Case study · October 2026
In September 2026 a tech industry group’s poll told Congress that voters oppose new limits on big online platforms. Thinkwell read what people said about the same laws in public. Here is what each found, and why they differ.
Congress is weighing laws that would put new limits on large online platforms and marketplaces, including how they can offer their own products and services. Supporters say the limits would increase competition. Opponents say they could raise prices and take away features people use every day.
In September 2026 a tech industry group paid a national polling firm to survey registered voters online: 4,103 people about how they use online services, then 1,999 people about the proposed laws. Each survey has a margin of error of 2 points.
What it found
Thinkwell read public social media threads and news coverage about the same laws, from August to October 2026. Nobody was asked anything. Thinkwell found the people who took a side in their own words, matched what they said to the poll’s 22 questions about the laws, and checked those readings blind. 243 people took a side, across 140 threads. The read took 45 minutes.
What it found
| The poll | Thinkwell | |
|---|---|---|
| Who | 1,999 registered voters, weighted to match the country | 243 people who took a side in public threads |
| Stands for | Registered voters | People who speak up in public, not a voter sample |
| How they answered | Chose from answers to 22 questions the sponsor’s poll wrote | Said it in their own words, unprompted |
| When | September 10, 2026 | August to October 2026 |
| On the side of the restrictions | 27% | 57% |
The two do not measure the same group, so neither number is the truth on its own. The six findings below show why they differ, and how far to trust each.
27% vs 57%
Among poll respondents who picked a side, 27% backed the restrictions. Among the 243 people in the open conversation who took a side, 57% did, most likely between 50% and 65%.
19 of 22
19 of the poll’s 22 questions were worded toward the sponsor’s side. Where the wording leaned, the conversation differed by 26 points; where it was neutral, by about 7 points the other way.
Two blind coders agreed on all 22.
BuiltSurvey wording audit
22 cells, one for each question in the poll: 19 worded toward the sponsor’s side and 3 neutral.
Rest on a question, or use the arrow keys, to read it.
95%
A blind second reading agreed with Thinkwell’s judge on 95% of checked pairs, and an independent AI model agreed on 92%.
Agree and disagree were never swapped.
BuiltThe judge used in this study; stance reading in Thinkwell Listen is Live
7.8%
243 people took a side across 140 threads. The busiest thread held 7.8% of them, and the five busiest 22% together.
Up to 35%
Thinkwell’s personas predict how people would answer, and every wave of verified people checks them again. Good personas make each verified person count for more, so answers get cheaper. Drifting personas widen the range instead of making the number wrong.
The noise drops without adding bias.
BuiltNo live panel yet; the first verified people are our own team
Built
69% vs 22%
Top-level comments posted in a thread’s first 15 minutes reached its top quarter 69% of the time, against 22% for those posted 3 to 12 hours in.
BuiltTiming and response windows
Hours, not weeks
About 45 minutes from setup to results, every number checkable.
Bring a poll, a claim or a debate you are working on, and see what Thinkwell measures today.
Findings from one study of one poll, October 2026. Every number comes from the study’s results file. Not a guarantee.