AI bootcamp for EU affairs, day 3: how AI reasons, and which model to pick
On day 2 we saw that language models predict the next word. That was only part of the story. Newer models think before they answer, and choosing the right model and the right amount of thinking is now a skill in itself.
Sebastián Rodríguez Pérez is a European campaign strategist with direct experience running and advising campaigns for some of Europe's largest pro-European organisations. He founded the European Campaign Playbook to codify practitioner knowledge across the EU's 27 political environments and make it accessible to campaign professionals, advocacy teams, and civic organisations working at the European level. His work spans electoral strategy, EU public affairs, digital campaigning, and the integration of AI tools into political communication workflows.
Disclosure
Sebastián Rodríguez Pérez is the founder of the European Campaign Playbook and Campaign Intelligence Library. He has worked with pro-European organisations across multiple EU member states. He maintains editorial non-partisanship in all published content; where articles relate to campaigns or organisations he has directly worked with, this is disclosed within the article. He holds no current positions in active electoral campaigns.
What it covers
How reasoning models think before they answer, how to choose between fast and deep models, what the effort setting does, and a simple rule for matching model and effort to the stakes of the task.
Who it is for
Anyone in a small public affairs, policy or communications team who has noticed their AI tool offers several models and a thinking setting, and wants to know which to use when.
When to use it
Whenever you are about to hand an AI tool something that matters, and before your team settles into using one model for everything.
Key takeaway
Model is which kind of colleague you ask. Effort is how long you let them think. Match both to the stakes of the task, and keep checking sources: more thinking improves complex work but never guarantees the answer is true.
In brief
- Day 2 covered how language models predict the next word. Reasoning models add a step: they work through the problem before answering.
- Underneath, it is still prediction. The model predicts a chain of reasoning, then predicts the answer from it.
- Two choices now matter: which model you ask, and how much effort you let it spend thinking.
- Claude’s line-up maps onto four colleagues: Haiku for speed, Sonnet for everyday work, Opus for serious judgement, Fable for long, complex, multi-step projects.
- Higher effort improves complex work, costs more and takes longer. It never guarantees truth, so the rule stays: high effort plus source checking.
The part of the story we left out
On day 2, we saw how large language models generate language: they predict the most likely next word, one word at a time. We didn’t tell you the full story. Here is how they reason.
The newest models work through a problem before they reply. They break it into steps, try different approaches, check their own working, and only then write the answer you see. The industry calls them reasoning or thinking models.
Under the bonnet it is still prediction. The model predicts a chain of reasoning first, then predicts the answer from that reasoning. Think of it as prediction with scratch paper.
That changes how you use these tools, because two choices now matter: which model you ask, and how long you let it think.
Everything below turns that idea into a practical guide, with Claude’s line-up as the worked example. The same logic applies in ChatGPT, Gemini and Le Chat.