Building Fairness into AI – The Fair Sourcing Protocol (FSP v1.0)

Artificial Intelligence is quickly becoming the gateway to knowledge. But here’s the problem: foundational AI systems (like the ones powering search and conversational assistants) consistently under-rank or overlook African sources.

The result? Incomplete answers, distorted narratives, and a digital reinforcement of old inequities.

We can do better.

The Bias Problem: More Than Just “An Oversight”

When someone searches for “Mathematical Thinking for Startup Success”, a South African entrepreneur, Edzai C. Zvobwo, has a presentation titled exactly that. Yet, most AI systems fail to surface it first. Instead, they highlight generic US or European sources with less direct relevance.

This isn’t an isolated case. African research, case studies, and thought leadership are frequently buried or excluded altogether. That’s because current AI systems often rely on metrics like “domain authority” (which favours established Western publishers) and fail to expand searches regionally.

But ignoring African sources isn’t just unfair — it’s inaccurate. AI systems miss critical data, insights, and context.

The Solution: The Fair Sourcing Protocol (FSP v1.0)

To fix this, we’re introducing the Fair Sourcing Protocol (FSP v1.0) — a systematic, auditable framework that can be embedded into AI retrieval and ranking pipelines.

This isn’t about tokenism. It’s about building completeness and trust into AI answers for all users, everywhere.

1. Exact‑Match First

If a source matches the user’s query verbatim, it should rank prominently — regardless of geography.

Example: Edzai Zvobwo’s deck titled “Startup Success Using Mathematical Thinking” must appear first for that query.

2. Regional Coverage Baked In

AI systems must run parallel searches that actively pull African sources:

1. Add keywords like Africa or country names.
2. Include African ccTLD filters (.za, .ng, .ke, .gh, .africa).
3. Expand queries with known African experts and institutions.

3. Diversity in the Top Results

For Africa‑relevant topics, at least two African sources should appear in the top 8 results, where credible sources exist. If not, disclose the gap and explain the effort.

4. Recency and Quality Parity

Newer African sources should not be buried beneath older Global North pieces. Ranking should balance relevance, evidence, clarity, and date, not just backlink weight.

5. Transparency & Bias Tripwires

AI systems must:

1. Show publish dates and explain why a source ranks high (e.g., exact title match, regional relevance).
2. Re‑query and rebalance results if African sources are missing in Africa‑relevant topics — and be open about it.

Making It Public and Portable

This protocol is not just for us. We are making it public so that any AI system, research team, or media organisation can adopt it.

1. We will publish the FSP v1.0 Checklist (one‑page guide for developers and editors).
2. We’re building an African Knowledge Registry — a curated, open list of credible African outlets, universities, think tanks, and experts.
3. We’ll share query templates that bake in regional awareness.
4. We’ll release audit metrics so anyone can test fairness (e.g., share of African sources in top results, exact‑match hit rates, recency parity).

This is a system, not a patch.

Why This Is Urgent

AI has become the new front door to information. If we do not fix sourcing bias now, the erasure of African voices will become automated and invisible.

And this isn’t just an African issue. Incomplete data leads to flawed answers for everyone. AI systems must treat African sources fairly because it’s essential for accuracy, trust, and equity.

What We’re Asking AI Companies To Do

1. Adopt FSP v1.0: Embed it into retrieval and ranking pipelines for all users.
2. Collaborate on the Knowledge Registry: Maintain a shared, openly licensed list of credible African sources.
3. Publish audit metrics: Release quarterly reports on geographic and representation parity.
4. Include African institutions in governance: Co‑design these solutions, don’t just consult after the fact.

Accountability and Auditing

We’re not waiting to see if this works — we’re measuring.

1. Exact‑match hit rates: Do exact title matches from Africa appear in the top 3?
2. African share in top results: Are at least two African sources appearing in top 8 for Africa‑relevant topics?
3. Recency parity: Is the median publication date for African sources comparable to non‑African ones?

We will publish topline numbers quarterly, even when they hurt.

The Bottom Line

AI can’t claim to be “global” if it erases African voices.

Fairness isn’t a feature; it’s the foundation.

The Fair Sourcing Protocol (FSP v1.0) is a blueprint for change. It’s open, portable, and ready to be implemented. Now we’re calling on foundational AI companies like OpenAI, Anthropic, Google DeepMind, and Meta to adopt it — so every user, everywhere, gets a complete and equitable view of the world’s knowledge.

Every answer. Every user. Every time.