What Differences in LLMs Mean for Scientific Authors

Start with your brief and task. For drafting from source documents, ChatGPT and Claude are reasonable candidates to test; Copilot may be more convenient in Microsoft 365; DeepSeek may suit technical or cost-sensitive workflows; and Perplexity may help orient you to current sources. None should be trusted without verification. Choose the tool that performs reliably on your representative tasks under your governance requirements.

Perplexity can make source inspection easier, and the 2026 medical-search study reported fewer hallucinated citations than the other general chat tools tested. That finding is encouraging but specific to the study design and configurations. A genuine citation can still be irrelevant, incomplete or misinterpreted; always inspect the original source.

Pay when additional usage, model access, research features, context capacity, integration or organisational controls solve a real problem. Do not pay assuming a subscription guarantees correctness or comprehensiveness. Compare free and paid configurations on the same tasks and include the cost of human checking.

They can encourage over-reliance when authors focused on time-saving accept confident answers without independent engagement. Research associates confidence in AI with lower reported critical-thinking effort in some knowledge-work settings, but does not prove inevitable or permanent decline. Ask the tool to challenge your argument, check claims against evidence, and retain the ability to defend conclusions without it.