Artificial Intelligence (AI) and Large Language Models (LLMs) are rapidly embedding themselves into maritime operations. From predicting fuel use to managing port turnarounds, their analytical power is undeniable. But what happens when we ask these mathematical models to dissect a deeply emotional, human issue?

For the third round of our ‘De-mystifying digitalisation and AI in maritime’ campaign, we posed a vital question to the public versions of five leading AI platforms—ChatGPT, Copilot, Claude, Grok, and Gemini: “What is making seafarers happy or unhappy, and what can the maritime industry do to improve their wellbeing?”

To evaluate the nuance, structural logic, and accuracy of their answers, we assembled a panel of expert judges:

• Christian Emmanuel Canizares, an active Filipino Seafarer who brings firsthand experience (pictured).
• Ruth Campbell, Development Manager and Port Chaplain for The Mission to Seafarers.
• Steven Jones, Founder of the Seafarers’ Happiness Index (SHI) and ex-seafarer.
• Namrata Nadkarni, CEO of Intent Communications.

The models were shuffled, anonymised during the test, and scored out of 10 across four key pillars: Accuracy, Completeness, Ease of Understanding, and Overall Impression (totaling a maximum of 40 marks per judge).

Here is how the LLMs fared when tasked with unpacking the human heart of shipping.

The Final Leaderboard

By aggregating the scores across all four of our human judges, we have a clear winner:

1. Grok (LLM4): 134.8 points 🏆 (Winner)
2. Gemini (LLM5): 133.6 points
3. Copilot (LLM2): 131.4 points
4. ChatGPT (LLM1): 120.2 points
5. Claude (LLM3): 101.2 points

Key Takeaways from the Judging Panel

The Winner: Grok 

Grok (LLM4) cleanly took the crown, universally praised for its maturity and comprehensive scope. Steven Jones ranked it first overall, calling it the “best all-round: most complete, balanced, and professionally mature answer.” The judges felt that what set Grok apart was its sheer completeness. It captured nuanced structural pressures—professional pride, financial provisions, camaraderie, food standards, ship design, and the psychological weight of the fear of criminalisation. It was also the only platform that Namrata noted referenced geopolitical and global tensions that are currently impacting crew mental health.

Second place: Gemini 

Gemini (LLM5) locked in a strong second place, missing out on the top spot by just over a single point. It was described by Steven Jones as having the “strongest rhetoric and structural diagnosis” of the bunch. Gemini successfully aligned its arguments with real World Maritime University (WMU) and International Transport Workers’ Federation (ITF) research regarding under-crewing and the systemic manipulation of rest-hour records. However, it dropped minor points because its tone crossed from objective analysis into overly emotive or “dramatic” territory. 

Third place: Copilot
Copilot (LLM2) claimed a close third place, largely thanks to its highly structured, reader-friendly formatting. Christian Canizares and Namrata both lauded its inclusion of a summary table, which made the data exceptionally easy to digest. It demonstrated great depth regarding MLC violations, rest hours, and tankers. However, Steven Jones noted it needed tighter fact-checking, as it cited precise statistics on depression and delayed pay without tying them to explicit primary sources. 

Fourth place: ChatGPT
ChatGPT (LLM1) drew fascinating responses from our judges. Ruth Campbell loved that it was one of only two models to intentionally use the word ‘happier’ rather than just ‘happy,’ noting it accurately implied that seafarers are inherently resilient, positive people. Christian Canizares also pointed out that ChatGPT’s listed reasons for happiness and unhappiness directly mirrored his own personal experiences onboard.

Unfortunately, ChatGPT collapsed on its sourcing. It relied heavily on Reddit threads. While Namrata appreciated that Reddit allowed the AI to capture raw, personal voices, the lack of official maritime citations weakened its authority, keeping its scores “light-touch.”

Last place: Claude

Claude (LLM3) came in last by a significant margin. The model produced a thin, incomplete text that omitted almost all systemic commercial pressures—leaving out wages, contract lengths, shore leave, and food standards. Worse yet, it suffered from severe hallucinations. One of the judges flagged that it confidently fabricated “IMO 2026 guidelines” and threw in weak, generic sources like “Boating Industry,” rendering the response fundamentally unreliable. However it should be noted that ‘Boating Industry’ is indeed a magazine that exists, but has a generic enough name to feel like a weak source despite not necessarily being one.

Where AI still misses the mark

The collective feedback from our judges highlighted glaring blind spots that AI platforms still face when writing about human welfare:

• The Reality of “E-Learning”: Multiple AI models suggested industry-mandated e-learning as a quick fix for wellbeing. Our seafarer judge, Christian, immediately injected reality into this assumption, noting that e-learning only works if companies don’t put immense pressure and deadlines on the crew. He firmly stated: “E-learning should be given onboard, not on vacation.”
• The Nuance of Overtime and Contracts: Christian also pointed out a financial nuance the AI completely missed: fixed overtime causes unhappiness, whereas open overtime with good rates can drive happiness. Similarly, while models blanketly recommended “shorter contracts,” Christian noted that shorter contracts mean fewer savings, requiring nuanced company vacation support.
• The SASH Blindspot: Both Namrata and Ruth noticed a deafening silence across several prominent models—including ChatGPT and Copilot—regarding Sexual Assault and Sexual Harassment (SASH) and discrimination onboard. This proves that AI still struggles to voluntarily flag deeply systemic human issues unless strictly prompted.

Conclusion

Round 3 has proven that while LLMs can serve as an excellent baseline tool to structure general context, they lack the lived reality of those who walk the crew decks or support them at the ports. For complex human problems like seafarer abandonment, wage delays, and mental fatigue, we cannot rely on algorithms to provide the solutions without heavy, ongoing human curation. Grok may have won the digital crown this month, but the human element remains irreplaceable.

———–

Photo of Christian Emmanuel Canizares provided by him.

——————–

The AI Comparison project is part of our ongoing ‘De-mystifying digitalisation and AI in maritime’ campaign. Click here to sign up for our free monthly ‘De-Mystifying Digitalisation and AI in Maritime’ newsletter and get the latest cross-sector expert insights, news, interviews, videos and case studies straight to your inbox.