The fear surge is mostly a coverage surge
Ten years of AI coverage measured directly. Volume rose 6.2x, tone fell 64 percent and never went negative, and fear coverage outran works coverage in exactly one year.
The short answer
What does the research show about The fear surge is mostly a coverage surge?
Evidence
- Cost framing appeared in 87.5 percent of articles quoting activists against 30.3 percent of articles quoting company-affiliated sources, and articles quoting only invested parties produced 71.2 percent benefit framing. Source selection swings framing by 57 points.
- Trust is a 14x lever on adoption. Among people who distrust AI, 3 percent embrace growing use. Among people who trust it, 43 percent do.
- Global sentiment posted its biggest single-year drop in the series: benefits outweigh drawbacks fell from 59 percent to 49 percent, n = 23,532 across 32 countries, fielded 20 March to 3 April 2026. The country set expanded from 30 to 32, so part of the move may be compositional.
Supporting pages
- Dictionary plain-language definitions
No published study gives a clean fear-to-works ratio in AI coverage, so we measured it. Three queries against the GDELT 2.0 DOC API, English-language sources, daily resolution, 3,504 daily observations from 1 January 2017 to 30 August 2026, averaged by calendar year.
Three readings. Volume is the story: AI coverage intensity rose 6.2x from 2017 and 4.5x from 2022. Tone fell hard and is still positive: average tone dropped from 1.161 in 2022 to 0.415 in 2026, a 64 percent decline, and every year in the series remains above zero. And fear coverage does not outgrow works coverage: the ratio sat below 1.0 in nine of ten years, crossing parity only in 2023 at 1.13, with the fear share of AI coverage staying inside an 11.4 to 16.0 percent band across the decade.
Caveats, stated plainly. Keyword baskets are a crude proxy for framing, the two baskets are not size-matched, GDELT's source pool drifts, GDELT tone is a lexicon score rather than human coding, and 2026 is a partial year. This is a defensible directional measurement, not a validated content analysis, and it is ours rather than a published result.
This is also where a starting assumption broke and is published as broken. We went looking for proof that fear content beats working content. The two largest engagement studies ever run on the question found no negativity advantage, and one found a penalty.
What this page is
A direct measurement of ten years of AI news coverage across 3,504 daily observations, testing whether fear framing outgrew working-deployment framing.
The argument
The perceived fear surge is mostly a volume surge. Coverage grew 6.2 times, tone fell 64 percent while staying positive, and fear coverage outran works coverage in exactly one year.
How to read it
- This is our own measurement, not a published result. Keyword baskets are a crude proxy for framing and the two baskets are not size-matched.
- GDELT tone is a lexicon score rather than human coding, and GDELT's source pool drifts over time.
- 2026 is a partial year. Treat the final point as directional.
AI coverage grew 6.2x. Its tone fell 64 percent and never went negative.
Two panels rather than two y-axes. Putting volume and tone on one plot with two scales would let the crossover sit wherever the designer wanted it.
AI coverage volume intensity · GDELT volume index
Average tone · GDELT tone, above zero is positive
2023 is the ChatGPT year: tone drops from 1.161 to 0.639. 2026 at 0.415 is the most negative year in the series and is still above neutral. The 2026 segment is dashed because the year is partial.
What this does not say
Volume and tone measure coverage, not outcomes. Nothing here says whether AI worked.
- Publisher
- GDELT 2.0 DOC API, queried and averaged by The Revenue AI Report
- Sample and method
- English-language sources, 3,504 daily observations
- Field dates
- 1 January 2017 to 30 August 2026
GDELT tone is a lexicon score rather than human coding, and the source pool drifts across the decade.
Fear coverage outran works coverage in exactly one of ten years.
Ratio of the fear keyword basket to the works keyword basket, by calendar year. Parity is 1.00.
Fear basket divided by works basket
- 20170.72
- 20180.83
- 20190.85
- 20200.81
- 20210.7
- 20220.66
- 20231.13
The only year above parity. ChatGPT shock year.
- 20240.82
- 20250.85
- 2026 to 30 Aug0.86
What this does not say
A stable fear share does not mean fear is unfounded. It means coverage mix did not move.
- Publisher
- GDELT 2.0 DOC API, queried and averaged by The Revenue AI Report
- Sample and method
- Same series as the panel above. Fear share of AI coverage stayed inside an 11.4 to 16.0 percent band across the decade
- Field dates
- 1 January 2017 to 30 August 2026
Keyword baskets are a crude proxy for framing and are not size-matched. This is a directional measurement, not a validated content analysis.
Ratio of the fear keyword basket to the works keyword basket, by calendar year. Parity is 1.00.
| Fear basket divided by works basket | Value | Note |
|---|---|---|
| 2017 | 0.72 | |
| 2018 | 0.83 | |
| 2019 | 0.85 | |
| 2020 | 0.81 | |
| 2021 | 0.7 | |
| 2022 | 0.66 | |
| 2023 | 1.13 | The only year above parity. ChatGPT shock year. |
| 2024 | 0.82 | |
| 2025 | 0.85 | |
| 2026 to 30 Aug | 0.86 |
Source: GDELT 2.0 DOC API, queried and averaged by The Revenue AI Report. Same series as the panel above. Fear share of AI coverage stayed inside an 11.4 to 16.0 percent band across the decade Fielded 1 January 2017 to 30 August 2026. Confidence: Medium.
Caveat: Keyword baskets are a crude proxy for framing and are not size-matched. This is a directional measurement, not a validated content analysis.
What this does not say: A stable fear share does not mean fear is unfounded. It means coverage mix did not move.
Nobody who measured AI headlines found a large fear ratio.
Negative against positive share in the peer-reviewed content analyses. The largest ratio anyone measured is 1.10 to 1.
- New York Times AI coverage 1956 to 2018ratio 0.44 to 1+30.58 pts
24.4% → 54.98%
- Global South headlines 2010 to 2024, n = 2,553ratio 0.44 to 1+28.8 pts
22.7% → 51.5%
- Global North headlines 2010 to 2024, n = 4,276ratio 1.10 to 1-3.9000000000000057 pts
41.7% → 37.8%
What this does not say
A framing difference between regions is not a difference in outcomes. It is a difference in who writes the coverage.
- Publisher
- Information, Communication and Society 2026, and Garvey and Maskal 2019
- Sample and method
- Global North against Global South, Mann-Whitney W = 4,157,516, p < 0.001. The New York Times series is a separate historical analysis
- Field dates
- Not published by the source.
The regime change is real. The flip is recent and lines up with the 2023 tone drop. Anyone quoting a large fear multiple is not quoting this literature.
The two largest tests of negativity found no advantage. One found a penalty.
Effect of negative tone on engagement, ordered by sample size. The width of each estimate is part of the finding.
- Facebook reactions. Negative posts received approximately 15 percent fewer reactions and 13 percent fewer comments.
- News-platform A/B tests. Beta = 0.003, p = 0.286, with highly significant cross-platform heterogeneity. No significant average effect.
- Upworthy headlines 2013 to 2015. Plus 2.3 percent per negative word, against a 1.39 percent base click-through rate. The base rate is the point.
- Four-outlet share counts 2019 to 2021. Plus 30 to 150 percent, on the smallest sample in the set and the widest interval.
What this does not say
A negativity effect in coverage does not mean the underlying events were positive. It measures selection, not reality.
- Publisher
- Noh and Soroka, and arXiv 2507.19300, with two smaller studies
- Sample and method
- Effect direction standardised for comparison. Ordered by sample size, descending
- Field dates
- Not published by the source.
No AI-specific replication of these experiments exists.
The cohort using AI the most is souring on it the fastest.
Two different survey houses and two different instruments. This is a descriptive pairing, not a within-respondent finding.
- Any personal AI use
- AI does more harm than good
What this does not say
Rising use alongside rising harm concern is not a contradiction. People use tools they do not trust.
- Publisher
- NORC AmeriSpeak and Gallup
- Sample and method
- Use figures move 54 to 70 to 73 percent across April, August and November 2025. Harm figures move 27 percent in 2024 to 36 percent in 2025 to 47 percent in 2026
- Field dates
- Not published by the source.
- Source
- https://news.gallup.com/
Two houses, two instruments. The lines are plotted on a shared percent axis and should be read as two separate measurements of one population, not as a paired panel.
Zero of 21 documented AI-native outcome claims were independently measured.
Every company-reported AI outcome checked in this research, grouped by evidence class. The empty class is the chart.
Independently measured: 0 of 21
- Company self-reported38%
- Company-reported through a journalist29%
- Vendor co-published19%
- Scored on a self-built benchmark14%
What this does not say
Weak evidence for a claim is not evidence the claim is false. It means the claim is unverified.
- Publisher
- Companies assessed by The Revenue AI Report
- Sample and method
- 21 documented AI-native outcome claims reviewed. Shares shown as percent of the 21 claims
- Field dates
- August 2026
- Source
- No primary URL reachable at research time.
The class boundaries are ours. The count of independently measured claims is not: it is zero in every reading of the set.
The claims that get walked back are the ones about headcount.
Every documented reversal in the window was a claim about people. None retracted a capability or a revenue figure.
Labor claims, February 2024 to February 2026
KlarnaFeb 2024 to May 2025
AI assistant framed as doing the work of 700 agents, followed by rehiring of human support.
DuolingoApr 2025 to May 2025
AI-first contractor memo, walked back within weeks after public reaction.
SalesforceSep 2025
4,000 of 9,000 support roles cut with AI named as the reason on the record, later reframed.
JPMorgan2024 to Feb 2026
$600M AI value figure subsequently hedged.
What this does not say
Four walkbacks in 24 months is a count of disclosed reversals, not a rate. Undisclosed ones are not in the frame.
- Publisher
- Compiled by The Revenue AI Report from primary coverage
- Sample and method
- Four documented reversals in a 24-month window, each traced to the original announcement and the revision
- Field dates
- Not published by the source.
- Source
- No primary URL reachable at research time.
Reversal is defined here as a public revision of a stated outcome, not as a retraction in the formal sense.
Also in the record
Figures that sit alongside these charts.
- Cost framing appeared in 87.5 percent of articles quoting activists against 30.3 percent of articles quoting company-affiliated sources, and articles quoting only invested parties produced 71.2 percent benefit framing. Source selection swings framing by 57 points.
- Trust is a 14x lever on adoption. Among people who distrust AI, 3 percent embrace growing use. Among people who trust it, 43 percent do.
- Global sentiment posted its biggest single-year drop in the series: benefits outweigh drawbacks fell from 59 percent to 49 percent, n = 23,532 across 32 countries, fielded 20 March to 3 April 2026. The country set expanded from 30 to 32, so part of the move may be compositional.
- We are not losing the narrative because journalists are hostile. We are losing it because the works side has never submitted its numbers for audit.
The brief
What is going on here, and why it matters.
The charts above are the evidence. This is the read: what the data describes, the mechanism behind it, where the argument could be wrong, and what a revenue team does about it.
What we measured and what it says
Three queries against the GDELT 2.0 DOC API, English-language sources, daily resolution, 3,504 daily observations from 1 January 2017 to 30 August 2026, averaged by calendar year. Volume is the story: AI coverage intensity rose 6.2 times from 2017 and 4.5 times from 2022. Average tone fell from 1.161 in 2022 to 0.415 in 2026, a 64 percent decline, and every year in the series stayed above zero.
The ratio result is the one that contradicts the common read. Fear coverage sat below parity with works coverage in nine of ten years, crossing only in 2023 at 1.13, with the fear share of AI coverage staying inside an 11.4 to 16.0 percent band across the decade. The feeling that fear took over is produced by a 6.2 times larger denominator, not by a changed mix.
A starting assumption that broke
We went looking for proof that fear content beats working content on engagement. The two largest tests of negativity found no advantage, and one found a penalty. That is published here as a broken assumption rather than quietly dropped, because the alternative is the exact selective-reporting behavior this library exists to document.
For a revenue team, the practical version is that alarm is not a reliable distribution strategy. The engagement premium people assume exists is not visible in the largest available tests.
Why this matters to a GTM narrative
Buyers are reading more AI coverage in a slightly less enthusiastic tone, not a hostile one. A message built on the assumption that the market has turned against AI will land wrong. So will a message built on 2022-level enthusiasm.
The defensible posture sits where the tone data sits: positive, cooler, more evidence-seeking than three years ago. That is also what the trust theme measures from the inside, which is two independent instruments pointing the same way.
What to do with it
The move, by seat.
- CMO
- Match the tone of the market rather than the tone of the loudest thread. Coverage is cooler and still positive.
- Content leads
- Stop optimizing for alarm. The two largest negativity tests found no engagement advantage and one found a penalty.
- Comms
- Assume a better-informed and more skeptical buyer. Volume rose 6.2 times, so exposure is high even where enthusiasm is not.
Questions this page answers
What the data says, in plain language.
- What does the research show about The fear surge is mostly a coverage surge?
- Ten years of AI coverage measured directly. Volume rose 6.2x, tone fell 64 percent and never went negative, and fear coverage outran works coverage in exactly one year. No published study gives a clean fear-to-works ratio in AI coverage, so we measured it. Three queries against the GDELT 2.0 DOC API, English-language sources, daily resolution, 3,504 daily observations from 1 January 2017 to 30 August 2026, averaged by calendar year.
- What does the figure "AI coverage grew 6.2x. Its tone fell 64 percent and never went negative" show?
- Two panels rather than two y-axes. Putting volume and tone on one plot with two scales would let the crossover sit wherever the designer wanted it. Source: GDELT 2.0 DOC API, queried and averaged by The Revenue AI Report. English-language sources, 3,504 daily observations Fielded 1 January 2017 to 30 August 2026. Confidence: Medium. Caveat: GDELT tone is a lexicon score rather than human coding, and the source pool drifts across the decade.
- What does the figure "Fear coverage outran works coverage in exactly one of ten years" show?
- Ratio of the fear keyword basket to the works keyword basket, by calendar year. Parity is 1.00. Source: GDELT 2.0 DOC API, queried and averaged by The Revenue AI Report. Same series as the panel above. Fear share of AI coverage stayed inside an 11.4 to 16.0 percent band across the decade Fielded 1 January 2017 to 30 August 2026. Confidence: Medium. Caveat: Keyword baskets are a crude proxy for framing and are not size-matched. This is a directional measurement, not a validated content analysis.
- What does the figure "Nobody who measured AI headlines found a large fear ratio" show?
- Negative against positive share in the peer-reviewed content analyses. The largest ratio anyone measured is 1.10 to 1. Source: Information, Communication and Society 2026, and Garvey and Maskal 2019. Global North against Global South, Mann-Whitney W = 4,157,516, p < 0.001. The New York Times series is a separate historical analysis Confidence: High. Caveat: The regime change is real. The flip is recent and lines up with the 2023 tone drop. Anyone quoting a large fear multiple is not quoting this literature.
- What does the figure "The two largest tests of negativity found no advantage. One found a penalty" show?
- Effect of negative tone on engagement, ordered by sample size. The width of each estimate is part of the finding. Source: Noh and Soroka, and arXiv 2507.19300, with two smaller studies. Effect direction standardised for comparison. Ordered by sample size, descending Confidence: High. Caveat: No AI-specific replication of these experiments exists.
- What else sits alongside these figures?
- Cost framing appeared in 87.5 percent of articles quoting activists against 30.3 percent of articles quoting company-affiliated sources, and articles quoting only invested parties produced 71.2 percent benefit framing. Source selection swings framing by 57 points. Trust is a 14x lever on adoption. Among people who distrust AI, 3 percent embrace growing use. Among people who trust it, 43 percent do. Global sentiment posted its biggest single-year drop in the series: benefits outweigh drawbacks fell from 59 percent to 49 percent, n = 23,532 across 32 countries, fielded 20 March to 3 April 2026. The country set expanded from 30 to 32, so part of the move may be compositional. We are not losing the narrative because journalists are hostile. We are losing it because the works side has never submitted its numbers for audit.
- Where does this data come from?
- Every figure is reproduced from a named publisher: GDELT 2.0 DOC API, Vicsek et al. 2026, source selection and framing, arXiv 2507.19300, negativity and engagement on Facebook, Ipsos AI Monitor 2026, Edelman Trust Barometer. Sample, field date, and confidence are shown on each chart. Sources marked as vendor research are labelled on the page.
- What could not be confirmed?
- A clean published fear-to-works ratio for AI coverage. None exists, which is why this theme carries our own measurement with its method in the open.
Cite this page
Permanent URL and suggested citation.
https://www.therevenueaireport.com/research/fear-versus-works
Kvarfordt, Jonathan. "The fear surge is mostly a coverage surge." The Revenue AI Report, Research Library. https://www.therevenueaireport.com/research/fear-versus-works
Figures on this page are reproduced from the publishers listed below. Cite the original publisher for the underlying data, and this page for the compilation and framing.
Sources
Every publisher used on this page.
If a metric, model term, or method on this page is unfamiliar, every one of them is defined in The AI and Revenue Dictionary. Sample size, field date, and confidence tags are explained there too.
GDELT 2.0 DOC API
English-language sources, 3,504 daily observations, 1 January 2017 to 30 August 2026. Queried and averaged by The Revenue AI Report
https://gdeltproject.org/Vicsek et al. 2026, source selection and framing
Cost framing in 87.5 percent of activist-quoting articles against 30.3 percent of company-quoting articles
https://www.tandfonline.com/journals/rics20arXiv 2507.19300, negativity and engagement on Facebook
6,081,134 posts across 97 news organizations. Approximately 15 percent fewer reactions on negative posts
https://arxiv.org/abs/2507.19300Ipsos AI Monitor 2026
n = 23,532 across 32 countries, fielded 20 March to 3 April 2026
https://www.ipsos.com/en/ai-monitorEdelman Trust Barometer
37 percent use AI weekly at work while 17 percent embrace growing use. 3 percent among distrusters against 43 percent among trusters
https://www.edelman.com/trust/trust-barometer
Could not confirm
What we looked for and did not find.
Claims found during research and not charted
- A clean published fear-to-works ratio for AI coverage. None exists, which is why this theme carries our own measurement with its method in the open.
How to cite this research
Written by Jonathan Kvarfordt, Founder and Principal Analyst, The Revenue AI Report. Published under CC BY 4.0.
APA
Kvarfordt, J. (2026). The fear surge is mostly a coverage surge. The Revenue AI Report. Retrieved from https://www.therevenueaireport.com/research/fear-versus-works
MLA
Kvarfordt, Jonathan. "The fear surge is mostly a coverage surge." The Revenue AI Report, 31 Aug. 2026, www.therevenueaireport.com/research/fear-versus-works.
BibTeX
@misc{kvarfordt2026fearversusworks,
author = {Kvarfordt, Jonathan},
title = {The fear surge is mostly a coverage surge},
year = {2026},
publisher = {The Revenue AI Report},
url = {https://www.therevenueaireport.com/research/fear-versus-works}
}Next theme
Six of eight seats are losing definition, not headcountThe postings baseline that has to run before anything gets attributed to AI, then each GTM seat with its own series, its own break point, and its own confidence tag.
Subscribe