Geopolitical AI-Trendslop

LLMs default to consensus mimicry. Start there and apply Human Judgment

After the Storm, c. 1700, Willem van de Velde the Younger

A new study published in the Harvard Business Review last week reminds us that large language models (LLMs) need to be paired with human judgment, especially in a field as hard to quantify as geopolitics.

Business school researchers tested seven leading large language models — including ChatGPT, Claude, Gemini, Grok, and DeepSeek — across seven core strategic tensions that required binary choices. The finding was consistent across all models: LLMs systematically recommend the same buzzword-aligned strategies regardless of business context.

The researchers call this “trendslop”: AI’s structural bias toward whatever sounds most current and defensible — differentiation over cost leadership, augmentation over automation, long-term thinking over short-term urgency — independent of what the specific situation requires.

The scale of the evidence is significant. Over 15,000 trials were run to determine whether better prompting could correct the bias. It could not: the strongest biases shifted less than 2%. Providing richer organizational context moved responses by only 11% on average.

The researchers also documented the “hybrid trap.” When not forced to choose, LLMs recommended pursuing contradictory strategies simultaneously, producing the appearance of analytical completeness while resolving nothing.

The study was designed around generic corporate strategy questions. The implications for AI-assisted geopolitical analysis are more serious still.

Why Geopolitics Is a Harder Case

Trendslop in corporate strategy produces suboptimal advice. A model that consistently recommends differentiation over cost leadership when cost leadership is correct wastes time and introduces drift.

Trendslop in geopolitical analysis can produce unrecognized exposure.

Geopolitical analysis is not generic by definition. It is inherently firm-specific, jurisdiction-specific, and time-horizon-specific. The material questions are not whether market fragmentation is accelerating in general, but whether it is accelerating in the specific configurations on which a particular firm depends.

An LLM trained on the accumulated discourse and commentary on international relations will have absorbed a set of consensus positions: the received wisdom on which risks are salient, which actors matter, and which scenarios are credible. It will reproduce those tendencies with confidence and structural coherence. The output will read as rigorous. It will not be firm-specific.

The hybrid trap compounds this. A model asked whether a firm should reduce China exposure or maintain it will, unless constrained, recommend doing both, presented as “de-risking while preserving optionality.” That is not a geopolitical judgment. It is a hedge dressed as analysis.

The same structural feature that makes LLMs unreliable strategic advisors for generic questions makes them actively misleading for geopolitical ones: they smooth the tensions that the analysis is supposed to resolve.

The Prior Question

Before asking what AI can contribute to geopolitical analysis, firms need to answer a more fundamental question: what is the business case for paying attention to geopolitics in the first place?

Geopolitical attention is not uniformly distributed across industries, geographies, or business models. A firm with concentrated revenue in a single domestic market, a domestically sourced supply chain, and no cross-border regulatory exposure faces a fundamentally different calculus than a firm whose operations span multiple jurisdictions, whose capital is denominated across currencies, and whose regulatory licenses are subject to extraterritorial reach.

The business case for geopolitical analysis is a function of exposure. That exposure must be mapped before analysis begins. Without it, the firm has no basis for evaluating which geopolitical developments are material and which are background noise.

This is the step AI-assisted analysis almost always skips. A model asked about, say, U.S.-China trade policy will generate a technically competent overview of tariff structures, export controls, and bilateral tensions. It will not tell the firm whether any of this is existential to its business, consequential but manageable, or essentially irrelevant to its specific configuration.

Only the firm can answer that. And the answer requires structured self-knowledge that many firms do not yet have: a mapped, maintained representation of their own geopolitical exposure across asset geography, supply-chain structure, regulatory architecture, capital denomination, and institutional relationships.

Without that foundation, AI-generated geopolitical analysis is trendslop applied to the wrong question.

The Human in the Loop

The HBR researchers’ prescription is precise: use LLMs to generate alternatives and stress-test assumptions, not to make choices. Actively prompt for opposing strategies to surface biases. Treat hybrid recommendations as a warning sign, not sophisticated thinking.

Applied to geopolitical analysis, this prescription has a specific institutional form.

The human in the loop is not a reviewer who checks AI outputs for factual errors. That function addresses the wrong failure mode. The problem is not that AI gets facts wrong. It is that AI gets the framing right and the judgment wrong.

The human function is threefold.

First, anchoring the analysis in the business. This means establishing, before any AI tool is engaged, which geopolitical configurations are material to the firm’s revenue, operations, and governance. This usually cannot be extracted from a strategy document, which is often written as a public-facing document. It requires structured elicitation from operational, legal, financial, and strategy leadership, surfacing the tacit and latent knowledge about how the firm actually works that no model can retrieve.

Second, forming and holding a conviction. Trendslop is, at its core, the absence of a committed analytical position. A model that recommends differentiation and cost leadership simultaneously has not taken a view. The human function is to take one, provisionally, revisably, but specifically. That conviction then becomes the input that directs AI-generated analysis rather than outsources it.

Third, forcing contact with reality. AI is most useful once a firm-specific conviction is in place: extending it across scenarios, tracing second-order effects, and identifying indicators that would confirm or contradict it. This is compute applied to a judgment already formed. It is not a substitute for forming that judgment.

The 11% finding from the HBR research is instructive here. Even with detailed organizational context, AI-generated analysis retained 89% of its trendslop bias. That residual is not a prompting failure. It is an architectural constraint. The model does not know what is existential to the business because that knowledge is not in the training data. It lives in the firm, in what is written in non-public documents and in the unwritten knowledge, aptitudes and practices of corporate decision-makers.

What This Produces

A firm that approaches AI-assisted geopolitical analysis this way — starting from mapped exposure, holding a working conviction, using AI to stress-test rather than originate — gets outputs that are qualitatively different from trendslop.

Instead of a balanced overview of geopolitical tensions, it gets a scenario-tested assessment of which of those tensions intersect with its specific exposure vectors.

Instead of a hybrid recommendation to both maintain and reduce presence in a given jurisdiction, it gets a structured stress-test of what maintaining that presence costs under different policy trajectories.

Instead of analysis calibrated to what sounds defensible, it gets analysis calibrated to what requires a decision.

Five recurring applications illustrate the difference:

1. Conditional market access. For firms exposed to China, what happens if regulatory approval or licensing becomes contingent on political alignment? For operations tied to the United States, how do outbound investment screening and export control expansions affect what can be sold, financed, or transferred? This determines which revenues are durable and where compliance becomes a geopolitical variable.

2. Instrumental interdependence. Which dependencies that appear efficient are no longer resilient? The disruption of Russian gas flows to Germany’s industrial base is the reference case. The question for any firm: where does your supply chain have a similar structure, and what is the unwind cost?

3. Regulatory sovereignty. How do data localization rules, export restrictions, and resource nationalism reshape operating models in specific jurisdictions? The question is not whether these trends are real. It is whether the firm’s business model aligns with state priorities in the markets where it operates.

4. Alignment over rules. Compliance is necessary but not sufficient. How does regulatory treatment vary within the EU depending on political positioning? Positioning matters independently of formal compliance.

5. Capital as leverage. How do U.S. financial controls affect access to dollar systems? How are hubs like the UAE positioning as alternative conduits? Where is the firm’s capital configuration exposed to state-directed restriction?

Gradual. Then All at Once.

Geopolitical systems are brittle.

They absorb pressure: incremental restrictions, localized disruptions, partial decoupling. Then thresholds are crossed and behavior changes more abruptly.

For firms, this is when exposure crystallizes: market access becomes conditional, supply chains must be redesigned, and capital flows are constrained.

Analysis anchored in continuity — which is precisely what trendslop produces — misses this transition. Not because the signals are absent. Because the framework assumes stability.

A firm operating from trendslop-derived geopolitical analysis is, structurally, a firm reading yesterday’s consensus into tomorrow’s decisions. The analytical product looks current. The underlying orientation is retrospective.

The Edge

The HBR study defines the mechanism of AI-generated analytical failure with precision: models produce what sounds most current and defensible, not what is most decision-relevant.

For geopolitical analysis, this is the difference between a firm that understands its exposure and one that has a well-formatted description of the world. This is the difference between making risk-based decisions and admiring the problems in world politics.

The differentiator is not access to better AI tools. It is the institutional capacity to use those tools correctly: starting from a mapped understanding of the firm’s own geopolitical exposure, forming a working conviction about the configurations that matter, and deploying AI to test and extend that conviction rather than generate it.

Firms that build this capacity will make better decisions than those substituting the tool for judgment. In a system where structural change accumulates quietly and then manifests abruptly, that difference compounds.

Romasanta, A., Thomas, L.D.W., & Levina, N. (2026). “Researchers Asked LLMs for Strategic Advice. They Got ‘Trendslop’ in Return.” Harvard Business Review, March 16, 2026.