AI Usage Policy

How The West Africa Journal uses artificial intelligence, what it never does, and how readers know.

Version 1.0 · Public statement and internal rules · September 2026 · Owner: Editor of Record · Read with the Newsroom Handbook v1.1 · Reviewed quarterly.

1. Statement to Readers

The West Africa Journal uses artificial intelligence as a newsroom tool — to gather and sort what the world is reporting about Africa, to help draft and check articles against our published standards, to translate, and to produce editorial illustrations under editorial direction. AI does not decide what is true. Every fact WAJ publishes traces to a named primary source; every article is held to the Newsroom Handbook by editorial gates and by editors; every generated illustration is labelled “WAJ Studios / AI Illustration”. AI never invents a quote, a source, a statistic or a scene. When we get something wrong, we correct it openly and say so.

This policy is published at wajournal.com/ai-usage and applies to everyone who produces work for WAJ — staff, correspondents, contributors and the automated systems the newsroom runs.

2. Principles

• Accountability is human. An Editor of Record is responsible for every published piece. No system is accountable for anything; a person always is.

• Truth comes from sources, not models. AI may summarise, structure and check; it may not be the origin of any factual claim. The Sourcing & Verification SOP applies unchanged.

• Disclosure is standing, specific and visible. This policy, the credit line on every generated image, and a note on any piece where AI did more than assist.

• The standard does not lower. AI-assisted work meets exactly the same handbook — structure, voice, sourcing, fairness, sensitivity — and passes the same gates.

• Cost discipline is editorial discipline. The newsroom runs on very little. Cheaper tools are preferred wherever quality holds, and every automated step states what it costs.

3. What AI Does at WAJ

• News judgement — How AI is used: Gathers headlines from approved sources as signals of importance; scores clusters by severity, breadth, watchlist and how many major outlets carry them. The human control: Editors set the weights, the watchlist and the source tiers; rate stories up or down; can overrule any ranking. The morning coverage-gap report is read by a person.

• Research — How AI is used: Assembles a fact table from primary sources — documents, filings, datasets, agency reports — with each claim linked to its source. The human control: Only verified claims reach the writer; a claim without a source is discarded.

• Drafting — How AI is used: Drafts articles in the Handbook's three-part structure and institutional voice from the verified fact table. The human control: The editor gate checks structure, length, voice, attribution and plagiarism; failures return for rewrite; nothing publishes on a safety net alone.

• Editing and checks — How AI is used: Deterministic checks for structure, word limits, exhortation, banned constructions, plagiarism, sensitivity flags. The human control: Editors read what publishes; ratings feed back into the system; corrections follow the Corrections Policy.

• Illustration — How AI is used: WAJ Studios concept and render, under the Visual Standards, gated before and after generation. The human control: Credit line on every image; an editor can reject or replace any image; sensitive stories in restrained mode.

• Translation and accessibility — How AI is used: French editions, captions, alt text. The human control: Proper nouns and cultural terms preserved; spot-checked by editors.

4. What AI Never Does

• Invent a fact, a figure, a quote, a document, a person, a source or an event.

• Present its own output as a primary source, or cite another outlet's text as WAJ reporting.

• Produce a photorealistic image of a real, identifiable person, or any image that could be mistaken for a photograph of a real event.

• Produce images that reproduce racist caricature, sexualise or infantilise subjects, dramatise suffering, or use flags, crests, maps and other lazy signifiers as the picture.

• Publish without passing every editorial gate. Output that cannot be parsed or verified is discarded, never “tidied up” and published.

• Decide the sensitivity treatment of a story involving death, conflict, disease or children on its own — a flag is derived deterministically and restrained mode is enforced.

• Alter a published article silently. Every change after publication follows the Corrections Policy.

5. Disclosure — how readers know

• Every generated illustration — Credit line: “WAJ Studios / AI Illustration”. Human-made house graphics say “Illustration: WAJ Studios”. The two are never confused because the credit is derived from the image's recorded origin.

• This policy — Linked from the site footer, the About page and the AI Usage Policy page.

• Articles — Standard reporting drafted with AI assistance under editorial control carries no per-article label beyond this policy. Where AI did more than assist — a full translation, a data analysis — a one-line note at the foot says so.

• Metadata — Generated images carry the industry AI-generated marker where the pipeline supports it. This is in addition to the visible credit, never instead of it.

6. Rules for People

Staff and contributors

• You may use AI to research, outline, transcribe, translate, check and tidy. You may not use it to generate quotes, facts, sources or scenes.

• You remain responsible for every word under your byline. “The tool said so” is not a source.

• Disclose AI assistance in your submission note, even if the answer is “none”. Undisclosed use is treated as fabrication.

• Never paste confidential documents, unpublished names, or sources' identities into a third-party AI tool. Use the newsroom's approved tools only.

• Do not use AI to write in the voice of a real person, or to produce images or audio of a real person.

Campus correspondents

The Video Submission Handbook v1.1, Section 7E, applies: no AI-generated or altered footage, audio or images presented as real; every use of AI disclosed at submission; stills follow the Visual Standards.

7. Governance

• Owner. The Editor of Record owns this policy and the newsroom's automated gates.

• Change control. Any change to a model, a prompt or a gate is tested offline against held-out articles before it goes live, and recorded in the editorial rulebook with its result.

• Learning loop. The system proposes weight changes weekly from its own scoreboard; an editor approves or declines each one; a kill switch freezes all learning to the last approved state.

• Vendors. Tools and providers are listed in the rulebook with what each is used for and what it costs. No provider is used for a purpose it is not listed for.

• Review. This policy is reviewed quarterly and whenever a tool changes.

8. Complaints

A reader who believes an AI-assisted piece is inaccurate, unfair or undisclosed writes to corrections@wajournal.com. Complaints are acknowledged within two working days and handled under the Corrections & Accountability Policy. The complaint, the finding and the action are logged.