Pulse Report 1
Week of September 7, 2026. Pulse is a weekly public-sentiment composite. It pulls a fixed set of free sources across three layers, attention (Google Trends, GDELT), opinion (the University of Michigan sentiment index, Civiqs, Gallup, Harris), and expectation (Polymarket, Kalshi), normalizes each series against its own trailing 52 weeks, and scores six topics on a ±100 scale. This is the first published edition.
Mean of the three topics with a scoreable opinion series. Both deltas compare to the July 6 reading, the last prior run.
Topic scoreboard
Each score is a weighted blend of three layers, opinion .4, expectation .4, attention .2, with every metric z-scored against its own trailing 52 weeks and mapped to ±100. A topic with no opinion series that has history is marked thin, gets no score, and stays out of the overall. Search volume plus market prices alone is not public mood. A positive score does not mean people feel good. It means the signals sit above their own past-year average, and the past year was itself a weak one for sentiment.
Divergence callouts
Attention and opinion point in opposite directions. Attention z = +1.1, opinion z = −2.0.
Expectation and opinion point in opposite directions. Expectation z = +1.2, opinion z = −2.0.
Synthesis
The overall Pulse sits at +11, in steady territory, after a one-week move of +23.7 that is identical to its four-week move, so the whole change is concentrated in this single week. AI sentiment is the strongest topic at +30 and the only one in the improving regime, lifted by a 72.2-point weekly gain, and the Civiqs good-or-bad tracker moved from more negative toward −13.2 with a +2.9 change even as Google Trends interest in AI risk, AI regulation, and AI taking jobs held flat at low levels. Economic mood is +12 and steady, though the opinion layer underneath it is soft: Civiqs' read on the economy now fell 2.2 to −32.4 and the Michigan index sat unchanged at 51.7, while prediction markets keep recession odds low on Kalshi at 0.012 and 0.076 and on Polymarket at 0.055, and search interest in layoffs dropped 10 points to 11. Presidential approval is the one negative composite at −9, and it carries both divergence callouts this week: opinion is the outlier at z=−2.0, with Civiqs approval down 1.7 to −24.4, while attention sits at z=+1.1 and expectation at z=+1.2, meaning the public's stated view is worse than either what it is searching for or what markets are pricing. Search interest in impeachment fell 6 points to 15 during that same stretch. Midterms, immigration, and institutional trust remain calibrating and thin, with no opinion series long enough to score, though the raw readings there are notable on their own, including Gallup's 89 percent saying corruption is widespread in government and 55 percent expressing at least a fair amount of confidence in it. The coherent story is a week where the composite rose on AI and a flat-to-firm economy layer while the one topic with real opinion history, presidential approval, saw stated sentiment sink below what attention and markets imply.
Evidence by topic
Economic mood
| Layer | Source | Metric | Value | 1 wk | 4 wk avg |
|---|---|---|---|---|---|
| attention | GDELT | avg_tone | −0.96 | – | – |
| attention | GDELT | news_volume_pct | 0.958 | – | – |
| attention | Trends | “inflation” | 33 | −1 | −3.8 |
| attention | Trends | “layoffs” | 11 | −10 | −8.2 |
| attention | Trends | “recession” | 7 | −1 | −1 |
| expectation | Kalshi | KXNBERRECESSQ (next recession start) | 0.012 | −0 | −0 |
| expectation | Kalshi | KXRECSSNBER (recession) | 0.076 | +0 | −0 |
| expectation | Polymarket | tag: fed | 0.141 | −0 | +0 |
| expectation | Polymarket | tag: inflation | 0.357 | −0 | +0 |
| expectation | Polymarket | tag: recession | 0.055 | −0 | −0.1 |
| opinion | Civiqs | economy_us_now (net) | −32.4 | −2.2 | −1.3 |
| opinion | UMich | Index of Consumer Sentiment, August final | 51.7 | +0 | −1.8 |
| opinion | HarrisLLM | share of Latino voters in battleground congressional districts | 73 | – | – |
Why economic mood scores +12 while the Michigan index is down year over year. The score compares each signal to its own trailing 52 weeks, not to a level. Stated opinion is below its past-year average: Michigan at 51.7 sits at z = −0.33 against a 12-month mean of 52.8, Civiqs at z = −1.13, so the opinion layer is −0.73. Searches for inflation, layoffs, and recession are well below their past-year averages (attention +1.54), and Kalshi recession odds are near their lows while Polymarket's fed and inflation tags pull the other way (expectation +0.42). With weights .4 opinion, .4 expectation, .2 attention, the composite z is +0.18, which maps to +12. Read it as less anxious than the past year on searches and markets, worse on what people say, not as the economy feeling good. Year over year is a different comparison: August 2025 was 58.2, so the index is down about 11 percent from a year ago and roughly at the average of a bad year.
Presidential approval
| Layer | Source | Metric | Value | 1 wk | 4 wk avg |
|---|---|---|---|---|---|
| attention | Trends | “impeachment” | 15 | −6 | −3 |
| attention | Trends | “president approval rating” | 2 | +0 | −0.5 |
| expectation | Polymarket | tag: trump | 0.075 | +0 | +0 |
| expectation | Polymarket | tag: white-house | 0.18 | −0 | −0.1 |
| opinion | Civiqs | approve_president_trump_2025 (net) | −24.4 | −1.7 | −2.2 |
AI sentiment
| Layer | Source | Metric | Value | 1 wk | 4 wk avg |
|---|---|---|---|---|---|
| attention | Trends | “AI regulation” | 3 | +0 | −1.2 |
| attention | Trends | “AI risk” | 8 | +0 | −2.2 |
| attention | Trends | “AI taking jobs” | 1 | +0 | +0 |
| expectation | Polymarket | tag: ai | 0.142 | +0 | +0 |
| opinion | Civiqs | ai_good_or_bad (net) | −13.2 | +2.9 | +2.9 |
| opinion | HarrisLLM | workers fear becoming obsolete | 58 | – | – |
| opinion | HarrisLLM | American workers feel underprepared for the next generation of AI | 54 | – | – |
| opinion | HarrisLLM | would be uncomfortable if AI guessed a personal care need | 79 | – | – |
| opinion | HarrisLLM | comfortable with AI helping them shop as long as they still decide | 72 | – | – |
2026 midterms (thin)
| Layer | Source | Metric | Value | 1 wk | 4 wk avg |
|---|---|---|---|---|---|
| attention | Trends | “midterm elections” | 53 | +0 | +10.8 |
| attention | Trends | “register to vote” | 38 | +0 | +3 |
| expectation | Polymarket | tag: house-of-representatives | 0.081 | +0 | −0 |
| expectation | Polymarket | tag: midterms | 0.411 | −0 | −0 |
| opinion | HarrisLLM | share of Latino voters in battleground congressional districts | 52 | – | – |
Immigration (thin)
| Layer | Source | Metric | Value | 1 wk | 4 wk avg |
|---|---|---|---|---|---|
| attention | Trends | “ICE raids” | 10 | +0 | +4.2 |
| attention | Trends | “border crisis” | 1 | +0 | +0 |
| attention | Trends | “deportation” | 26 | −3 | −29.2 |
| expectation | Polymarket | tag: immigration | 0.04 | +0 | +0 |
Institutional trust (thin)
| Layer | Source | Metric | Value | 1 wk | 4 wk avg |
|---|---|---|---|---|---|
| attention | Trends | “corruption” | 32 | +0 | −2 |
| attention | Trends | “deep state” | 7 | −1 | −0.8 |
| attention | Trends | “rigged” | 13 | +0 | +1.8 |
| opinion | GallupLLM | great deal or fair amount of confidence in the U.S. government | 55 | – | – |
| opinion | GallupLLM | corruption is widespread throughout the government | 89 | – | – |
| opinion | GallupLLM | corruption is widespread within businesses in the U.S. | 71 | – | – |
Rows marked LLM were extracted from Gallup and Harris release pages by Claude with a strict schema and have no history yet. Dashes mean the metric has no prior reading to compare against. The AI sentiment table shows four of sixteen Harris rows from one release on AI in personal-care shopping.
Sources this run: Google Trends, GDELT (partial, rate-limited), University of Michigan Surveys of Consumers, Civiqs, Gallup, Harris, Pew, Polymarket, Kalshi. The Michigan index comes from the survey's own monthly table rather than FRED, which carries it one month delayed. Polymarket and Kalshi carry 52 weeks of backfilled history. Google Trends history was re-pulled so it shares one normalization window with this week. The three thin topics need an opinion series with history before they score.
The composite is a directional instrument, not a measurement. When a divergence callout and the composite disagree, the callout wins the headline.
Scott Ewalt is the founder of Cardinal Element. Pulse is Cardinal Element's own research instrument.