Washington Post AI-generated podcasts contain errors and fabrications
The Washington Post launched an AI-generated podcast feature that within 48 hours was found to contain pronunciation errors, invented quotes, misattributions, and added commentary misrepresenting sources. Staff internally called for the tool to be pulled, with the head of standards acknowledging the errors and editors criticizing the launch as damaging to the publication's journalism.
What happened
The Washington Post launched an AI-generated podcast feature within its app, described as a 'personalized, AI podcast experience' that allowed users to customize topics, lengths, and hosts. Within 48 hours of launch, internal staff reported the podcasts contained pronunciation mistakes, misattributed or invented quotes, and added commentary that presented sources' remarks as the newspaper's own position.
Staff raised concerns in internal Slack messages about the errors and their implications. The head of standards, Karen Pensiero, acknowledged the errors had been 'frustrating for all of us.' One editor stated it was 'truly astonishing that this was allowed to go forward at all' and said the Post was 'deliberately warp[ing] its own journalism,' recommending the tool be pulled immediately.
Reported impact
The launch sparked internal backlash and embarrassment within the newsroom. Staff expressed concern about putting words in people's mouths and potential litigation, particularly in coverage areas described as 'incredibly litigious and fraught.' One staffer called the rollout 'a total disaster.'
Index analysis
The AI system generated inaccurate content including pronunciation errors, false quotations, and misattributions. It also added interpretive commentary that blurred the line between reporting and editorial position, potentially misrepresenting sources' statements and the publication's journalism.
Governance controls
One staffer questioned the guardrails in place to ensure accuracy, asking 'What are the guardrails to ensure accuracy in this podcast?' Another questioned the product's governance, writing that while the tool was labeled 'product not journalism,' the Post was 'putting words in peoples mouths' and journalists faced the consequences.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.