The queryable war: When AI turns ‘could’ into ‘must’

The US–Iran war reveals a quieter power of generative AI: it can turn unfolding reports into plausible futures, reshape what citizens expect, what leaders think they can politically afford and what adversaries prepare to resist—without falsifying a single fact.

The question “what happens next?” is as old as war. What is new is that almost anyone can now ask a machine to build the answer while reports are still arriving. Ask a web-connected chatbot what would follow another round of US strikes on Iran and it may return several orderly futures: Iranian retaliation in the Persian Gulf, a wider regional confrontation, an energy shock or a negotiated exit.

These are not reports of events already established; they are constructed futures. The system has not merely retrieved information about capabilities, intentions and previous behaviour. It has selected facts, connected them causally and assembled worlds that do not yet exist. Users can vary the premises: What if Washington narrows its aims? What if Israel escalates while the US seeks an exit? Much has been written about AI targeting and synthetic propaganda. The deeper change is that machines are becoming everyday producers of the futures through which war itself is judged. The US–Iran war is no longer only watched, reported and searched. It is queried.

The industrialisation of the possible

Television made distant war visible. Search made information about it retrievable. Social media made war immediate, participatory and viral. Generative AI adds a different capacity: it lets users question reports, vary assumptions and simulate what has not happened. A search engine retrieves sources about the world. A generative system can assemble a possible world. It is emerging not merely as a channel of information, but as a mass medium of the possible.

When Russia launched its full-scale invasion of Ukraine in February 2022, ChatGPT did not yet exist; many early chatbots had dated training horizons. Leading assistants now search the web, while scheduled monitoring tasks can revisit a standing question and notify users of meaningful change. The query can outlive the conversation: users delegate interpretation and decisions about what merits attention. The interval between event, report, retrieval and scenario is collapsing. The model supplies inherited causal patterns; web retrieval supplies fresh triggers. An event can become a narrated future before the evidence has settled.

This remains a minority practice. The Digital News Report 2026 found that weekly use of chatbots for news had risen from seven to ten per cent, while only one per cent treated them as their main news source.

Yet 42 per cent of chatbot news users ask follow-up questions. They are not merely requesting headlines; they are interrogating what the news means and where it may lead. Strategic foresight is not new; mass queryability is.

Scenario-making is escaping specialist institutions and entering everyday news consumption.

Call this possibility inflation: coherent futures are becoming cheaper to produce than to audit. A chatbot can generate another scenario in seconds; testing its assumptions, omissions and probabilities can take hours—and remain impossible while conflict unfolds. The supply of politically usable futures therefore expands faster than our ability to rank them.

Web access does not mean access to the whole present. A system may reformulate the user’s question into searches, retrieve and rank part of an uneven information record, then decide what futures to construct from it. A same-day news evaluation of six commercial chatbots traced more than 70 per cent of their errors to retrieval rather than reasoning and exposed an Anglophone source bias. Selection therefore happens twice: first among sources, then among the futures built from them. The interface hides two politically consequential absences: the sources never retrieved and the futures never shown. Omitted evidence can become omitted futures.

Consider two answers built from the same evidence about the US–Iran conflict. One foregrounds prolonged attrition, wider regional escalation, economic disruption and an eventual negotiated stalemate. Another foregrounds Iranian military exhaustion, restored US deterrence, growing bargaining pressure and negotiated concessions.

Neither answer needs to invent a fact. Yet the first makes continued force look increasingly futile, while the second makes the same costs look like the unfinished price of success.

Their political divergence lies in emphasis, order and where each causal story ends. A user is rarely given a defensible basis for weighting the scenarios displayed. Generate enough fluent futures and at least one catastrophe may look too plausible to ignore. It need not be the most likely outcome; its consequences need only appear grave enough that declining to prepare for it looks irresponsible. An increase in imaginable danger is not necessarily an increase in danger, but politics may react to both. A model can change the politics of war without falsifying a single fact; it need only change the menu of futures.

When “could” becomes “must”

The decisive movement may occur inside a handful of verbs. A response might run as follows: Iran can disrupt shipping in the Persian Gulf. Under continued pressure, it may do so. If confrontation persists, such disruption becomes likely. The United States therefore must prevent it. Preventive force then appears justified. But can describes capability, may identifies possibility, likely makes a probabilistic judgement, must introduces a prescription and justified asserts legitimacy. These categories do not entail one another. Capability is not probability. Probability is not necessity. Necessity does not create legitimacy.

Generative prose can make the passage between those categories feel frictionless. It can fill each gap with reasons, precedents and qualifications until a conceivable danger arrives as a policy imperative. The mechanism is not inherently hawkish: the model might reason that further strikes could widen the war, the risk may become intolerable and leaders should therefore negotiate. Its power lies in altering the threshold at which a possible future becomes a reason for present action. Historical analogy performs similar work in compressed form. An analogy is a scenario with its ending already attached: call a crisis “Munich” and restraint resembles appeasement; call it “Vietnam” and persistence resembles entrapment.

Citations alone do not stop this movement. A response may accurately source missile ranges, oil flows, deployments, official statements and previous conduct. Those sources can support a scenario’s premises; they cannot certify the future built from them. Nor does a report’s recency guarantee its truth amid the fog of a developing war. Sources document the bricks. They do not certify the house. A named analyst can be questioned about assumptions and method. A chatbot answer can carry authority without an accountable forecaster. Coherence is not calibration.

Why should this affect a war? Because wars are judged prospectively. Research on US war support has long shown that expectations of eventual success shape the public’s willingness to tolerate casualties. Policy-persuasion experiments show that AI-generated messages can shift attitudes, although they do not test war scenarios. An August Reuters/Ipsos poll found that 50 per cent of Americans expected US military action to bring greater regional instability, while 17 per cent expected more stability and 35 per cent supported the war. Those figures do not implicate AI. They show the terrain on which it may matter: public judgement of war is partly a judgement about futures.

The path from a private answer or alert to policy is neither automatic nor singular. These private, personalised exchanges are mediated by a handful of systems; their effects may surface in polls, media demand, party rhetoric and mobilisation. Professional intermediaries offer another route.

A broadly representative survey of UK journalists found that 56 per cent used AI professionally at least weekly and documented its use in story research, brainstorming and drafting. This gives AI a place inside the questions, frames and texts through which scenarios reach a wider public.

Analysts and officials may use such systems to select contingencies for briefings and preparation. Recent work on synthetic foresight has begun theorising how GenAI scenarios might reshape strategic crisis decision-making.

What matters is political permission. Public opinion does not issue orders to leaders. It changes the anticipated political price of their choices. Expectations of victory may enlarge the room for escalation or endurance; expectations of stalemate, economic pain or regional chaos may narrow it. Nor must a leader personally trust an AI-generated scenario. Leaders act on expectations about others: what voters will tolerate, whether allies will remain committed, how markets will react and what the adversary thinks domestic audiences can bear. A generated future gains force when actors believe other actors are beginning to organise around it.

Imagine that a quagmire scenario becomes salient across chatbot answers and public debate. Citizens lower expectations of success; US leaders anticipate shrinking support and seek an exit. Iranian decision-makers infer that delay may improve their position. Resistance prolongs the confrontation, moving events towards the scenario. Escalatory feedback is equally possible: precaution becomes observable military movement, the opponent reads hostile intent and responds.

A second loop is informational. Once AI-assisted analysis is published, it can enter the retrievable record used to produce later answers. AI systems may therefore search an information environment they have partly helped to write. Repetition can increase retrievability without increasing probability. No available evidence shows that chatbot answers caused a particular US, Iranian or Israeli decision in this war. The argument identifies an emerging mechanism, not a retrospective verdict. Yet an imagined future that triggers action can begin producing its own evidence.

The traditional fog of war arose from too little reliable information. Generative systems may reduce some of it through retrieval and synthesis, while adding another problem: too many coherent futures and too few grounds for ranking them. The task is not to suppress scenarios. It is to keep capability separate from possibility, possibility from probability, probability from prescription and prescription from legitimacy. The decisive question is not merely which futures can be generated, but which one gains enough authority to reorganise the present. In the queryable war, power operates in the grammar between “could” and “must”.

The views expressed in this article belong to the author and do not necessarily reflect the editorial policy of Middle East Monitor.