49 of 50 were legible in the screenshot. None of them was a new technique.
Not because they are bad prompts. Most are perfectly good. But a prompt that names a job is not the same thing as a technique that changes how the model reasons, and almost all of these name jobs. Here is where every one of them lands, and why.
Source: a Facebook graphic headed "Steal these 50 Claude AI prompts", seen 2026-10-20, screened 2026-10-21.
The same four questions this site publishes, run in the same order. Most rows are settled by the first and the fourth.
Question one asks you to state the mechanism without using its name. A row that can only be described by repeating its own label has nothing underneath the label yet. Question four asks whether a thing changes the thinking or the output — wanting an answer in a particular shape is a formatting request, wanting the model to reason differently is a technique.
These describe behaviours the Builder already has a mode for. In every case the existing mode is the narrower of the two, which is usually what makes it work.
Explain simply, intermediate and expert level.
Already a mode. Three depths instead of one is a choice about how much is said, not about how the answer is worked out — which is exactly how the taxonomy classified it: presentation, not reasoning.
Covered by ELI10
Break problems into fundamentals.
Already a mode, under the same name. This is the clearest match on the sheet.
Covered by FIRST PRINCIPLES
Identify hidden assumptions and risks.
Already a mode, and the catalogue splits it in two: surfacing the assumptions is one job, attacking the plan is another. This row asks for both at once.
Covered by ASSUMPTION CHECK
Rank tasks and create a schedule.
The ranking half is already a mode — work out which few items carry most of the result, and say what that leaves out. The schedule half is a deliverable you can ask for on top.
Covered by IMPACT PRIORITIZATION
Extract key points and takeaways.
Already a mode, and the version in the catalogue is the stricter one: cut the length without losing anything load-bearing, rather than cut the length.
Covered by TIGHTEN
Define your brand voice and style.
Already a mode. The catalogue version matches a voice from samples you give it; this row asks it to write the style guide first. Same behaviour, one extra artifact.
Covered by VOICE MATCH
Get honest responses to objections.
Already a mode, and the catalogue version does the harder half: it finds the objections you have not heard yet, rather than answering the ones you already know about.
Covered by BUYER OBJECTION LENS
The taxonomy went through a long list of candidate techniques and split them into ones that change reasoning and ones that are something else — a persona, a deliverable, a framework. Each of these rows matches something on that list, and the reason it was set aside is quoted as it was written.
Explain key concepts with examples.
Asking for an expert is asking for a voice, not a method. The taxonomy sorted the role prompts — CEO, CFO, product manager — into a persona library on purpose, because none of them says what the model should do differently.
Sorted in the taxonomy as CEO — “Persona library”
Compare options and get a recommendation.
A comparison with a recommendation at the end is a deliverable. The reasoning move nearby — forcing the real trade-offs out rather than producing a flawless-looking pick — is a known gap here, and it is not what this row asks for.
Sorted in the taxonomy as DECISION MATRIX — “Analytical artifact, not a mode”
Turn processes into step-by-step SOPs.
A standard operating procedure is a document. Asking for one tells the model what to hand back, not how to think about the process first.
Sorted in the taxonomy as ACTION PLAN — “Deliverable type”
Convert meeting notes into action items.
Same deliverable as the row above, from a different input. Worth knowing: a rewrite like this is where a model will quietly add detail the notes never contained, which is a live open question here rather than a settled one.
Sorted in the taxonomy as ACTION PLAN — “Deliverable type”
Analyze your week and plan next steps.
Looking back at finished work to learn from it is classified, and classified as narrow — it only applies once something is over. That is a point in its favour as a technique and the reason it is not in the core set.
Sorted in the taxonomy as POSTMORTEM — “Useful after completion, narrower use”
Get reliable, sourced information.
Checking where a claim came from is classified and rated highly, and it is not built here yet. This row asks for the result of that habit rather than the habit itself.
Sorted in the taxonomy as SOURCE CHECK — “High practical importance”
Compare options with a weighted matrix.
Named identically to a row that was sorted out of the library years before this post, for the reason quoted. A weighted table is a thing you produce, not a way of thinking.
Sorted in the taxonomy as DECISION MATRIX — “Analytical artifact, not a mode”
Develop creative concepts.
A job title, like row 1. The taxonomy keeps these somewhere — they are useful — but it keeps them apart from the techniques, because a title does not describe a method.
Sorted in the taxonomy as DEVELOPMENTAL EDITOR — “Persona library”
Two of the fifty have been through this process as candidates in their own right. Both were examined twice and neither survived.
Critique and improve your content.
Critique then rewrite was proposed here as its own technique and did not survive a second look. The critique step is a mode; bolting the rewrite onto it describes an order of work rather than a different way of working.
Screened as a candidate in its own right. Recorded decision: rejected.
Review and improve Claude’s output.
The same proposal as row 39 with a stronger name, and it was already decided. Reviewing your own output is a real and valuable habit; it is the step the catalogue already has, not a new one.
Screened as a candidate in its own right. Recorded decision: rejected.
This is the bulk of the sheet, and the argument is the same for all of them, so it is made once here rather than repeated twenty-eight times. Every row below names something to produce. Try question one on any of them — state the mechanism without using its name — and the answer is "do the task". That is not a criticism: these are useful things to ask for, and a good prompt for any of them beats a bad one. It is a category difference. A job tells the model what to hand back. A technique tells it how to work the problem before it hands anything back.
| # | Prompt | What it produces |
|---|---|---|
| 6 | Viral Hook Generator | A list of hooks |
| 7 | X Post Writer | A post |
| 8 | LinkedIn Thought Leader | A post |
| 9 | Content Repurposing | The same content, reformatted |
| 10 | Storytelling Framework | A story |
| 11 | Personal Brand Strategy | A plan |
| 12 | Competitor Analysis | An analysis |
| 13 | Landing Page Copy | Copy |
| 14 | Offer Builder | An offer |
| 15 | 30-Day Content Calendar | A calendar |
| 16 | Business Idea Validator | An evaluation |
| 17 | Find a Profitable Niche | A list of niches |
| 18 | Business Process Audit | An audit |
| 19 | Customer Persona | Personas |
| 20 | Growth Strategy | A plan |
| 22 | Automate Repetitive Work | A list of candidates |
| 26 | Build a Website | A website |
| 27 | Debug My Code | A fix |
| 28 | Code Review | A review |
| 29 | API Integration | Integration code |
| 30 | Test Case Generator | Test cases |
| 31 | Analyze a Dataset | An analysis |
| 34 | Market Opportunity Analysis | An assessment |
| 37 | Visual Design Brief | A brief |
| 38 | Video Script | A script |
| 41 | Personalized Outreach | Messages |
| 43 | Proposal Writer | A proposal |
| 45 | Customer Feedback Analyzer | An analysis |
These four are neither jobs nor techniques. They are about making work repeatable — turning something you did once into something you can run again. That is a real and growing category, and nothing in the catalogue addresses it, which is correct: it is not a way of reasoning. It is closer to what a prompt builder is for.
Row 44. The post’s own caption sits over this row in the screenshot. It was not read, so it is not judged.
It is counted in the fifty and excluded from every verdict above. Filling it in with something plausible would have been the exact mistake this project is set up to catch.
It does not show the sheet is worthless. Several of these prompts are the fastest way to get a decent first draft of something, and the five that match a mode match it because whoever wrote them had the same instinct the catalogue was built on.
What it shows is narrower and more useful: a collection assembled for sharing optimises for breadth, and breadth is the enemy of the third question — say when you would not use it. Fifty prompts that each apply to one task are fifty jobs. The techniques worth learning are the handful that change how any task gets worked, and there are far fewer of them than fifty.
Every verdict above is checked against the catalogue on each deploy. If a mode named here is renamed or removed, this page stops building rather than going on claiming coverage that moved.
The four questions used on this page, with the follow-ups that make them work and all four run on a candidate that did not survive them.