How journalists can verify sources suggested by AI
A research tool offers a neat statistic about podcast listening: “Sixty percent of campus students listen every week.” It looks ready for a story. First, though, a reporter needs to know which students were asked, when the survey happened, how many people answered and whether the research exists at all.
That figure is invented for this example. The reporting problem is real: an AI tool can suggest leads or summarise documents while also introducing a false date, a nonexistent source or a finding stripped of its limits. Its answer is a starting point for verification, not the final source.
Follow the number into the report
Before publishing the example percentage, find the actual report. Check the organisation, date, wording of the question and population behind the result. A survey answered by one student club does not automatically represent an entire campus.
Another article repeating the same figure is not necessarily a second measurement. If both stories point to one report, they may share the same source. Associated Press's discussion of verification in data journalism highlights the value of tracing information to its origin. An independent study can add perspective, but its date, participants and question still need comparison.
Consider a real figure from Türkiye: TurkStat’s 2026 household ICT survey reports internet use of 92.3% among people aged 16–74. It is not a rate of podcast listening or a rate for university students. If a tool turned the number into “92.3% of students listen to podcasts,” both the population and the measured behaviour would have changed, even though the number itself came from a real source. The linked bulletin is in Turkish; checking its definition and scope prevents the misleading claim.
Listen around the quotation
Listen while exploring this story
Tune into campus radio with non-stop independent music and culture.
An automated transcript can help locate a passage in a long recording. Before quoting it, listen to the original: a name may have been misheard, or a short extract may change the meaning of the surrounding answer. Accuracy includes context as well as the words themselves.
Associated Press's AI standards announcement says tools can assist with specific tasks while editorial judgment and verification remain with AP journalists. Those are AP's standards; each newsroom needs its own editorial and correction practices.
Does the image show the claimed event?
A genuine photograph can still be shared with the wrong date or location. Investigate the earliest available version, the claimed place and other records of the event. Reverse-image search and frame analysis can provide leads, but no single tool establishes the whole context. An AI-detection result alone should not be treated as a definitive authenticity verdict either.
What if the evidence is missing?
If the example report cannot be found, or the percentage describes a different group, do not publish it as a finding about the whole campus. Recording the source, relevant page, scope and editorial decision makes later review easier. Sensitive documents need to be handled through the newsroom's approved secure process.
If an error is discovered after publication, a clear correction helps readers understand what changed and when. The important question is not how confident the tool sounded, but what evidence supports the information the audience receives.
