Verify in under two minutes
For most everyday tasks, run a quick verification pass that catches the things that actually matter — without turning a five-minute task into a fifty-minute one.
Intermediate training
Getting Reliable Outputs
Turn one-shot prompting into a reliable workflow. Verification, iteration, and lightweight critique — the difference between an answer that looks right and an answer that's ready to ship.
Start here
What you'll be able to do after this lesson
Improve a weak answer in two or three rounds
Name what's wrong, ask Claude to fix exactly that, and converge fast. Don't blow up the chat and start over for every flaw.
Know when to stop iterating
Recognize when revisions stop adding value — and when starting clean is faster than the next round.
Three reliability moves
Iteration, verification, self-critique
Iteration
Cheap revisions inside the same chat. The first answer is rarely the right one — but it's a starting point. Iteration is asking Claude to fix the specific thing instead of starting over.
Verification
Cross-check against the source. The five things that always need verification — names, dates, numbers, regulated claims, and any confident statement without evidence — are the bones of every reliability pass.
Self-critique
Ask Claude to find what's wrong with its own answer before you do. The model is better at critique than at first-pass writing — use that asymmetry.
Trainer note: The second draft is where Claude becomes useful for real work. The first
draft is a starting point. Employees who ship round one ship slop. Employees who run a quick second pass
ship work.
Where reliability comes from
Three places — and only one of them is Claude
The prompt
A clear brief is the cheapest reliability investment you can make. Most "Claude is unreliable" complaints are "I sent a thin brief" complaints in disguise.
Brief qualityThe source
The attachment, the labeled excerpt, the linked file. A good answer rests on a good source. Garbage in, fluent garbage out.
Source qualityThe loop
The verification and critique steps you run before sending. The loop is where unreliable output gets caught — and where reliable output proves it's reliable.
Your disciplineA reliable loop
Five rules for the second draft
Ask for the draft and the verification checklist together
"Draft X. Then list the five things I should verify before sending." The verification list is free, comes from the model that wrote the draft, and reliably surfaces the riskiest claims.
Ask Claude to critique its own draft before you do
"What's wrong with this draft on tone, specificity, and accuracy?" Claude will find real issues — sometimes ones you would have missed.
Compare two approaches on the same task
"Draft this two ways: one short and direct, one warmer and longer. Tell me which you'd ship and why." Two-option prompting is a forcing function for explicit trade-offs.
Re-run in a fresh chat to check stability
If the answer changes wildly between runs, the brief is too thin. If it stays consistent across two fresh chats, you've got something solid.
Stop iterating when revisions stop improving
Two or three rounds is usually plenty. By round four, you're either polishing diminishing returns or chasing a problem that needed a different brief.
Prompt upgrade
Weak prompt
Make it better.
Loop-ready prompt
Work-ready prompt
The opening line of your last draft — "I hope this email finds you well" — is an AI tell. Rewrite the first sentence only. Give me two options: one that opens with a specific detail from the closing, and one that opens with a direct question. Keep everything else.
Patterns that compound
Four reliability moves worth memorizing
Self-critique-then-revise
"Critique the last draft on tone, specificity, and accuracy. Then rewrite once, applying every critique." The fastest path from round one to ship-ready.
Two-options-then-pick
"Give me two ways to do this. For each: a pros list, a cons list, and your recommendation." Removes false dichotomies and surfaces hidden trade-offs.
Draft-with-citations
"For every fact-bearing sentence, cite the page or section of the attached source. Use the format [page X]." Verification turns into a five-minute spot-check.
Side-by-side comparison
"Here's the current version. Here's the rewrite. Show me a numbered list of every meaningful change with a one-line reason for each." Makes the diff visible.
Verify on every output
Five things that always need a real check
Employee rule: A reliability loop that takes two minutes for a five-minute task is the
right loop. A loop that takes twenty minutes for a five-minute task is a sign you should have used a
different tool. Match the loop to the stakes.
Drill the loop
Six exercises that build the habit
The goal is muscle memory. Run the self-critique pattern enough times that you don't think about it — you just do it.
- Open the iteration loop in this lesson. Click through all three rounds and read the verdict on each. Note what changed between rounds two and three — that's the move you most need to internalize.
- Run the self-critique pattern on three real tasks this week. Track total time and how much improved between rounds.
- Ask Claude for two options on a real decision you're carrying. Pick one and write down why — that "why" is the work you'd skip otherwise.
- Verify a Claude answer against the source document end-to-end. Note which kinds of claims needed the most checking.
- Time a one-shot prompt versus an iterated prompt on the same task. The point isn't to prove iteration is faster — it's to feel where the trade-off actually lives.
- Ask Claude where it might be wrong before you ask anything else. Make this the last move you run on any high-stakes draft.
Completion standard
You've finished this module when you have a default reliability loop — two or three moves you run on every shippable output — and can name, for any given task, the right verification depth in under thirty seconds.