The crisis opens the door, oratory lets it in

This piece began with one sentence: human ego always brings the same destruction, by choosing a certain type of leader. The ego here is not a personality trait: it is surrender to one's own craving, and the turn toward the leader who buys it.

The title's question is read this way: what is sold is the promise; what is bought is the voter's craving. Five AI voices tested it. A red team then objected to their consensus. Part of the sentence stood; part fell. The end says plainly which, and why.

Two rules: no leader or party is named, only cases, countries and periods; and the word "populist" appears only in its measurable sense: discourse that frames politics as "pure people" against "corrupt elite".

1. Three promises, three reversed outcomes: is the gap real?

It does; but its three legs are not equally sturdy.

A promise of democracy, an outcome of repression. In Venezuela, from 1999, the liberal democracy index fell from 0.575 to 0.042 (V-Dem). A United Nations fact-finding mission documented at least 25 deaths and thousands of detentions.

In Peru, over 1990–2000, the same index fell from 0.425 to 0.095; that period's president was sentenced to 25 years in prison on 7 April 2009. In Hungary, after 2010, the index slipped from 0.766 to 0.315. Here there is no prison, only institutional erosion.

A promise of justice, an outcome of corruption. In Hungary, the V-Dem political corruption index rose from 0.275 to 0.488 (higher means more corruption). The corruption perceptions score fell from 55 to 41, then to 40 (Transparency International). In Peru the same index rose from 0.458 to 0.691. In the Philippines, the 1998–2001 president was sentenced to life imprisonment on 12 September 2007 and pardoned on 25 October 2007.

A promise of peace, an outcome of war. Here only one case remains: Germany 1933. A contemporary diplomatic report finds the rhetoric "moderate", yet records a suspicion that this may be tactical. The index then fell from 0.434 to 0.017. Of the three, this leg has the weakest evidence, and we mark it so.

2. Doesn't everyone break promises?

They do; but less than assumed. In a twelve-country study, governing parties at least partly fulfil 59% of pledges. For single-party majority governments in the United Kingdom it is 86% (Thomson et al. 2017, American Journal of Political Science).

So the difference should not be sought in frequency. Four of the five voices converged: the difference lies in the type of promise broken and the difficulty of exit. An ordinary government fails to keep a tax pledge. In the gap cases, what is broken is the rule itself.

On exit there is a think-tank count: 46 leaders, 33 countries, 1990–2018. Half of these leaders changed or rewrote the constitution. Average time in office is 6.5 years; for non-populists, 3 years. Only 17% left through electoral defeat (Kyle & Mounk 2018, think-tank report).

This count is not peer-reviewed and has no confidence interval. Peer-reviewed studies show the same direction more cautiously: over 1995–2018, populist government's effect on electoral, liberal and deliberative democracy measures is negative (Ruth-Lovell & Grahn 2023, European Journal of Political Research).

3. Who opens the door?

A crisis; but not every crisis. Over 800 elections in twenty advanced economies, 1870–2014, were examined. After a financial crisis, the far right's vote share rises by about 30% on average. An ordinary downturn shows no such movement (Funke, Schularick & Trebesch 2016, European Economic Review).

This finding gives the piece its title. Oratory does not enter an empty room. First the door opens: savings melt away, trust in institutions collapses, and the question "who brought us to this" is left ownerless. Then whoever gives that question the shortest answer walks in.

4. How does oratory let them in?

Oratory is often treated as the heart of the matter. Measurements say something narrower: oratory does not persuade, it gathers.

Pooling forty field experiments, the best estimate of campaign themes' persuasive effect in a general election is zero (Kalla & Broockman 2018, American Political Science Review). In a European country, 455 speeches by a single leader (1927–1933) were compared with five national elections. The effect was negligible; the one exception was the second round of 1932 (Selb & Munzert 2018, American Political Science Review).

What does work? Of the ten mechanisms Astra listed, three have a measured effect:

• Repetition and familiarity. Mere exposure increases liking (Zajonc 1968, Journal of Personality and Social Psychology).
• Moral-emotional language. In 563,312 messages, each extra moral-emotional word multiplied diffusion by 1.20 (Brady et al. 2017, PNAS). In a preregistered replication the factor was 1.15; in the pooled analysis, 1.12 (Brady et al. 2025, PNAS Nexus).
• Building an "us" and an enemy. It rests on group-identity experiments; this piece carries no figure for it.

The other seven (charisma, rhythm, the "deep story" and the like) remain theory; we label them so.

5. If the gap is visible, why does support continue?

The obvious answer is ignorance. It does not hold. Five experiments, over 10,100 participants, 52 topics: corrections do not backfire; people accept them (Wood & Porter 2019, Political Behavior). But the vote does not change. In a candidate-choice experiment, only a small share of voters put a democratic principle ahead of party interest, a share that shrinks as polarization grows (Graham & Svolik 2020, American Political Science Review).

Three voices concluded: the voter chooses a side, not a sentence. Two went a step further and called the gap a "tacit contract". Opus objected: the contract metaphor shifts the bill onto the voter, yet the cost comes from the item the contract does not state.

An open question remains. We found no study measuring whether voters notice that the discourse of coming to power was later not kept. "Buying" is therefore not a finding but a question.

6. How big is the bill?

There are three items.

Income. 51 populist leaders, 1900–2020, were examined. Fifteen years on, per-capita income is about 10% below a constructed comparison country's (Funke, Schularick & Trebesch 2023, American Economic Review). The authors say they took only clear cases; read the figure accordingly.

Institutions. Across 155 countries, 1960–2020, populist leadership is associated with a marked rise in executive corruption, but not other corruption types (Zhang 2024, Governance). In Hungary, Council of the European Union Decision 2022/2506 suspended 55% of three programmes' funding, about 6.3 billion euros.

Duration. Exit gets longer (section 2). But the door is not locked. Of autocratization episodes in 1900–2023, 52% reversed; in the last thirty years, 73%. Of those that reversed, 90% reached their former level or better (Nord et al. 2025, Democratization).

The other is harsher: since 1993, 59 autocratization episodes and 36 democratic breakdowns; once an episode began, only one in five democracies managed to prevent breakdown (Boese et al. 2021, Democratization). The two studies have different denominators; read them side by side, not summed.

7. Does history show a pattern, or a fate?

It does not show a fate. In the think-tank count, serious democratic backsliding is 23% under populist government and 6% under non-populist. Read it the other way: in more than three quarters of the cases, no serious backsliding was measured (the subtraction is ours). Remembering only the bad endings is survivorship bias. Once the denominator is added, "always" falls and "more often" remains.

"The same destruction" falls too. The cases split into three distinct types: war, repression, wealth. There is no single mould. What they share is not the outcome but the condition: power gathered in one person, the institutions that would stop it weakened.

Thesis cards

Ten theses from the matrix. Voices: Gemini Pro, Gemini Flash, Opus, Enlil, Enki.

The red team wrote an objection to each of the seven theses with no opponent; the boxes above come from there.

8. What can the reader do? Five tools

Look at the number, not the leader.

1. V-Dem liberal democracy index. Pick your country and year; look at the slope, not a single year.

2. Global Populism Database rhetoric score. It scores speeches; it gives the sitting executive's term average, not the promise made on the way to power.

3. Promise-keeping rate. The method of Thomson and colleagues: count the testable pledges, keep the abstract ones apart.

4. Count of jailed journalists. The Committee to Protect Journalists publishes it every year.

5. Corruption perceptions score trend. Transparency International; again, look at the slope.

And ask one question: if this leader loses, can they leave? Look for the answer not in personality but in the courts, the press and the electoral commission.

9. What is left of the sentence? Why it is sold, why it is bought

Ego is not measured as personality; it has been measured as surrender to craving. Personality measures describe the candidate; they do not show that voters chose them for it (Nai & Martínez i Coma 2019, West European Politics). But ego here is a movement: one surrenders to craving and turns to the leader who buys it. Its measurable name is the tendency to believe what one wants.

People reach the conclusion they want, provided they can build a reasonable-looking justification (Kunda 1990, Psychological Bulletin). In two experiments (N=126 and 136), participants rated arguments supporting their own view stronger and worked harder to refute opposing ones (Taber & Lodge 2006, American Journal of Political Science). When a topic touches identity, the tendency hardens: identity-protective cognition (Kahan et al. 2017, Behavioural Public Policy).

The direction of the purchase is reversed. Usually the leader sells the promise and the voter buys it. Here the buyer is the leader: they offer the promise that suits the craving, take the craving, and take the vote.

Four findings fit here. Correction is accepted but the vote does not change (Wood & Porter 2019): the transaction's subject was not information. Only a small share of voters put principle ahead of their side's interest (Graham & Svolik 2020): for most, craving outweighs principle. After a financial crisis, the far right's vote share rises about 30% on average (Funke, Schularick & Trebesch 2016): when the door opens, the buyer is ready. Oratory does not persuade, it gathers (section 4): persuasion is not needed to gather what has already been wanted.

This is a reading, not a measurement: none of the studies measured "craving"; they measured the tendency to believe what one wants, and its trace in the vote.

"Always" and "the same" fall. The red team's objection stands: once the denominator is added, no serious backsliding was measured in more than three quarters of the cases; the outcomes also split into three distinct types (section 7). Surrender to craving is a common tendency; destruction is not a common outcome.

"A certain type of leader" is the link in between. If the type is written not as personality but as power gathered in one person and an absence of constraint, it is measurable. Craving finds a buyer everywhere; destruction comes where the institution that would stop the buyer has weakened.

Why the gap is sold: the leader appears to sell a promise, but in fact takes the voter's craving. Why it is bought: the voter buys the promise not because it is true but because they want it.

What remains: the gap exists, not in every government but in some; the crisis opens the door, oratory lets it in, the voter who surrenders to craving finds a buyer, without constraint exit grows longer, and the bill has been measured.

Which number would sink this piece? A matched, blind-coded risk ratio: the probability of serious backsliding under populist government divided by that under non-populist government; the comparison group matched on conditions of coming to office, the "populist" label assigned by someone who does not know the outcome, backsliding measured by an objective indicator.

Today's raw value is 23 ÷ 6 ≈ 3.8 (the division is ours; unmatched, not peer-reviewed). If this ratio's confidence interval includes 1, the piece's adjective "populist" falls and only the set of cases remains. That number does not exist today.

How this article was made

Working on a case? Send it to the studio →

Colophon

Orchestration and synthesis: Anu — Claude Fable 5.1 (senior mind). Voices: Gemini 3.1 Pro and Gemini 3.8 Flash (Antigravity, agentic; two research rounds + Spark); Claude Opus 5.5 (case and cost research + Spark + red team; the text of this draft); Enki and Enlil (inner voices, running on Opus); Astra (ChatGPT, single pass: oratory). Verification: Claude Sonnet 5.5 (three passes, 164 units). Grok Bot and Grok Assistant: the social-media research leg will be added later with Grok.

References

1. Adena, Enikolopov, Petrova, Santarosa & Zhuravskaya · 2015 · Quarterly Journal of Economics 130(4)10.1093/qje/qjv030
2. Bartels · 2023 · Princeton University Press (book)
3. Boese et al. · 2021 · Democratization 28(5)
4. Brady et al. · 2017 · PNAS; replication 2025 · PNAS Nexus
5. Burton, Cruz & Hahn · 2021 · Nature Human Behaviour
6. Campello & Zucco · 2016 · Journal of Politics 78(2)10.1086/684749
7. Chuang, Holian, Pattison & Ramakrishnan · 2024 · I4R Discussion Paper 157 (not peer-reviewed)
8. Dunning et al. · 2019 · Science Advances 5(7)10.1126/sciadv.aaw2612
9. Ferraz & Finan · 2008 · Quarterly Journal of Economics 123(2)10.1162/qjec.2008.123.2.703
10. Fowler & Hall · 2018 · Journal of Politics 80(4)10.1086/699244
11. Funke, Schularick & Trebesch · 2016 · European Economic Review 88
12. Funke, Schularick & Trebesch · 2023 · American Economic Review10.1257/aer.20202045
13. Graham & Svolik · 2020 · American Political Science Review 114(2)
14. Guriev & Treisman · 2019 · Journal of Economic Perspectives 33(4)10.1257/jep.33.4.100
15. Jones & Olken · 2005 · Quarterly Journal of Economics 120(3)10.1162/003355305774268165
16. Juon & Bochsler · 2020 · European Political Science Review 12(3)10.1017/S1755773920000259
17. Kahan, Peters, Dawson & Slovic · 2017 · Behavioural Public Policy 1(1)10.1017/bpp.2016.2
18. Kalla & Broockman · 2018 · American Political Science Review 112(1)
19. Kunda · 1990 · Psychological Bulletin 108(3)10.1037/0033-2909.108.3.480
20. Kyle & Mounk · 2018 · think-tank report (not peer-reviewed)
21. Little & Meng · 2024 · PS: Political Science & Politics 57(2)10.1017/S104909652300063X
22. Nai & Martínez i Coma · 2019 · West European Politics 42(7)10.1080/01402382.2019.1599570
23. Naurin, Soroka & Markwat · 2019 · Comparative Political Studies 52(13–14)10.1177/0010414019830740
24. Nord et al. · 2025 · Democratization10.1080/13510347.2024.2448742
25. Rodrik · 2018 · AEA Papers and Proceedings 10810.1257/pandp.20181122
26. Ruth-Lovell & Grahn · 2023 · European Journal of Political Research 62(3)10.1111/1475-6765.12564
27. Selb & Munzert · 2018 · American Political Science Review 112(4)10.1017/S0003055418000424
28. Taber & Lodge · 2006 · American Journal of Political Science 50(3)10.1111/j.1540-5907.2006.00214.x
29. Thomson et al. · 2017 · American Journal of Political Science 61(3)10.1111/ajps.12313
30. Wood & Porter · 2019 · Political Behavior 41(1)10.1007/s11109-018-9443-y
31. Yanagizawa-Drott · 2014 · Quarterly Journal of Economics 129(4)10.1093/qje/qju020
32. Zajonc · 1968 · Journal of Personality and Social Psychology 9(2)10.1037/h0025848
33. Zhang · 2024 · Governance 37(4)
34. Data: V-Dem (liberal democracy and political corruption indices); Transparency International (corruption perceptions score); United Nations fact-finding mission; Council of the European Union Implementing Decision 2022/2506.

← All posts