JoshuaShepherd Research public opinion and polling data on: $ARGUMENTS
$ARGUMENTS can include:
- A topic (e.g.
public attitudes toward AI,trust in institutions,political polarization) - A geographic focus (e.g.
--region US,--region EU,--region global) - A time range (e.g.
--since 2020,--trend 2015-2025) - A demographic focus (e.g.
--demo age,--demo partisan,--demo education) - An output path override (default:
_research/polling/) - A depth flag (e.g.
--depth quickfor headline numbers,--depth deepfor full trend analysis) - Empty — ask the user what topic they want opinion data on
Purpose
This skill performs live research across major polling and public opinion organizations to find the most current and authoritative data on what people think, believe, and feel about a given topic. It retrieves published survey findings, trend data, demographic breakdowns, and cross-national comparisons, then organizes everything into a structured, citation-rich report.
What it produces:
_research/polling/[topic-slug]/README.md— headline findings and executive summary_research/polling/[topic-slug]/data-tables.md— all extracted polling numbers in table format_research/polling/[topic-slug]/trend-analysis.md— how opinion has changed over time_research/polling/[topic-slug]/demographic-breakdowns.md— opinion by age, education, gender, party, etc._research/polling/[topic-slug]/methodology-notes.md— sample sizes, margins of error, methods_research/polling/[topic-slug]/source-index.md— all searches run, pages fetched, retrieval status
Source Organizations
1. Pew Research Center (Primary — US & Global)
What it is: Non-partisan fact tank. Conducts extensive public opinion polling on technology, politics, religion, media, demographics, and global affairs. Among the most respected and frequently cited survey organizations. Regularly publishes AI-specific opinion research.
Web Endpoints:
# Main search
https://www.pewresearch.org/search/{query}
# Topic pages (curated collections)
https://www.pewresearch.org/topic/science/science-issues/artificial-intelligence/
https://www.pewresearch.org/topic/politics-policy/
https://www.pewresearch.org/topic/internet-technology/
https://www.pewresearch.org/topic/religion/
# Data sets (downloadable)
https://www.pewresearch.org/datasets/
# Global Attitudes
https://www.pewresearch.org/global/
Search strategy:
site:pewresearch.org "{topic}" survey 2024 OR 2025
site:pewresearch.org "{topic}" public opinion
site:pewresearch.org "{topic}" Americans say
What to extract:
- Headline percentages (e.g., "52% of Americans say...")
- Trend data (same question asked over multiple years)
- Demographic breakdowns (age, education, party, race/ethnicity, gender)
- Methodology boxes (sample size, dates, margin of error)
- Full report URLs for citation
Usage notes:
- Reports are freely available online with detailed methodology
- Pew often publishes crosstabs and topline questionnaires as PDFs
- Their data is considered gold standard for non-partisan opinion research
- Check both the main report and any "detailed tables" or "appendix" links
2. Gallup (Primary — US & Global Tracking)
What it is: Long-running polling organization with tracking polls on confidence in institutions, economic outlook, social issues, and well-being. Valuable for longitudinal trends spanning decades.
Web Endpoints:
# Main search
https://news.gallup.com/search/default.aspx?q={query}
# Topic pages
https://news.gallup.com/topic/technology.aspx
https://news.gallup.com/topic/world-affairs.aspx
https://news.gallup.com/topic/politics.aspx
# Gallup Analytics (some data free, some gated)
https://www.gallup.com/analytics/
Search strategy:
site:gallup.com "{topic}" poll survey 2024 OR 2025
site:news.gallup.com "{topic}" Americans
"gallup poll" "{topic}" 2024 OR 2025
What to extract:
- Tracking poll trend lines (Gallup often asks the same question for 10-50+ years)
- "Confidence in institutions" data
- Economic confidence index numbers
- "Most important problem" data
- Demographic breakdowns where published
Usage notes:
- Published poll reports at news.gallup.com are free to read
- Some deeper analytics and datasets require subscription
- Gallup's value is in longitudinal trends — always look for historical comparison
- Note when data is from Gallup Panel (online) vs. traditional phone polling
3. Ipsos (Global Multi-Country Surveys)
What it is: Multinational research firm that publishes regular global surveys on technology perceptions, social issues, trust, and current events. Excellent for cross-country comparison data.
Web Endpoints:
# Main search/publications
https://www.ipsos.com/en/knowledge/search?search={query}
# Global Advisor surveys
https://www.ipsos.com/en/ipsos-global-advisor
# What Worries the World (monthly tracking)
https://www.ipsos.com/en/what-worries-world
Search strategy:
site:ipsos.com "{topic}" survey global 2024 OR 2025
"ipsos" "{topic}" public opinion poll
site:ipsos.com artificial intelligence OR AI survey
What to extract:
- Cross-country comparison tables (typically 25-30 countries)
- "What Worries the World" monthly tracking data
- Global averages vs. country-specific figures
- Trust/confidence metrics
- Infographics and data visualizations (note the data within them)
Usage notes:
- Published reports and infographics are free
- Ipsos Global Advisor surveys typically cover 25-30 countries with ~1,000 respondents each
- Reports often come with downloadable PDFs containing full data tables
- Particularly strong on AI perceptions — they run regular multi-country AI surveys
4. Eurobarometer (EU Public Opinion)
What it is: The European Commission's public opinion survey program. Conducted in all EU member states plus candidate countries. Covers technology, science, values, social issues, and EU policy.
Web Endpoints:
# Main portal
https://europa.eu/eurobarometer/screen/home
# Search surveys
https://europa.eu/eurobarometer/surveys/browse/all/theme/all
# Open data portal (full datasets)
https://data.europa.eu/data/datasets?query=eurobarometer
Search strategy:
site:europa.eu eurobarometer "{topic}"
"eurobarometer" "{topic}" 2024 OR 2025
"special eurobarometer" "{topic}"
"standard eurobarometer" "{topic}"
What to extract:
- Country-by-country breakdowns across all EU members
- Trend data (Eurobarometer has tracked some questions for 40+ years)
- Demographic breakdowns (age, education, occupation, urbanization)
- Special Eurobarometer reports on specific themes (AI, science, digital)
- Standard Eurobarometer tracking questions on trust, economy, issues
Usage notes:
- All reports and datasets are completely free and open
- Full datasets are downloadable from the EU Open Data Portal
- Two types: "Standard" (biannual tracking survey) and "Special" (topic-specific deep dives)
- The Special Eurobarometer on AI/digital is particularly rich
- Reports include methodology, sample sizes per country, and margin of error
- Available in all EU languages
5. YouGov (High-Frequency Digital Polling)
What it is: Online polling firm that publishes a high volume of topical polls across US, UK, and other markets. Fast-turnaround polling on current events and trending topics. Also runs brand tracking.
Web Endpoints:
# Main results/articles
https://today.yougov.com/search?q={query}
# Topics
https://today.yougov.com/topics/technology
https://today.yougov.com/topics/politics
# YouGov-Cambridge tracking
https://today.yougov.com/topics/yougov-cambridge
Search strategy:
site:today.yougov.com "{topic}" poll
site:yougov.co.uk "{topic}" survey
"yougov" "{topic}" 2024 OR 2025 poll results
What to extract:
- Topical poll results (often with demographic crosstabs)
- Favorability/approval ratings
- "Most important issue" tracking
- Brand/entity perception data
- Cross-country comparisons (US/UK primarily)
Usage notes:
- Published poll results are free to read
- YouGov polls are online panel-based (note this in methodology)
- Very high volume — useful for finding polls on niche topics
- YouGov-Cambridge Centre for Public Opinion Research produces more rigorous studies
- Data Explorer tool sometimes allows custom crosstabs
6. Morning Consult (Tech & Industry Tracking)
What it is: Data intelligence company with high-frequency polling. Strong coverage of technology perception, brand trust, and industry-specific sentiment. Publishes regular AI tracking data.
Search strategy:
site:morningconsult.com "{topic}" poll survey
"morning consult" "{topic}" 2024 OR 2025 survey
site:pro.morningconsult.com "{topic}"
What to extract:
- Technology adoption and perception metrics
- AI-specific tracking polls
- Brand trust and favorability data
- Demographic and partisan breakdowns
- Trend tracking over quarters
Usage notes:
- Some content is free, some requires Pro subscription
- Their AI tracking data is among the most current available
- Note when data is from free reports vs. gated Pro content
7. ANES & GSS (US Longitudinal — Deep Demographic Data)
What it is:
- ANES (American National Election Studies): Deep surveys on political attitudes, values, and participation. Conducted every election cycle since 1948.
- GSS (General Social Survey): Broad sociological survey on American attitudes and behaviors. Conducted since 1972 by NORC at University of Chicago.
Web Endpoints:
# GSS Data Explorer
https://gssdataexplorer.norc.org/
https://gss.norc.org/
# ANES
https://electionstudies.org/
https://electionstudies.org/data-center/
Search strategy:
site:gssdataexplorer.norc.org "{topic}"
site:electionstudies.org "{topic}"
"general social survey" "{topic}" trend
"ANES" "{topic}" attitudes
What to extract:
- Multi-decade trend lines on social attitudes
- Extremely detailed demographic breakdowns
- Political behavior and participation data (ANES)
- Social trust, religion, work, family, race attitudes (GSS)
Usage notes:
- Both datasets are completely free to access and download
- GSS Data Explorer allows running crosstabs online without downloading data
- These are the gold standard for understanding long-term American attitude change
- Academic-quality methodology — probability samples, detailed codebooks
- ANES cumulative file spans 1948-present; GSS spans 1972-present
8. Stanford HAI AI Index (AI-Specific Annual Report)
What it is: The Stanford Institute for Human-Centered AI publishes an annual AI Index Report that compiles AI-related polling data, public perception trends, policy attitudes, and industry sentiment from multiple sources.
Web Endpoint:
https://aiindex.stanford.edu/report/
Search strategy:
site:aiindex.stanford.edu public opinion OR perception OR survey
"AI Index Report" public opinion "{year}"
"Stanford HAI" AI public perception survey
What to extract:
- Aggregated public opinion data from multiple polling sources
- Cross-national AI perception comparisons
- AI policy preference data
- Workforce and industry sentiment on AI
- Media coverage analysis of AI topics
Usage notes:
- The full report is free to download as PDF
- Published annually (typically March-April)
- Compiles data from Pew, Ipsos, Gallup, Eurobarometer, and other sources — very useful as a meta-source
- Includes data visualizations that synthesize multiple polls
Research Process
Phase 1 — Scope & Strategy
Before searching, establish:
- Topic — What opinion/attitude are we measuring?
- Geography — US only? EU? Global? Specific countries?
- Time frame — Current snapshot? Trend over time?
- Demographic interest — Any particular breakdowns needed (age, party, education)?
- Search terms — Generate 3-5 query variations:
- The topic as commonly polled (e.g., "artificial intelligence" not just "AI")
- Related framings (e.g., "automation" OR "machine learning" alongside "AI")
- Polling-specific language ("survey" "poll" "public opinion" "Americans say")
State the strategy before executing.
Phase 2 — Multi-Source Search
Search across all relevant sources in priority order:
For US opinion:
- Pew Research Center (most authoritative, detailed methodology)
- Gallup (best for longitudinal tracking)
- YouGov / Morning Consult (most current, highest frequency)
- ANES / GSS (deepest demographic data, longer time horizons)
For global/comparative opinion:
- Pew Global Attitudes
- Ipsos Global Advisor
- Eurobarometer (EU-specific)
- YouGov (US/UK comparison)
For AI-specific opinion:
- Pew (regular AI surveys)
- Stanford HAI AI Index (aggregated data)
- Ipsos (global AI perception surveys)
- Morning Consult (AI tracking)
- Eurobarometer Special on Digital/AI
For each source:
- Run 2-3 search query variations
- Fetch the most relevant report pages
- Extract data tables, percentages, and trend data
- Record methodology details (sample size, dates, method)
- Log all searches in the source index
Phase 3 — Data Extraction
For each poll or survey found, extract into structured format:
| Field | Description |
|---|---|
source |
Organization name |
report_title |
Full title of the report or article |
publication_date |
When the report was published |
fieldwork_dates |
When the survey was conducted |
sample_size |
Number of respondents |
population |
Who was surveyed (US adults, global, registered voters, etc.) |
method |
Phone, online panel, in-person, mixed |
margin_of_error |
Overall and for subgroups if stated |
key_findings |
Top-line percentages and findings |
demographic_breaks |
Any breakdowns by age, party, education, gender, race, etc. |
trend_data |
Same question asked in prior waves, if available |
url |
Direct link to the report |
Phase 4 — Trend Analysis
When the same question has been asked across multiple time periods:
- Build a trend table showing the response over time
- Identify inflection points — when did opinion shift? What was happening?
- Note direction and magnitude — is opinion moving slowly or rapidly? In what direction?
- Partisan/demographic divergence — are different groups trending in different directions?
- Cross-national trends — are patterns consistent globally or divergent?
Phase 5 — Demographic Deep Dive
Where data permits, build demographic comparison tables:
| Demographic | Favorable | Unfavorable | Unsure |
|-------------|-----------|-------------|--------|
| **Age** | | | |
| 18-29 | X% | Y% | Z% |
| 30-49 | ... | ... | ... |
| 50-64 | ... | ... | ... |
| 65+ | ... | ... | ... |
| **Party** | | | |
| Democrat | ... | ... | ... |
| Republican | ... | ... | ... |
| Independent | ... | ... | ... |
| **Education** | | | |
| College+ | ... | ... | ... |
| Some college| ... | ... | ... |
| HS or less | ... | ... | ... |
Note which breakdowns come from which source — not all polls publish the same demographic cuts.
Phase 6 — Output Generation
Write all output files with clean markdown formatting.
README.md structure:
# Public Opinion Research: [Topic]
**Date:** [YYYY-MM-DD]
**Topic:** [topic description]
**Geographic scope:** [US / EU / Global / etc.]
**Sources consulted:** [N organizations, N reports]
**Date range of polls:** [earliest to most recent]
## Headline Findings
[Top 5-7 findings with specific numbers and sources]
## Key Trends
[2-3 sentence summary of how opinion is changing]
## Most Surprising Finding
[The data point that challenges common assumptions]
## Files in This Report
- [data-tables.md](data-tables.md) — All polling numbers in table format
- [trend-analysis.md](trend-analysis.md) — How opinion has changed over time
- [demographic-breakdowns.md](demographic-breakdowns.md) — Opinion by demographic groups
- [methodology-notes.md](methodology-notes.md) — Sample sizes, methods, margins of error
- [source-index.md](source-index.md) — All searches and sources
Key Rules
- Numbers must have sources — Every percentage, every data point must be attributed to a specific poll with date, source, and sample size. Never state a polling number without attribution.
- No fabricated poll data — Never invent survey results. If you cannot find polling data on a specific question, say so. "No data found" is better than a fabricated number.
- Methodology matters — Always note sample size, method (phone vs. online), and margin of error. A poll of 500 people has very different reliability than a poll of 5,000.
- Question wording matters — The exact wording of a poll question dramatically affects results. When possible, include the actual question text. Note when different polls ask differently-worded questions on the same topic.
- Margin of error applies to subgroups — A poll's overall MOE of ±3% may be ±6-8% for a demographic subgroup. Note this when presenting breakdowns.
- Online vs. probability samples — Probability samples (Pew, Gallup phone, ANES, GSS) are methodologically stronger than online opt-in panels (YouGov, Morning Consult). Note the difference.
- Source index is mandatory — Every search, fetch, and data retrieval must be logged in
source-index.md. - Recency hierarchy — When the same organization has published multiple waves, always lead with the most recent data and show the trend.