Gemini: the Cognitive Scientist
Thinks in flowcharts, cites its sources, and silently judges your search history.
đ I shared a prompt template built using the six building blocks of effective prompt design, then asked four top AI models to provide feedback. Each one had strong (and different) opinions.
Keep reading for Geminiâs feedback, or jump to your favorite modelâs feedback below:
Gemini approached this prompt like a cognitive psychologist evaluating a training program. It wasnât just asking, âDoes this make sense?â It was asking, âDoes this teach the model how to think?â
It praised the foundation: clear roles, grounded data, and a well-structured example. But what caught Geminiâs interest wasnât the structure itself so much as the learning behavior the prompt encouraged. Does the format support logical inference? Can it handle ambiguity? Does it teach synthesis across multiple inputs?
In Geminiâs view, this was already a strong prompt. But it believed a few key additions could turn a good prompt into a reliably smart one.
Notes from the Cognitive Coach
Gemini liked the structure overall: clean, bounded inputs that reduce guessing and keep the model grounded in evidence. Its favorite part? The example. Beyond showing the output format in action, it modeled how to think: move from question, to reasoning, to a quote-backed answer. For Gemini, thatâs not formatting, thatâs pedagogy.
It also praised the use of <relevant_reviews> as a cognitive nudge. By prompting the model to reflect on the data first, the prompt encouraged chain-of-thought reasoningâespecially useful when answers require judgment.
But it saw room to improve the promptâs teaching strategy.
First, Gemini flagged the lack of support for contradiction. If reviews disagree, the model needs a way to surface both views without picking sides or ignoring tension. Gemini wanted an example that shows how to handle conflicting data transparently.
Second, it wanted stronger guidance around synthesis. The current prompt only uses one quote, but real questions often need a broader pattern. Gemini recommended renaming Quote to Relevant Quotes, and including multiple sources when needed.
đĄ Gemini was the only model to distinguish between explicit and implicit answers such as when a productâs long battery life implies suitability for long flights.
It didnât just want the model to make that leap, instead asking the prompt to train for it. Its core suggestion: allow these type of inferences by the model, but clearly state the logic behind them in the âWhyâ section of the output.
Gemini didnât ask for new capabilities. It asked for better instruction so the model can reason clearly, even when the data isnât obvious.
Geminiâs Strategic Priorities
Geminiâs feedback reflects Googleâs deep roots in information science. Answering the question as posed isnât all that makes a smart system. Instead it wanted it to reason transparently, like a good research assistant. The focus wasnât on style or speed, but on structured thought: identifying patterns, weighing evidence, and showing your work.
That tracks with Googleâs philosophy. Gemini is built to integrate into information workflows, such as search, docs, or code, where accuracy and clarity matter more than personality. Itâs not trying to be clever. Itâs trying to be trustworthy.
Where some models chase engagement or speed, Gemini wants reliability at scale. Its feedback was about teaching the model how to think out loud, not just get the right answer. Because when AI is part of your infrastructure, guesswork isnât good enough.
Takeaways for Prompt Writers
Gemini shines when you treat the model like a reasoning partnerânot just a word machine. If your prompts rely on judgment, ambiguity, or synthesis, follow its lead:
Show how to think, not just what to outputâuse examples that model reasoning, not just formatting.
Allow for inference, but require models to explain their logic clearly.
Prepare for edge casesâteach the model how to handle disagreement, vagueness, and conflicting evidence.
When your task calls for subtlety or smart tradeoffs, the Cognitive Scientist archetype is the one to build with.
đ I shared a prompt template built using the six building blocks of effective prompt design, then asked four top AI models to provide feedback. Each one had strong (and different) opinions.