Verasight

Trump's expansionist use of executive powers is dramatically unpopular with voters. 55% of American adults say he is "acting like he is above the law, consolidating power like a king."

More in our October Strength In Numbers/ @verasight.bsky.social poll:
www.gelliottmorris.com/p/new-poll-v...

Final pre-election Strength In Numbers / @verasight.bsky.social poll here!

- Dems lead on handling the economy
- Trump issue approval low, stable
- Voters oppose Trump's use of executive powers
- Dems up 7 in generic ballot
- Voters want Ds to win in NJ, VA

www.gelliottmorris.com/p/new-poll-v...

The findings from The Voter Poll by SSRS are shaping LIVE election coverage across major news networks including ABC News, CNN, NBC News, MSNBC, and FOX News.

Read more about our involvement here: www.verasight.io/post/verasig...

#thevoterpoll #electionresearch #surveydata #election2025

So excited that @verasight.io is part of this important polling effort! A wonderful validation of the incredible Verasight team!!

Verasight is a primary data contributor to The Voter Poll by SSRS For more than two decades, the leading media networks in the US have pooled their resources to commission a single voter poll to guide their insights and commentary on election results. For the 2025 e...

We are proud to announce that we will be a data contributor to The Voter Poll by SSRS on behalf of ABC News, @apnews.com, CBS, @cnn.com, Fox News Media, & @nbcnews.com. We are excited to support these decision desks & provide Americans with real-time election data.

www.verasight.io/post/verasig...

While many physicians see potential in reducing administrative burden and improving diagnosis, others cite obstacles like lack of training, poor data quality, limited funding, and mistrust in AI’s reliability.

http://www.verasight.io/reports/physicians-and-ai?utm_source=bluesky&utm_medium=organic

Verasight is heading to APHA2025 in DC, hosted by @apha.org. Come by Booth #1019 to ask a free question to 1,000 Americans and see how we power public health research with verified surveys. See you there.

Nearly half of U.S. physicians now use AI tools, mostly to cut down on paperwork and support clinical decisions. Our latest survey of U.S. physicians finds that AI adoption in healthcare is no longer experimental.

http://www.verasight.io/reports/physicians-and-ai?utm_source=x&utm_medium=organic

#Meettheteam Monday: Kayla just finished her PhD and is already diving into our sampling and data analysis. She loves collaborating with teammates and clients who care deeply about data quality and making sure respondents have their voices heard. Welcome, Kayla. We’re thrilled you’re here.

We asked 3,000 Americans questions submitted by APSA 2025 attendees:
• Should high schools teach more about misinformation? 88% said yes
• What drives your vote? 44% said economic interests matter most

See what political researchers are asking: https://reports.verasight.io/reports/apsa-2025

Congratulations to Makenzie J. Krocak, Joe Ripberger, Rachel Carr, Burrell Montz, Kathryn Semmens, Keri Maxfield, Lauren Harvey, Abby Bitterman, and Zoey Rosen on their new publication “Introducing the Flooding and Society Survey Dataset”. We are proud these researchers chose to work with Verasight.

93% of U.S. physicians say insurance companies have too much influence over patient care.

Verasight’s new national survey of 839 practicing physicians captures how doctors view the forces shaping American healthcare, and where AI fits into the picture.

www.verasight.io/reports/phys...

Proud to support the new Biological Conservation paper, “Mitigating collision-caused bird mortality through message framing,” by Shelby C. Carlson and Tina B. Phillips, which reports results from a survey they conducted with Verasight of 2,000+ adults.

Synthetic data is gaining traction in market research, but it’s still far from a reliable stand-in for real consumer behavior. Grateful to @kwcollins.bsky.social for highlighting our latest findings; part of an ongoing series investigating how synthetic samples stack up.

Such an important and timely topic. Thank you for bringing it to the table. We’re glad we could field your question!

I have new survey data! Democrats are more likely to disagree that masculinity is under threat than Republicans in the US—a bigger divide than gender! Super interested what this means in the context of online environments and politics. Thanks to @verasight.bsky.social for fielding the question!

Thanks for sharing this, James. It’s really interesting to see how views on masculinity differ so strongly by political affiliation. Grateful we could help field the question. Curious to see where this conversation goes.

In our latest synthetic sampling study, @gelliottmorris.com & Ben Leff add behavioral attributes to LLM personas, creating “digital twins”. This improved errors (mean error on Starbucks prices fell from ~20 to ~9 points), but synthetic data still missed big, with zero “about the same” responses.

#MeettheTeam Monday: Mike joined Verasight in September as a Client Research Associate, where he ensures surveys run seamlessly from start to finish. He thrives on solving complex challenges and making sure every detail is in place so clients get high-quality results they can trust.

In our synthetic sampling report, we asked respondents and their AI personas to pick among six hypothetical fall coffee drinks. The LLM sample made Autumn Ember the runaway winner while consumers were evenly split.

This kind of distortion risks misleading product teams and misallocating investment.

Excited to share the new Neuroethics article “Revolutionizing Brain Research Using Portable MRI in Field Settings” that includes a nationally representative Verasight survey of 2,001. Congrats to the authors on this important contribution.
https://link.springer.com/article/10.1007/s12152-025-09606-4

LLMs Misread Real Consumer Behavior | Verasight Researchers relying on so-called “synthetic samples” risk misestimating brand awareness and market demand, for both existing and, notably, potential products‍. By G. Elliott Morris and Ben Leff

www.verasight.io/reports/coff...

In our latest synthetic data analysis, @gelliottmorris.com and Ben Leff tests the ability of LLMs to simulate consumer behavior.

While synthetic samples might offer cheaper, faster market research, they dramatically misestimate consumer demand and brand awareness.

Link in the thread.

Thanks @kwcollins.bsky.social for sharing. When it comes to understanding consumers, real survey data still wins. LLM-generated “synthetic samples” missed the mark on everything from coffee consumption to brand recall. More on our new synthetic sampling analysis for market research coming soon.

Appreciate the shoutout, Chris. We’re glad the findings resonated. Synthetic data’s limitations are clearer the more we test them. Great to see others advancing the discussion!

The Risks of Using LLM Imputation of Survey Data to Produce `Synthetic Samples’ | Verasight The addition of administrative data and attitudinal markers does not always improve, and can decrease, the performance o

Subgroup errors of 10+ points can flip conclusions about electoral coalitions, policy support, or partisan divides. Our study with @gelliottmorris.com, Ben Leff and @peterenns.bsky.social shows that synthetic samples risk misleading researchers on exactly the questions where accuracy matters most.

The Risks of Using LLM Imputation of Survey Data to Produce `Synthetic Samples’ | Verasight The addition of administrative data and attitudinal markers does not always improve, and can decrease, the performance o

In our latest synthetic sampling analysis, neither additional information nor more recent LLMs improve the accuracy of the LLM results. For Trump approval, all models with additional info performed worse than the original model. GPT-5 + the most information was the least accurate.

The Risks of Using LLM Imputation of Survey Data to Produce `Synthetic Samples’ | Verasight The addition of administrative data and attitudinal markers does not always improve, and can decrease, the performance of LLMs. By G. Elliott Morris, Benjamin Leff, and Peter K. Enns

www.verasight.io/reports/synt...

In Verasight’s second synthetic data paper, @gelliottmorris.com, Ben Leff and @peterenns.bsky.social find that the performance of synthetic samples does not consistently improve (and can perform worse) with additional administrative data or real survey responses. Link in the thread.

#Meettheteam: @peterenns.bsky.social is our Co-founder & Chief Data Scientist. As a Prof. at Cornell & former Exec. Dir. of the Roper Center, Peter has spent his career advancing how we understand public opinion. At Verasight, he’s helping make validated, representative survey data more accessible.