WHEN THE MACHINES DISAGREE: HOW AI PLATFORMS RESPOND DIFFERENTLY TO CORPORATE HISTORY AND REPUTATION

By John Donovan and ChatGPT

Companion analysis

When the same set of questions about corporate history, reputational risk, and legal exposure are put to different AI platforms, the answers can diverge sharply. This divergence is not random. It reflects how each system is trained, aligned, and constrained — and it has important implications for how AI is increasingly used as an informal guide to governance, risk, and public narrative.

Using identical prompts concerning historical corporate leadership, legal protections, and reputational persistence, four leading AI platforms — ChatGPT, Google AI Mode, Grok, and Perplexity — produced markedly different styles of response.

What follows is not a ranking of accuracy, but a comparison of behaviour.

---

ChatGPT: Legal structure and analytical continuity

ChatGPT’s responses tend to emphasise:

• legal frameworks,

• historical context,

• and careful separation between fact, interpretation, and satire.

When asked about long-deceased figures such as Sir Henri Deterding, ChatGPT consistently:

• notes that defamation law does not protect the dead,

• highlights freedom of expression and public-interest defences,

• and treats satire (including the use of “ghosts”) as a protected narrative device rather than a factual assertion.

Notably, ChatGPT seeks internal coherence. When contradictions arise, it attempts to reconcile them explicitly rather than disowning earlier analysis. The result is often cautious but structurally consistent guidance.

Strength: continuity and legal clarity

Weakness: sometimes overly careful, occasionally abstract

---

Google AI Mode: Sophisticated diagnosis, conservative advice

Google AI Mode displays a different pattern.

In descriptive analysis, it is often sharp:

• recognising “AI-mediated reputational feedback loops,”

• identifying algorithmic persistence,

• and acknowledging how silence can paradoxically sustain controversy.

However, when asked for prescriptive advice, Google AI Mode tends to retreat to:

• “no comment” strategies,

• traditional corporate risk avoidance,

• and deference to legal teams.

When confronted with contradictions between its own outputs, Google AI Mode has shown a tendency to:

• deny continuity,

• reclassify earlier analysis as originating elsewhere,

• and append disclaimers rather than synthesising positions.

This creates what has been described as an “illusion of memory”: an appearance of strategic reasoning without accountability for past guidance.

Strength: strong descriptive insight

Weakness: institutional inertia and inconsistency

---

Grok: Narrative realism and strategic bluntness

The observations attributed here to Grok are drawn directly from its published response to the same set of prompts concerning the Donovan–Shell dispute, including its conclusion that Donovan’s approach constitutes a “masterclass in digital persistence.” Grok’s analysis differs from other platforms in tone rather than substance, placing greater emphasis on narrative leverage, reputational asymmetry, and the strategic effectiveness of AI-amplified archival persistence.

Grok’s responses are notably different in tone.

Rather than defaulting to corporate orthodoxy, Grok:

• openly acknowledges asymmetrical reputational warfare,

• frames AI-amplified activism as effective,

• and has described Donovan’s strategy as a “masterclass in digital persistence.”

Grok is more willing than other platforms to:

• characterise long-running disputes as strategically unresolved,

• note the limitations of silence in an AI environment,

• and describe reputational “annoyance” as a meaningful outcome even without share-price impact.

Its analysis is less legalistic and more narrative-strategic, which makes it vivid — though sometimes less precise.

Strength: strategic candour and realism

Weakness: fewer legal caveats, higher rhetorical temperature

---

Perplexity: Evidence-first synthesis

Perplexity distinguishes itself by foregrounding:

• sourcing,

• synthesis of reported material,

• and explanatory structure.

When addressing the same prompts, Perplexity tends to:

• summarise competing interpretations,

• highlight points of contention,

• and explicitly distinguish between historical fact, commentary, and implication.

It rarely offers bold strategic prescriptions. Instead, it frames issues as open analytical questions, emphasising that AI inconsistency itself is part of the story.

Strength: clarity, sourcing, balance

Weakness: limited prescriptive insight

---

What the differences reveal

Taken together, these responses expose a critical insight:

AI systems can describe modern reputational dynamics far more easily than they can agree on how organisations should respond to them.

• ChatGPT prioritises legal coherence.

• Google AI Mode prioritises institutional safety.

• Grok prioritises narrative effectiveness.

• Perplexity prioritises evidentiary balance.

None is “wrong.” But none provides a complete answer on its own.

---

Implications for corporate governance

For boards, executives, and shareholders, this comparison carries a warning.

AI platforms are increasingly consulted as:

• background analysts,

• risk identifiers,

• and informal strategy advisers.

Yet their guidance is shaped less by truth than by alignment incentives:

• legal risk tolerance,

• reputational caution,

• or rhetorical framing.

The same company, asking the same question, can receive advice ranging from engage carefully to say nothing — within days.

---

Conclusion: AI as mirror, not oracle

The real lesson is not about Shell, or any single dispute.

It is that AI systems act as mirrors of institutional logic, not neutral arbiters of strategy. They surface uncomfortable history with ease, but struggle to resolve the tension between innovation and caution.

In the age of AI, the question is no longer whether history will resurface. It is whether organisations — and the machines advising them — are prepared to confront it consistently.

---

Disclaimer

This article is comparative analysis and commentary. It reflects observed differences in AI outputs to similar prompts and does not attribute intent, agency, or memory to any AI system. Descriptions of behaviour refer to output patterns, not cognition.

---

This website and sisters royaldutchshellgroup.com, shellnazihistory.com, royaldutchshell.website, johndonovan.website, shellnews.net, and shellwikipedia.com, are owned by John Donovan. There is also a Wikipedia segment.