13 task categories. For each one: the right model, exactly what context you need to supply, what output to request, how to QA it, and what to do when the output is garbage.
Generic output caused by weak context or vague success criteria.
Add audience, offer, examples, constraints, evidence, and the exact desired format before running.
Generic output — sounds like every other DTC brand. Words like "crafted," "premium," "unleash," and "discover" everywhere.
Name a competitor URL and say "write copy that is deliberately different in tone." Add banned words and the exact voice reference.
Output is generic creative direction with no brand specificity. Could be for any business.
Add your exact hex codes, a reference image URL, the channel, and one "this is what we never do" rule. Specificity breaks genericness.
Output describes what the data shows but not what to do. All observation, no decision.
End your prompt with: "What is the single most important action to take, and by when?" Forces a decision, not a description.
Output is vague because the goal was vague. Garbage in, garbage out applies more here than anywhere.
Start over with: "My goal is [X]. I want the output to be [format]. Here's an example: [Y]." Three sentences. Runs every time.
Output looks technically fine but could belong to any brand. No visual identity carries through.
Add your exact brand hex colors to the prompt. Add "consistent with [reference URL]." Add one negative prompt: "watermark, text overlay." Three changes, dramatically more on-brand.
Campaign plan reads like a textbook. "Build awareness, nurture leads, convert." Useless for execution.
Add: "My current cost-per-lead is $X. My goal is $Y. My top-performing ad shows [description]. Beat it." Now the model has a target and a reference. Output gets specific.
SOP is written for a generic business with generic tools. Doesn't match how your team actually works.
Name the actual tools: "We use Notion, Slack, and Stripe. Not Asana, not HubSpot." Add "Write the SOP as if you're handing it to a new hire who has never done this job." Forces operational specificity.
Prompt works once, fails on the second input. Built around the example, not the objective.
Test with the edge case first: the hardest, most unusual input you'll ever feed it. If it breaks, fix the prompt for that case. Don't fix for the easy case and assume it generalizes.
Article is technically on-topic but says nothing new. No original angle. Will not rank above what already exists.
Add your stance: "Our take on this topic is [X] — which disagrees with the common advice to [Y] because [Z]." Give the model an argument to make, not just a topic to cover.
Diagnosis is a list of everything that could be improved. No priority order. Useless for a founder who has 4 hours.
End the prompt with: "If I can only fix one thing before next week's campaign launch, what is it and what exactly should it say?" Forces a decision, not a wishlist.
Generated video looks nothing like the concept. Too abstract. No direction on motion or camera.
Describe the camera move first: "Slow zoom in on [subject] from wide to medium. No cuts." Then describe the subject. Camera direction before subject description consistently produces tighter results.
Ready to run these workflows?
Open the Vault — 1,096 prompts free
Social Media
Browse prompts →What to Provide
What to Ask For
Quality Checks
Posts sound like a corporate newsletter. No personality. No hook that stops the scroll.
Paste in your best-performing post and say "match this voice exactly." Then add: "The first word cannot be 'I,' 'We,' or 'Our.'" Watch the hooks improve immediately.