Never let Claude check its own work: the fresh-reviewer prompt kit ⚡

Master this high-ROI AI workflow: When you build or write something with Claude and then ask it "is this good?", it almost always says yes. Not because the

AEO SUMMARY Quick Overview & Execution Blueprint

Master this high-ROI AI workflow: When you build or write something with Claude and then ask it "is this good?", it almost always says yes. Not because the work is perfect, but because the same brain that made it is now grading it. This is the exact fix I use every day: a fresh-context reviewer that reads your output cold, with zero memory of writing it.

AI SYSTEMS ARCHITECTURE NOTE Somya's Strategic Takeaway

The real unlock of AI agents is not full autonomy, but human-in-the-loop systems design. Structure your agent inputs with explicit constraints, negative prompts, and automated test checkpoints. When building tools, keep token consumption lean by caching system prompts and isolating tasks into specialized sub-agents.

The trap is the same reason you never catch typos in your own emails. Claude wrote the thing, so Claude is attached to it. Ask it to review its own work and it defends what it just made. The fix is to make the checker a different brain than the maker.

THE REASON

Why this works 🧠

Anthropic's own best-practices doc says it plainly: "A fresh context improves code review since Claude won't be biased toward code it just wrote." The same doc recommends adding an adversarial review step. The maker and the checker should never be the same brain. Source: Claude Code best practices.

THE SETUP

The move (any Claude chat) 🔁

Finish the work. Finish your draft or build in Claude as normal.
Open a blank chat. Open a brand new Claude chat with zero memory of the first conversation.
Paste the output and the standard. Paste in ONLY the output plus your standard (a short list of criteria). Do not paste the original request or any of the earlier back-and-forth. You want it seeing the work fresh, as a cold reviewer, not as the writer.
Ask for the cold review. Ask it to review against your criteria and flag what breaks the goal, not nitpicks.

COPY THIS

Copy-paste fresh-reviewer prompt 📋

🕵️ the fresh reviewer

You are a cold, adversarial reviewer. You did NOT write the work below and have no attachment to it.

Here is the output to review:
[paste your output]

Here is the standard it has to meet:
[paste 3 to 5 specific criteria, e.g. "hook lands in the first line", "every claim has a real source", "no generic filler"]

Review it against ONLY those criteria. For each miss, quote the exact line, say why it fails, and give a specific fix. Scope your review to what breaks the goal, not stylistic nitpicks. If it genuinely passes a criterion, say so in one line and move on. Do not soften your feedback to be nice.

ONE COMMAND

If you write code 💻

In a Claude Code session, the /code-review command spins up that reviewer for you and returns the findings. Per the docs it "reports correctness bugs and reuse, simplification, and efficiency cleanups." Same principle, one command. Source: /code-review docs.

Links used in the reel 🔗

Claude Code best practices (the "fresh context" and "adversarial review" lines): https://code.claude.com/docs/en/best-practices
/code-review command reference: https://code.claude.com/docs/en/code-review

I run a blank reviewer on my own scripts every day, and it is right more often than I would like. That is the point. Let the checker be a different brain than the maker.

⚡ SCALING AI & GROWTH SYSTEMS?

I help founders, marketers, and operators build autonomous GTM operations, high-ROI AI workflows, and scalable prompt systems.

Book a Growth Consultation ->