Iro AIWhat you'll be able to do
- Get to a working understanding of a new topic in minutes, not hours
- Write research prompts that return sources, dates, and confidence levels
- Synthesize several sources into one clear, non-contradictory summary
- Spot a hallucinated citation before it lands in your report
- Use tools like Perplexity to get answers you can actually trace and cite
Inside the path
A focused set of five-minute lessons — each one ends with a hands-on exercise, not a quiz you can guess.
Scope the question first 4 min
Turn a vague topic into a sharp research question AI can actually answer well.
Prompt for sources, not just answers 6 min
Get cited claims with dates and confidence levels instead of a confident wall of text.
Synthesize multiple inputs 5 min
Combine several articles, docs, or search results into one coherent view — and surface where they disagree.
Catch the hallucinated citation 6 min
Learn the tells of a made-up source and the checks that catch it every time.
Verify before you cite 5 min
A repeatable check for load-bearing facts so nothing invented reaches your final work.
Try a sample exercise
This is the kind of card you'd practice inside Iro — you do the thinking, then get feedback.
◆ Sample exercise · Prompt practice
You're researching the European e-bike market for a go-to-market memo due tomorrow. You want AI to get you up to speed fast, but the memo will be read by your leadership team — a made-up statistic in it would be a disaster.
Your task: Choose the research prompt that gets you useful facts without inviting invented ones.
- "Tell me everything about the European e-bike market."
- "I'm researching the European e-bike market for a 2026 go-to-market memo. Give me the 5 most important facts about market size and growth. For each: cite the source and its date, rate your confidence high/medium/low, and flag anything you couldn't verify. If you don't have a reliable source for a number, say so instead of estimating."
- "What's the exact market size in euros and the top 5 companies by revenue in the European e-bike market?"
- "You're a world-class market analyst. Confidently summarize the European e-bike market for my leadership team."
See why the second prompt wins
The winning prompt builds verification into the request itself. It asks for a source and date on every claim, a confidence rating so you know which facts to double-check, and an explicit flag for anything unverified — and it gives the model permission to say "I don't know" instead of estimating. It's also scoped to a real deliverable, so the output stays focused. The other three do the opposite: "tell me everything" and "confidently summarize" reward fluent guessing, and demanding "exact" numbers pressures the model to fabricate precise-looking figures. In Iro you practice writing prompts like this and get feedback on whether they'd actually protect you from a hallucinated fact.
Why AI research goes wrong (and how to stop it)
AI models are trained to sound helpful and fluent, which means their failure mode isn't saying "I don't know" — it's confidently inventing a plausible answer, complete with an author, a year, and a link that doesn't exist. That's fine when you're brainstorming and fatal when you're citing.
The fix is to change what you ask for. Instead of requesting an answer, request an answer plus its evidence: the source, the date, a confidence level, and an honest flag on anything the model couldn't stand behind. When you make the evidence part of the deliverable, you can see at a glance which claims are solid and which need a human check — and you give the model an easy, honest exit instead of forcing it to guess.
Synthesis is the real superpower — verification is the price
The single biggest research win from AI isn't finding one fact — it's synthesis: pointing it at five sources and getting back one coherent summary that names where they agree and where they conflict. That collapses hours of reading into minutes.
- Hand off: the first-pass scan of a topic, combining multiple inputs, drafting a neutral summary, and surfacing disagreements between sources.
- Keep: the final judgment on which source to trust, and the verification of every fact that your conclusion actually rests on.
Live-search tools like Perplexity make this faster because they return real links you can open, but they don't remove the check — they just make it quicker. The rule holds either way: synthesize with AI, verify the load-bearing facts yourself.