Five frameworks, five acronyms, one idea: every slot is a question the AI would otherwise answer with its most likely guess. AI Central's guide, 5 Prompt Frameworks Pros Use, matches each framework to a kind of job, RT-CROS for work tasks, CARE when you have a model to copy, APE for a quick ask, PAS or FAB for copy that must sell and PTCF for Gemini in Google Workspace, then adds two habits that work on top of any of them.
Why a vague prompt gets a generic answer
The guide opens with a failure everyone has seen. Ask for a follow-up email for your webinar and the reply thanks attendees, hopes they found it valuable and invites questions. Nothing in it is wrong, and nothing in it could only have come from your webinar. AI Central's diagnosis fits in one line, and a framework is simply a way to close the gaps before the model has to guess.
AI fills every gap you leave with its most likely guess
Pick by the job, not by the acronym
The second page is a 30-second chooser, and it is the most useful page in the guide. Its rule is one job, one framework:
- RT-CROS for most work tasks: analysis, plans and documents.
- CARE when you have a model to copy, an email or report you liked.
- APE for a quick ask with one clear output.
- PAS or FAB for copy that must sell, such as ads and landing pages.
- PTCF for Gemini inside Google Workspace: Docs, Gmail and Sheets.
RT-CROS, six slots for real work
RT-CROS is the heavyweight: Role, Task, Context, Reasoning, Output and Stop. The guide's test for each slot is practical, since every slot answers a question a new colleague would ask you. Context is what the AI cannot know. Stop is when to stop and what not to invent, the slot most people never write and the one that keeps made-up figures out of the draft.
The worked example rebuilds the webinar email. A B2B email marketer role, a task with a target (at least 8% of attendees book a demo), context with real numbers (212 attended, 140 left early, last quarter's single email got 1%), a reasoning step that decides what each email must achieve before writing it, a table as the output, and a stop rule: three emails, no invented customer names or numbers. The difference from the one-line version is not length. Every decision the model used to guess is now made.
CARE and APE, the lighter frameworks
CARE (Context, Action, Result, Example) is for work that already has a model. The guide's example is a monthly update to 30 clients, written from your notes, with last month's update pasted in so the AI matches its tone and length. AI Central's verdict on that last slot is short:
The example does half the work
APE (Action, Purpose, Expectation) is the quick version, three slots for one clear output: summarize this email thread in 5 bullets, because I brief my manager in 2 minutes, decisions first and open questions last. The purpose line tells the model who reads the answer and how much time they have, which shapes the summary more than any adjective would.
PAS, FAB and PTCF for copy and for Gemini
For selling copy, the guide gives a simple fork. PAS (Problem, Agitate, Solution) when the reader has a pain, FAB (Features, Advantages, Benefits) when the product leads. Its one-line example asks for a LinkedIn ad with PAS: the problem is that reports take a full day, the solution is a template that cuts it to an hour.
PTCF (Persona, Task, Context, Format) is the framework for Gemini in Docs and Gmail, typed into the Help me write box. The example is an account manager drafting a reply that moves a kickoff to next week, as a short email of 3 paragraphs. AI Central notes that Google's own prompting guide uses these same 4 slots.
Make it check in first
The first pro tip is one paragraph you add to any framework: ask up to 3 questions about anything unclear, list the rules you will follow and a 3 to 5 step plan, say so if two rules conflict, and wait for an OK. The reason the guide gives is the thread of the whole document, because it finds the gap you missed before it fills it with a guess. AI Central calls it ten seconds that save the rewrite.
Drafting habits that compound
- Ask for 3 versions with different angles, because one answer gives you nothing to compare.
- Pick one and say why. The reason teaches the next draft.
- Refine one thing at a time: shorter, then warmer, then a clearer ask.
- Save the prompt that worked, and next time change only the context.
Where to start
Do you need all five? No. Start with RT-CROS on one real task this week, since it covers most work, add the check-in paragraph, and save the prompt once it works. The guide closes on the same three moves: open ChatGPT or Claude, fill one framework, save your best prompt.