Practice

Try the three methods on your own work

Three of the course's form-shaped methods, built as fill-in forms. Type on the left, the prompt assembles on the right, copy it into any AI tool. Everything else in the course is prompt-and-go: open the class page and copy the prompt.

Build a GOLD block

From Class 4, Prompt Design for LLMs (Sydney Seifert-Gram)

GOLD is the four-part shape for any request to a language model: Goal, Output, Limitations, Data. Write each part, and the block assembles on the right. The closing question at the end is part of the method: it makes the model confirm the brief before it starts.

GoalWhat you want done, in one or two sentences. The job, not the format.OutputFormat, length, persona, style. What the result should look like when it lands.LimitationsWhat to avoid or emphasize. Sydney's rule: tell it to stop before the next step until you say go.DataPaste or describe the source material it should work from.

Your prompt, assembled as you type

Fill the fields on the left.

You will know it worked whenThe first reply is a short restatement of your brief or a clarifying question, not the finished work. If it jumps straight to output, your Limitations line did not land.

Build a campaign system block

From Class 14, Copy Craft and AI (Antonio Fragoso)

Antonio's point was that AI does not need a better prompt, it needs better context. This assembles the eight-part system block he taught: concept, audience, brand voice with rationale, territories, human tension, emotion, guardrails, and the output you want. Paste the result at the top of a fresh chat and hold every piece of copy against it.

ConceptThe campaign idea in one sentence.AudienceWho it is for, specifically.Brand voiceThree to five adjectives, each with a one-line reason. The rationale is the part most people skip.TerritoriesWhere the work lives: channels, moments, formats.Human tensionThe true, slightly uncomfortable thing the idea resolves.EmotionThe single feeling the work should leave behind.Guardrails, alwaysWhat every output must do.Guardrails, neverWhat no output may do.OutputWhat you want it to produce now.

Your prompt, assembled as you type

Fill the fields on the left.

You will know it worked whenEvery line it gives you can be traced back to the concept and passes the guardrails without you editing. If a line feels generic, the weak field is usually brand voice without rationale, or a human tension that is really just a benefit.

X-ray a brief

From Class 20, Brief Mining with an End-to-End Creative Collaborator (Angella Tapé)

Stage one of Angella's five-stage brief-mining method. You hand the model the brief and ask it to find the tension the brief is too polite to say, then return spark ideas in a table you can actually compare. Paste the brief, set how many ideas you want, and add any constraint the brief already names.

The briefPaste the whole thing. Do not summarize it first; the point is to let the model find what you skimmed past.How many spark ideasAngella's range was five to seven. Fewer than five and you are guessing, more than seven and you stop reading.Constraints the brief already setsAnything it bans or requires, so the ideas do not break the rules on page nine.

Your prompt, assembled as you type

Fill the fields on the left.

You will know it worked whenThe named tension is something the brief implies but never states, and at least one spark idea makes you uncomfortable. If every idea could have come from the brief's own summary paragraph, ask it again with 'more honest, less safe.'

Pratt AI Design Certificate · Summer 2026