The AI Mirage: Big Claims, Thin Capability, Weaker Firms

The corporate race to appear "AI-ready" has created a new credibility problem. Companies can announce ambitious strategies, highlight intelligent products and fill annual reports with artificial-intelligence language long before they build the data infrastructure, technical talent and operational systems needed to support those claims. The immediate reward may be attention, but the longer-term cost may be a weaker organization, warns a new study published in the journal Systems.

Authored by Yufei Xia, Jikang Sun, Jiarun Liu, Kun Fang, Huiyi Shi and Na Li of Jiangsu Normal University's Business School, the study titled "Beyond AI Narratives: AI Washing and Organizational Resilience," examines whether companies become more fragile when their AI narratives move faster than their actual AI investment.

Using 35,764 firm-year observations from Chinese A-share listed companies between 2010 and 2024, the researchers created an "AI washing" index by comparing how heavily a company discussed AI with how much it invested in identifiable AI-related software and hardware. A one-standard-deviation increase in AI washing was associated with a decline in organizational resilience equivalent to about 3.276% of the average annual change in the study's resilience measure.

AI communication becomes dangerous when it creates expectations that the company cannot operationally support. The gap between promise and capability can weaken investor confidence, strain financing relationships and destabilize supply chains, turning a communications strategy into an organizational risk.

The Real AI Risk: False Preparedness

AI washing resembles greenwashing, but the technology's opacity makes it especially difficult to detect. A company's environmental claims can sometimes be tested against emissions, energy consumption or waste data. AI capability is harder to verify from outside. Businesses may use familiar terms such as machine learning, intelligent decision-making, smart platforms or automation without revealing whether those systems are deployed, commercially relevant or integrated into core operations.

The study defines AI washing as a mismatch between AI-related narratives and substantive investment or capability formation. Disclosure itself is not treated as deceptive. The problem emerges when external communication implies a level of readiness that the company's spending, personnel, infrastructure and implementation cannot substantiate.

AI can genuinely improve organizational resilience. Properly deployed systems may strengthen information processing, identify emerging risks, support resource allocation and help firms respond more quickly to disruption. But those benefits depend on real capabilities.

When the narrative advances without the underlying system, management and stakeholders may develop a false sense of technological security. Investors may overestimate the company's capacity to adapt. Executives may misjudge their own operational boundaries. Resources can flow toward presentation and impression management instead of data quality, employee skills, cybersecurity, infrastructure and process redesign.

This makes AI washing more than a misleading marketing practice. It can alter decisions inside the firm. A company that convinces itself it has become technologically resilient may delay the investments required to make that resilience real.

The study measures resilience through a composite of medium-term operating-revenue recovery and financial stability, with stock-return volatility reverse-coded to represent greater stability. That approach captures both sustained performance and the capacity to remain financially steady under uncertainty.

Its baseline analysis found a negative and statistically significant relationship between AI washing and resilience. The estimated coefficient remained negative when the researchers added industry-year and province-year controls, used conventional fixed-effects models and applied alternative machine-learning approaches.

The relationship also persisted in propensity-score-matched samples and under a deep instrumental-variable framework designed to address concerns that weaker firms might simply be more likely to exaggerate their AI capabilities. Those tests strengthen the result, although they do not eliminate every possible source of bias.

What Companies Build Matters More Than What They Announce

The most revealing finding is not merely that AI washing is associated with weaker resilience. It is that the direction of the gap matters. When companies talked more about AI than their investment suggested, resilience was lower. When substantive AI investment exceeded the scale of external AI narratives, resilience was higher.

The study reports a negative coefficient of –0.347 for firms whose disclosure ran ahead of investment and a positive coefficient of 0.469 for companies whose investment exceeded their public narrative. More severe AI washing was associated with a larger resilience penalty. That asymmetry changes the policy and management interpretation.

The research is not evidence that AI disclosure is inherently harmful. Nor does it imply that companies should understate genuine technological progress. It suggests that resilience is associated with capability formation, not promotional intensity.

For corporate leaders, this creates a straightforward test. Every material AI claim should be traceable to something operational: an approved investment, a functioning system, a trained technical team, an accessible data resource, a deployed business use case or a measurable implementation milestone.

The study's AI-disclosure measure was built using a large language model that reviewed corporate annual reports. The researchers manually audited 3,000 sentences and reported a Cohen's kappa of 0.771 among human coders. Qwen-max, the selected model, achieved precision of 0.9557 and an F1 score of 0.9540 against the manually coded benchmark.

Actual AI investment was measured through spending on AI-specific software and hardware relative to total assets. General information technology, office software, enterprise-management systems and broad digital transformation were excluded unless the description clearly identified an AI function or application.

This is a stronger approach than simply counting AI-related words, but it also exposes a measurement challenge. Important capability-building may occur through cloud services, data cleaning, staff training, external consulting, workflow redesign or internally developed tools that do not appear neatly as AI software or hardware investment.

The Credibility Gap Spreads Into Capital, Governance and Supply Chains

The study also explains how AI washing becomes organizationally damaging. The authors identify three legitimacy-related pathways: pragmatic, moral and cognitive.

Pragmatic legitimacy concerns whether stakeholders believe the company can create real value. AI washing was associated with lower patient capital and weaker net trade-credit support. Long-term investors may become less willing to provide stable financing when they doubt a firm's technological foundations. Suppliers and customers may also become less willing to extend credit when inflated claims raise concerns about operating quality and reliability.

The estimated indirect effect through patient capital accounted for 11.475% of the total relationship, while net trade credit accounted for 2.298%. These are model-based mediation estimates, but they suggest that unsupported AI claims can shrink the financial buffers firms need during disruption.

Moral legitimacy concerns whether management is viewed as credible and responsible. The study links AI washing with indicators of opportunistic executive share reductions. The interpretation is not that every share sale signals misconduct, but that unsupported technology narratives combined with insider selling may intensify doubts about managerial motives.

Cognitive legitimacy concerns whether stakeholders can understand the company's strategy and trust its operational coherence. AI washing was associated with higher supply-chain risk and lower supply-chain resilience. The mediation share through supply-chain resilience was estimated at 6.185%.

AI washing reduces stakeholder recognition of practical value, weakens governance credibility and lowers operational coherence, ultimately undermining a company's ability to absorb, restore and adapt after shocks. The effect was not evenly distributed. In high-tech industries, a one-standard-deviation increase in AI washing was associated with a resilience decline equivalent to 4.176% of the average annual change, compared with 2.253% among non-high-tech firms.

The penalty was also greater among companies with established bank relationships. The corresponding magnitude was 4.461% for firms with bank ties, compared with 1.889% for those without. Banks may initially provide stability, but they also have more information and stronger incentives to punish a credibility gap once it becomes visible.

The Next AI Governance Fight Will Be About Proof, Not Promises

The study points toward a new phase of corporate AI governance. So far, much of the public debate has focused on safety, bias, privacy, intellectual property and workforce disruption. Corporate disclosure presents another problem: how should investors, lenders, regulators and employees distinguish genuine deployment from sophisticated technology storytelling?

The authors recommend that companies treat AI disclosure as a capability-backed process rather than a communications exercise. Internal review should involve technology, finance, strategy, compliance and investor-relations teams. Firms should be able to support major claims with evidence of budgets, systems, staff, data, applications and measurable deployment progress.

For investors and creditors, AI terminology should not be treated as evidence of technological readiness. Assessment should compare public claims with AI-specific investment, hiring, patents, research activity, business applications and operational integration.

Regulators could encourage companies to separate AI-specific spending from general digitalization, identify where systems are actually deployed and explain how AI-related risks are governed. The study does not establish that a particular disclosure rule would work, but it makes a persuasive case for greater verifiability.

There are also reasons for caution. The evidence comes only from Chinese listed firms and may not generalize directly to private companies or businesses in different regulatory and capital-market systems. Unobserved factors, including strategic foresight, management quality, risk culture and digital-governance maturity, may affect both AI washing and resilience. The legitimacy indicators are proxies, not direct measurements of what stakeholders believed.

Future research should test the relationship across countries, ownership structures and disclosure regimes. Surveys and interviews could directly measure how employees, investors, banks and suppliers respond to AI claims. Regulatory shocks or new disclosure requirements could offer stronger causal evidence.