Certified into the record
The exact text we saved
This is the version exactly as it was saved, character for character. The SHA-256 hash below is a 64-character code worked out from that text. Run the same text through SHA-256 and you get the same code; change a single letter and the code comes out completely different. Each version also stores the previous version’s hash, which chains them together so no version can be swapped out quietly.
TRUVACE RECORD VERSION record: TRV-2026-3055 version: 1 kind: certified reason: Certified into the record timestamp: 2026-11-19T18:02:07.751778Z status: published lens: p_space sector: science headline: Can AI Stand In for Human Survey-Takers? dek: Methods & Innovation A quarterly digest of the Center’s latest methodological research and data science discoveries · Subscribe ↗ Preelection polling this year accurately depicted the public’s dissatisfaction with national conditions and where Americans placed the blame. It revealed that Democrats were highly motivated to vote, as well as dissatisfaction among parts of Trump’s 2024 electoral coalition. […] gain_title: (none) problem_title: When Pew Research Center replicated three American Trends Panel surveys with AI respondents, the synthetic surveys reproduced the real results poorly and were especially error-prone for Republicans and Black adults. trace_subject: (none) gain_reading: (none) gain_evidence: (none) problem_reading: When Pew Research Center replicated three American Trends Panel surveys with AI respondents, the synthetic surveys reproduced the real results poorly and were especially error-prone for Republicans and Black adults. problem_evidence: These synthetic surveys did a poor job of reproducing the real results and were especially error-prone for some subgroups, including Republicans and Black adults. quick_read: Pew Research Center tested whether AI can stand in for real respondents in public opinion surveys. It replicated three human surveys from its American Trends Panel using AI respondents, and the synthetic surveys did a poor job of reproducing the real results. The errors were not spread evenly. The synthetic surveys were especially error-prone for some subgroups, including Republicans and Black adults, which matters for any pollster considering AI respondents as a cheaper substitute for people. Pew describes the test as a response to growing interest in the method as AI becomes more advanced. The finding appears in the center's quarterly Methods & Innovation digest, alongside work on election polling accuracy and bogus respondents in online opt-in polls. limitation: tag: Evidence-backed problem key_points: Pew replicated three human surveys from its American Trends Panel using AI respondents instead of people. | The synthetic surveys did a poor job of reproducing the real results. | Errors were larger for some subgroups, including Republicans and Black adults. | Pew said interest in AI stand-ins for survey respondents is growing as AI becomes more advanced. | The same digest reports that preelection polls in 2026 generally understated the size of the Democratic Party's advantage. | A separate Pew study of three ways to remove bogus respondents from online opt-in samples found that a surefire solution remains elusive. rundown: The AI-respondent test is one item in Pew's quarterly Methods & Innovation digest. Pew built synthetic versions of three American Trends Panel surveys and compared them with the answers real panelists gave. Pew reports that the synthetic surveys reproduced the real results poorly, with higher error for subgroups including Republicans and Black adults. Pew presents this as an assessment of using AI to stand in for public opinion respondents. The digest also covers other threats to poll quality. Preelection polls this year captured the direction of the election but generally understated the Democratic advantage, and almost 70% of national election polls as of 2024 weighted on party affiliation or past vote choice. sources: - journalism | Pew Research Center | https://www.pewresearch.org/newsletter/methods-and-innovation/methods-innovation-newsletter-nov-2026/ | 2026-11-19 prev: 0000000000000000000000000000000000000000000000000000000000000000
- SHA-256 hash
- 4d66b225e5c845d92321e6a4635c44ef06d0f7ae9c67c2b81ee9fd682eea7c42
- Previous version’s hash
- 0000000000000000000000000000000000000000000000000000000000000000
How to check it yourself, without trusting this page
Download any version from /api/record/TRV-2026-3055 and hash its canonical field yourself, for example with shasum -a 256. You should get its content_hash. Each version’s text also ends with prev:and the hash of the version before it (version 1 points to 64 zeros). If anyone had changed a single character of any version after it was saved, those hashes wouldn’t line up.
ace