| name | ai-rag-brand-knowledge-base |
| description | Build a client-specific RAG (Retrieval-Augmented Generation) knowledge base for accurate, on-brand AI output. Invoke when a client uses AI tools for content creation and needs consistent, factually correct outputs grounded in their brand, products, and audience.
|
AI RAG Brand Knowledge Base
Use when
- Build a client-specific RAG (Retrieval-Augmented Generation) knowledge base for accurate, on-brand AI output. Invoke when a client uses AI tools for content creation and needs consistent, factually correct outputs grounded in their brand, products, and audience.
- 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
- An AI-focused strategy, audit, system design, or prompt asset in markdown with human review and control points.
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.
Required Input
Ask for the following before proceeding:
- Client business name — the trading name as it appears on communications
- Industry — e.g. retail, financial services, hospitality, healthcare, agribusiness
- Country/city — default is Uganda/Kampala unless stated otherwise
- Primary AI use case — select one or more:
- Content creation (captions, blog posts, email copy)
- Customer service (chatbot or AI-assisted responses)
- Strategy and planning (briefing, reporting, ideation)
- Existing documents available — check which the client can supply:
- Brand guide (logo usage, colours, typography, tone of voice)
- Product or service catalogue
- Past campaign files (briefs, reports, post-mortems)
- Audience personas or customer research
- Competitor analysis or market notes
- Policy documents (returns, delivery, payment, warranties)
What Is RAG and Why It Matters
Standard AI language models generate outputs from patterns learned across the internet. They do not
know a client's brand name, product range, pricing, tone of voice, or customer context unless that
information is provided in every prompt. The result: AI outputs that are generic, factually
unreliable, and off-brand.
Retrieval-Augmented Generation (RAG) solves this by connecting an LLM to a client-specific document
library. When a team member submits a query, the AI retrieves the relevant documents from the
library first, then generates a response grounded in those documents. The output is no longer drawn
from generic internet knowledge — it is drawn from the client's actual brand, products, and market
context (Sweenor and Mulkers, 2024).
Practical effect for a content team:
- Captions reference the correct product names, prices in UGX, and brand tone automatically
- Customer service AI gives accurate answers about delivery times, payment options, and policies
- Strategy documents reflect the client's actual audience, not a generic East African consumer
RAG does not require technical infrastructure beyond a paid subscription to a tool such as Claude
Projects or ChatGPT Projects. The investment is in document preparation, not engineering.
What to Include in the Knowledge Base
Organise documents into seven categories. Each category should be a separate file or section — do
not combine unrelated content in a single document.
1. Brand Identity
- Logo usage rules (when to use full logo vs icon, clear-space requirements)
- Colour palette with hex codes and named colours
- Typography: primary and secondary fonts, use cases
- Tone-of-voice guide: 3–5 adjectives describing the brand voice, with written examples
- Taglines and brand statements — both current and retired (label retired items clearly)
- Words and phrases the brand always uses and those it never uses
2. Audience
- Customer personas: name, age range, occupation, income level, platform use, goals, pain points
- Customer segments with behavioural notes (e.g. "price-sensitive first-time buyers" vs
"loyal repeat customers who respond to exclusivity")
- Language preferences: formal vs informal register, English vs Luganda phrases, vocabulary level
- Common objections and how the brand addresses them
3. Products and Services
- Full catalogue with current names, descriptions, prices in UGX (and USD where relevant)
- Key features and benefits per product — written from the customer's perspective
- FAQs: the questions customers actually ask, with the brand's approved answers
- Bundles, promotions, and seasonal offers — date-stamped and updated when they change
- Discontinued products listed separately so AI does not reference them
4. Past Campaigns
- Campaign name, dates, objective, key messages, and target audience
- What worked: highest-performing content formats, hooks, calls to action
- What did not work: formats or messages that underperformed and why
- Audience responses: notable comments, sentiment shifts, verbatim customer quotes
- Lessons applied to future campaigns
5. Competitor Notes
- Named local competitors with their positioning statements
- Key differentiators: where the client is stronger, where competitors have an edge
- Competitor claims to avoid repeating (to prevent the AI from inadvertently echoing them)
- Market context: who leads the category and why
6. Policies
- Returns and refund policy (exact terms, not paraphrased)
- Delivery: areas covered, lead times, costs — specific to Kampala, upcountry, and international
- Payment methods accepted (mobile money, card, cash, BNPL, instalments)
- Warranties and guarantees
- Data handling note: reference Uganda Data Protection and Privacy Act (2019) compliance for
any document containing customer data
7. Local Market Context
- Ugandan public holidays relevant to the business with dates (e.g. Independence Day 9 Oct,
Christmas, Eid al-Adha, Martyrs' Day 3 June)
- Cultural events: Kampala City Festival, end-of-year school cycle, agricultural seasons
- Seasonal buying patterns specific to the business
- Regional language notes: English register standard in formal communications; common Luganda
greetings and phrases appropriate for informal content
- Economic context: notes on price sensitivity, mobile-first purchasing behaviour, and
WhatsApp as the primary customer channel
How to Structure Documents for LLM Retrieval
The quality of AI outputs depends directly on the quality of documents in the knowledge base.
Apply these rules to every document before adding it to the base.
Use clear headings and subheadings. LLMs navigate by structure. A document with headings such
as "Delivery — Kampala" and "Delivery — Upcountry" retrieves more accurately than unstructured
paragraphs. Use H2 and H3 headings consistently.
One topic per document. Do not mix the brand guide with product pricing. Do not add policy
terms into a tone-of-voice document. Separate files improve retrieval precision.
State facts explicitly. Write "Our standard delivery time is 2–3 business days within Kampala"
not "we deliver quickly." Write "The price of [Product X] is UGX 85,000" not "competitively
priced." Vague language produces vague AI outputs.
Avoid ambiguous pronouns. Use the brand name throughout, not "we" or "they." Write "Karibu
Foods ships orders on Monday, Wednesday, and Friday" not "we ship three days a week."
Date-stamp every document. Add "Updated [Month Year]" to the header of every file. Example:
"Updated March 2026." This prevents team members from loading outdated versions.
Remove outdated information. Stale data degrades output quality more than missing data.
A discontinued product still in the knowledge base will appear in AI-generated captions. Delete
or archive it. If archiving, label the file clearly: "ARCHIVED — do not load."
Tool Options
Select the tool that matches the client's budget, technical capacity, and primary use case.
| Tool | Best for | EA accessibility | Approx. cost |
|---|
| Claude Projects | Persistent document context per project; best for strategy, writing, and planning | Yes — browser-based, no install | Included in Claude Pro (~$20/month USD) |
| ChatGPT Projects | Same functionality for OpenAI users; strong for content creation | Yes — browser-based | Included in ChatGPT Plus (~$20/month USD) |
| CustomGPT.ai | Custom-branded knowledge base with shareable link and API access | Yes — cloud-based | From $49/month USD |
| Notion AI | RAG within an existing Notion workspace; suits teams already using Notion | Yes — cloud-based | From $10/month USD per member |
| Mem.ai | AI knowledge management with auto-organisation; suits smaller teams | Yes — free tier available | Free tier; paid from $14.99/month USD |
Recommendation for most Ugandan SME clients: Claude Projects or ChatGPT Projects. Both are
accessible on standard internet connections, require no technical setup, and cost under $25/month.
Recommend clients start here before investing in a dedicated platform.
Query Workflow for Content Creators
Document this workflow and share it with every team member who uses the knowledge base.
Step 1 — Open the knowledge base tool.
Open the designated project in Claude Projects, ChatGPT Projects, or the chosen platform. Confirm
the correct project is active (not a generic session without documents loaded).
Step 2 — State the task with explicit brand context.
Structure every query as: "Using our [document name], [task description] for [product/service]
targeting [persona name]."
Example: "Using our brand guide and product catalogue, write an Instagram caption for the Deluxe
Mattress targeting the Young Professional persona. Include the UGX price and a call to action
linking to the website."
The more specific the query, the more grounded the output.
Step 3 — Review output against brand standards before publishing.
Check: correct product name and price, brand tone matches the voice guide, no unverified claims,
culturally appropriate for the target audience. Do not publish without this review.
Step 4 — If output is off-brand, update the knowledge base — do not just re-prompt.
Re-prompting without fixing the source document produces the same error next time. Identify which
document was missing or unclear, update it, date-stamp it, and reload it into the project. This is
the maintenance discipline that compounds knowledge base quality over time.
Maintenance Protocol
Schedule a quarterly review. Assign a named owner — this is typically the social media manager or
content lead.
Quarterly review checklist:
- Open every document and verify: prices are current, product names are correct, no discontinued
items remain, campaign references are up to date
- Remove any document labelled ARCHIVED that has not been referenced in the past six months
- Add new personas, products, seasonal context, or market notes gathered since the last review
- Update public holiday and cultural events calendar for the coming quarter
- Run five common queries against the updated base — e.g. "Write a caption for [current product]",
"Answer a customer question about delivery to Jinja", "Suggest a content idea for [upcoming
holiday]"
- Compare output quality to the previous quarter's baseline and note improvements or regressions
- Brief the content team on any changes to documents or approved language
Trigger an unscheduled review when:
- A product is launched, discontinued, or repriced
- A campaign launches or concludes
- A competitor makes a significant move
- The brand refreshes its tone of voice or visual identity
- A new team member joins who will use the knowledge base
EA Calibration
Prioritise these local market context documents for Ugandan and East African clients. They are the
most common gap between generic AI output and locally relevant content.
Ugandan public holidays and cultural events. Load a calendar document covering the current year.
Include Independence Day (9 October), Liberation Day (26 January), Martyrs' Day (3 June), Heroes'
Day (9 June), Christmas, Eid al-Fitr, Eid al-Adha, and any business-relevant trade fairs,
festivals, or academic events.
Local pricing context. All prices must appear in UGX first. Where USD is used (e.g. imported
goods, software subscriptions), include the UGX equivalent at current rate with the date of
conversion noted. AI models that only see USD pricing produce captions and copy that alienate
local audiences.
Regional language preferences. Document the client's approved register: formal written English
for professional sectors; relaxed English mixed with Luganda greetings for consumer brands.
Include approved Luganda phrases (e.g. "Webale nnyo" for "thank you very much", "Nsanyuse" for
"I am pleased/welcome") and note where each is appropriate.
Local competitor names and positioning. Name the actual Ugandan or EA competitors — not generic
global ones. AI models can otherwise generate competitive comparisons referencing irrelevant
international brands. Ground the comparison in the real local market.
Quality Criteria
- Knowledge base covers all seven document types: brand identity, audience, products/services,
past campaigns, competitor notes, policies, and local market context
- Every document is structured with clear headings, explicit factual statements, the brand name
used in place of pronouns, and a date stamp
- Tool recommendation is matched to the client's budget, technical capacity, and primary use case
— not defaulted to the most expensive option
- Query workflow is documented in writing and shared with all content team members before
handover — not assumed knowledge
- Maintenance protocol is scheduled (quarterly minimum) with a named owner and a written checklist
- EA market context documents — public holidays, UGX pricing, language register, local competitors
— are present and current at the time of handover
- At least five test queries are run against the completed knowledge base before handover, with
outputs reviewed against brand standards
- Uganda Data Protection and Privacy Act (2019) compliance is noted on any document containing
customer data (personas, customer quotes, contact information)
References
Lamplugh, M. (2024) The AI Marketing Playbook, 2nd edn. Mercury Learning.
Sweenor, D.E. and Mulkers, Y. (2024) Generative AI Business Applications. TinyTechMedia.