AI Guides › Getting Started
The Onboarding Mistake: Treating Claude Like A Search Engine
By Nigel Guy · 2 min read
New users type one dense query, read the single answer that comes back, and
either accept it or give up and try a different phrasing — exactly the habit
built by two decades of search boxes. It feels natural, and it's the fastest
way to get mediocre results from something built for a completely different
kind of interaction.
The rule: a search engine gives you one shot at one query; a conversation
is meant to be worked, corrected, and built on — and treating it like the
former throws away most of what it's good at.
The mechanism: work it, don't just query it
- Send the rough version first, on purpose. You don't need the perfect
phrasing before you start — an imperfect first message you refine beats a
perfectly worded one you spent ten minutes drafting.
- Read the answer, then react to it specifically. "The second point is
wrong because X" or "shorter, and skip the caveats" does more than
rephrasing the original question from scratch.
- Let it ask you back. If it asks a clarifying question, answer it —
that's the conversation doing its job, not a sign you asked badly.
- Build across turns, not just within one. A good result is often the
fourth message in a thread, not the first — that's normal, not a failure
of the first attempt.
- Keep going in the same thread when the topic's the same. Starting a
fresh chat to "get a better answer" throws away the context that made the
previous answer better in the first place.
What to skip
Skip trying to write the one perfect prompt before sending anything — that's
search-engine thinking wearing a different outfit. Skip starting a new chat
every time an answer isn't quite right; correct the one you're in first.
And skip judging the tool by a single exchange — a search engine's one
result is the whole product; here, the one result is the opening move.
What this actually feels like once it clicks
Less like typing a query and reading a result. More like working with
someone who got the first draft roughly right and needs two rounds of notes
— because that's closer to what's actually happening underneath.
Guardrails
- This doesn't mean every answer needs several rounds — some questions
genuinely are one-shot. The point is not assuming that's the norm.
- Iterating well still requires you to actually read the answer critically,
not just ask again until something sounds plausible.
- If you find yourself doing five rounds to fix something a single clearer
first message would have solved, that's worth noticing too — iteration
isn't a substitute for asking clearly, it's a complement to it.
All 751 AI guides · JulieMango plans from £17/mo