# Can AI Stand In for Human Survey-Takers?

> Source: <https://www.pewresearch.org/newsletter/methods-and-innovation/methods-innovation-newsletter-nov-2026/>
> Published: 2026-11-19 17:00:00+00:00

**Recommendation:** Before substituting AI respondents for human survey-takers, replicate existing human surveys and check subgroup-level error, since Pew's synthetic runs diverged most for Republicans and Black adults.
More in the playbook "Evals and testing for LLM apps": <https://wpnews.pro/playbooks/evals-testing>

## 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. In the coming weeks and months, further analysis will help us better understand why the election unfolded the way it did and hopefully reveal why polling understated the forces that brought Democrats back into power in Congress and in many states.*

Scott Keeter

Senior Survey Advisor

## [**A First Look at Polling Performance in 2026**](https://www.pewresearch.org/newsletter/methods-and-innovation/methods-innovation-newsletter-nov-2026/)

 **A First Look at Polling Performance in 2026**

The most common question pollsters hear around election time is “Who’s going to win?” A close second in 2026 was “Are the polls going to be wrong again?” With final vote tallies nearly complete, it’s clear that preelection polls [captured the direction of this year’s election](https://www.pewresearch.org/short-reads/2026/11/06/a-first-look-at-polling-performance-in-2026/) while generally understating the size of Democratic Party’s advantage, writes Senior Survey Advisor Scott Keeter.

## [**Can AI Stand In for Human Survey-Takers? Not Really**](https://www.pewresearch.org/newsletter/methods-and-innovation/methods-innovation-newsletter-nov-2026/)

 As artificial intelligence becomes more advanced, there is growing interest in using it to [stand in for real respondents in public opinion surveys](https://www.pewresearch.org/data-labs/2026/09/30/can-ai-stand-in-for-human-survey-takers-not-really/). To learn more about this method, we replicated three human surveys from our American Trends Panel using AI respondents. 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.

## [Weighting Polls on Party Affiliation and Past Vote](https://www.pewresearch.org/methods/2026/09/16/weighting-polls-on-party-affiliation-and-past-vote/)

 As of 2024, almost 70% of national election polls weight on political party affiliation or report vote choice in a previous election. Though this kind of weighting is controversial, evidence shows that it has [made polls in recent elections modestly more accurate](https://www.pewresearch.org/methods/2026/09/16/weighting-polls-on-party-affiliation-and-past-vote/). But it can sometimes lead to an overstatement of similarity to past elections, which might have contributed to polls underestimating the extent of Democratic gains in 2026.

## [No Easy Fix for Bogus Respondents in Online Opt-In Polls](https://www.pewresearch.org/methods/2026/08/27/no-easy-fix-for-bogus-respondents-in-online-opt-in-polls/)

 One of the most urgent problems in online opt-in polling is bogus respondents, or survey-takers who make no effort to answer questions truthfully and instead are just looking to finish them quickly and collect rewards. [Our recent study compares three methods](https://www.pewresearch.org/methods/2026/08/27/no-easy-fix-for-bogus-respondents-in-online-opt-in-polls/) for identifying and removing bogus respondents from survey samples but finds that a surefire solution remains elusive.

## [How Americans Are Engaged With News, Politics, Religion and Civic Life](https://www.pewresearch.org/journalism/2026/07/16/how-americans-are-engaged-with-news-politics-religion-and-civic-life/)

 A recent Pew-Knight Initiative report uses a multimethod approach to explore [how Americans show up in public life](https://www.pewresearch.org/journalism/2026/07/16/how-americans-are-engaged-with-news-politics-religion-and-civic-life/). The project draws from a nationally representative survey of 5,393 U.S. adults, which asked about their political activity, civic involvement, religious attendance and attention to news.

## [New on Decoded](https://www.pewresearch.org/decoded/) 

 Pew Research Center’s [Decoded methodology blog](https://www.pewresearch.org/decoded/) focuses on the “how” behind our research. We regularly dig into interesting questions related to survey methods, data science, data visualization and more, so researchers can build on and engage with our work. Read some of our latest posts:

## [WATCH: Methods 101](https://www.youtube.com/playlist?list=PLZ9z-Af5ISavJpPlvdMU4T-etIDOUmldk)

 Check out our [Methods 101 video series](https://www.youtube.com/playlist?list=PLZ9z-Af5ISavJpPlvdMU4T-etIDOUmldk), an approachable introduction to the basics of survey research that answers common questions about polls and explores how pollsters are tackling challenges in the industry.

## [Featured datasets](https://www.youtube.com/playlist?list=PLZ9z-Af5ISavJpPlvdMU4T-etIDOUmldk)

 Pew Research Center makes the data used for much of our research available to the public for secondary analysis after a period of time. All available datasets can be [downloaded from our website](https://www.pewresearch.org/datasets/). Contact [info@pewresearch.org](mailto:info@pewresearch.org) with any questions. Featured selections from our archive include:

**2026 National Public Opinion Reference Survey (NPORS)**

Survey of 5,511 U.S. adults conducted Jan. 30- to June. 18, 2026, via web, paper and phone.

**Spring 2025 Global Survey Data**

Nationally representative surveys of 25 countries conducted Jan. 8-April 26, 2025.

**Global Restrictions on Religion 2007-2022**

Data on government restrictions and social hostilities involving religion in 198 countries and territories from 2007 to 2022.
