Different tools for different jobs
What this is
In the next twenty to thirty minutes you’re going to run the same task through two different AI tools and pay attention to what’s different. You saw Eladia do this in the book: she almost published an article with fake regulation names because she’d done her research in a tool that wasn’t built for it, then tried Perplexity and got linked sources she could actually verify. Her routing rule came from experience, not from a guide.
By the end you’ll have a one or two sentence routing rule of your own — based on what you actually noticed, not what someone else told you to think.
What you’ll need
- 20–30 minutes
- A real task that requires both factual accuracy and a logical structure (a presentation to prepare, a briefing note for a new client in an unfamiliar industry, a piece of writing where structure matters and sources need to be real)
- Two AI tools — and the right pairing matters. To make the contrast visible, pair a generalist chat AI (Claude, ChatGPT, or Gemini) with a search-grounded AI (Perplexity is the cleanest example; Gemini with search grounding also works). These two categories of tool are built differently and the differences will show up immediately. Comparing two generalist chat AIs against each other often won’t surface useful contrast — both default to similar bulleted outlines with similar polished claims
If you only have one tool, you can still do a useful version: run the same task twice in the same tool — once with a one-line request, once with a structured brief like the one from Chapter 4. You’ll see a different kind of routing — not between tools, but between approaches.
Before you start
Caution: General-purpose AI tools will confidently invent sources, regulation names, citations, and statistics that look real and aren’t. This exercise asks you to compare how different tools handle sources, but at no point in this exercise should you treat a generated source as verified. Click through and check anything you’d actually use.
The exercise
Pick your task. Something where the quality of the output matters to you. Open both tools in different browser tabs.
Round 1 — generalist chat AI
Open your generalist tool first. Adapt this template to your task:
I need to prepare [a presentation / a briefing note / a piece of writing] on [topic] for [audience]. I know roughly what the topic is but I haven’t gone deep on it. Give me a concise outline, the specific sources I should read to ground it, and one counter-intuitive angle I could take that would make this more interesting than the obvious version.
Spend about ten minutes working with the response. Push back. Ask follow-ups. Treat it like a real working session, not a test.
Round 2 — search-grounded AI
Open the second tool. Start a fresh conversation. Use the exact same opening prompt. Spend another ten minutes working with the response.
You’re not looking for which tool is better in some general sense. You’re looking for differences in three places specifically: the outline (did the structure differ?), the sources (did one give you real linked references while the other gave you plausible-sounding names?), and the counter-intuitive angle (which one took a genuine risk, and which one defaulted to the conventional take?).
The routing-rule prompt
Now open whichever of the two tools you preferred overall. Paste this and fill in what you actually noticed — be honest, not diplomatic.
I just ran the same task through a generalist chat AI and a search-grounded AI. Two observations:
The most obvious difference in how they answered was: [your observation] I would trust Tool A more for [a specific kind of work], and Tool B more for [a specific kind of work].
Based on these observations, help me draft a one or two sentence personal rule for when I should use each. Don’t be generic. Base it strictly on what I just noticed. If my observations don’t yet support a clear rule, say so and tell me what to test next.
The point of this prompt isn’t to produce a routing rule you publish on LinkedIn. It’s to force you to articulate what you noticed. The act of writing the observations down is most of the value — the rule that comes back is a useful artefact, but the observations are the learning.
What you’ll notice
Three things usually happen if you run this on a real task with the right pairing.
The sources will differ visibly. The generalist will produce a confident list of authors, papers, and reports — some real, some misattributed, some invented. The search-grounded tool will produce a smaller list with links you can click. That difference alone is most of what Chapter 7 is teaching.
You’ll have a preference within ten minutes of Round 2. That preference is real data, but it’s not yet a routing rule. A preference says “I like this one better today”. A routing rule says “I use Tool X for tasks that look like Y”. The third prompt above is the bridge between the two.
Your rule won’t match Mario’s, or anyone else’s, and shouldn’t. Mario’s routing in the book is what works for him given his stack, his projects, and his money — “Claude for nearly all work, Perplexity for searching broadly, ChatGPT for images and second opinions”. Yours will be different. The point of the chapter is the method, not the answer.