-voice-ai-training — how to train AI tools on a specific brand voiceprompt-engineering-library — ready-made prompt templates for common marketing content typestraining-client-team — general social media team training workbook for content creation and community management
Co-Pilot vs Co-Thinker (Farri and Rosani, 2025)
The most important distinction for any marketing team new to AI:
Co-Pilot mode — AI handles speed tasks:
- Summarising documents and reports
- Drafting first versions of captions, emails, and briefs
- Generating slide content from bullet points
- Taking notes in meetings
- Formatting data into tables
Co-Thinker mode — AI acts as a thought partner for reflection-heavy work:
- Pressure-testing campaign strategy logic
- Mapping stakeholder perspectives the team may have overlooked
- Identifying assumptions in a brief that should be validated
- Framing the client's core marketing challenge as a solvable problem
- Generating alternative strategic options for evaluation
When to use which: Use Co-Pilot when you know what you want and need it done faster. Use Co-Thinker when you are not yet sure what the right answer is. Most marketing teams default to Co-Pilot only — they are leaving the most valuable AI capability unused.
Training exercise: Ask participants to list their last five AI interactions. Classify each as Co-Pilot or Co-Thinker. Discuss: what proportion were Co-Thinker? What would they have done differently?
The Three Waves of AI in Marketing (Nayebi, 2025)
Help participants understand where they and their clients currently sit:
Wave 1 — Automation (Most EA businesses today)
Rules-based tools that follow fixed instructions. Examples: scheduled social posts, auto-reply chatbots, email drip sequences. No learning or adaptation. Reliable but rigid.
Wave 2 — Predictive ML (Growing in EA)
Systems that learn from data and predict future behaviour. Examples: audience segmentation models, engagement rate prediction, A/B test optimisation, sentiment analysis. Requires sufficient data. Improves over time.
Wave 3 — Agentic AI (Horizon for EA)
Autonomous agents that perceive their environment, reason about it, decide on actions, and learn from outcomes — without waiting for a human prompt. Examples: a content agent that monitors trending topics and drafts posts for approval; a campaign agent that detects low engagement and automatically triggers a response.
Training exercise: Ask participants to identify one marketing activity in their business at each wave level. Where is the gap between Wave 1 and Wave 2? What data or tools would be needed to close that gap?
Quality Criteria
- Augmented intelligence framing is used consistently throughout — AI assists humans, it does not replace them; the junior assistant analogy is included
- All three AI types (Mechanical / Thinking / Feeling) are explained with Uganda/East Africa marketing examples
- The "What AI cannot do" section is specific and EA-calibrated — not a generic global list; Luganda/Swahili limitations and cultural intelligence gaps are named explicitly
- All 5 hands-on tools (ChatGPT, Gemini, Canva, FeedHive, Otter.ai) are verified as accessible on Android, free tier, and 3G connection; bandwidth guidance is included
- The human quality standard section includes the banned vocabulary list, the 5 signs of AI text, and the 3-step edit process
- Platform AI applications are presented as a table with EA-specific notes per channel — not a single generic list
- Output is structured so a non-technical marketing manager can facilitate the session without additional preparation
- British English spelling throughout; imperative language used in all instructions
