Better Prompts, Sharper Answers: 13 Frameworks for Smarter Use of AI
- The single biggest cause of weak AI output is a weak prompt. The UK government’s 2025 AI Skills review found 42% of the UK workforce lacks AI literacy, and the most common gap is the inability to frame a clear question.
- Structure beats length. A 30-word prompt with a framework attached almost always beats a 300-word prompt without one. The framework gives the AI scaffolding to fill in.
- The best frameworks for prompting AI are the ones consultants, analysts and managers already use for thinking: MECE, SMART, Jobs to Be Done, 5 Whys, MoSCoW, SWOT, PESTLE, RACI, Balanced Scorecard, McKinsey 7S, Porter’s Five Forces, Ansoff Matrix and the Risk Impact Matrix.
- Each framework solves a specific problem. You do not need all 13. You need to know which one fits the task in front of you. This guide explains each in plain English with a real scenario.
- Every section shows the same thing in two versions. A coral box for the prompt without the framework, and a teal box for the prompt with the framework. Compare the two and you will see the lift the structure gives you.
- You can combine frameworks. SMART inside MoSCoW for prioritised goals. SWOT plus PESTLE for context and self-assessment. The frameworks were built to layer.
You open ChatGPT, Claude, Gemini or Copilot. You type a question. The answer comes back fast. It is fine. Not great, not terrible, just fine. You copy a bit, close the tab and get on with your day. By the afternoon you cannot remember what you asked. Most of us have had this conversation with an AI dozens of times.
The problem is rarely the model. The newest large language models are extraordinary. They write, summarise, plan and analyse at a standard that surprises people who have not used them lately. The problem is the prompt. A vague question gets a vague answer. The biggest single lift you can give an AI is to ask it a sharper question, and the easiest way to ask sharper questions is to borrow a framework you would already use to think with a colleague.
This guide walks through 13 of those frameworks. Each one has a single, narrow job. None of them is new. Consultants, managers and analysts have used them for decades. Apply them to the prompt and the output changes shape. The guide shows the same prompt twice for each framework: once without, once with. Read the comparison and the principle becomes obvious.
| Framework | One-line use form |
|---|---|
| MECE | Use MECE to split a topic into clear, non-overlapping sections that cover the full picture. |
| SMART | Use SMART to turn vague goals into specific, measurable, realistic, relevant and time-bound objectives. |
| Jobs to Be Done | Use Jobs to Be Done to understand what the audience wants to achieve in a real situation. |
| 5 Whys | Use 5 Whys to move from a surface problem to the real root cause. |
| MoSCoW | Use MoSCoW to prioritise what must, should, could and will not be included. |
| SWOT | Use SWOT to assess strengths, weaknesses, opportunities and threats in a balanced way. |
| PESTLE | Use PESTLE to analyse external political, economic, social, technological, legal and environmental factors. |
| RACI | Use RACI to clarify who is responsible, accountable, consulted and informed. |
| Balanced Scorecard | Use the Balanced Scorecard to review performance across financial, customer, process and learning areas. |
| McKinsey 7S | Use McKinsey 7S to check whether strategy, structure, systems, people, skills, style and values are aligned. |
| Porter’s Five Forces | Use Porter’s Five Forces to analyse market competition and pressure from buyers, suppliers, rivals, entrants and substitutes. |
| Ansoff Matrix | Use the Ansoff Matrix to compare growth options across existing and new products and markets. |
| Risk Impact Matrix | Use a Risk Impact Matrix to rank risks by likelihood, impact, priority and response. |
Use MECE to split a topic into clear, non-overlapping sections that cover the full picture. MECE stands for Mutually Exclusive, Collectively Exhaustive. It is a way of organising any topic so that every part of it gets covered exactly once.
Scenario: you are preparing a 15-minute board update on options for your charity’s next year of growth.
Typical output: a long, unstructured list of suggestions ranging from new fundraising campaigns to a redesigned logo to social media tips. Many overlap. Some sit at different levels (a high-level strategy next to a tactical post idea). The board would have to spend the meeting reorganising the list before deciding anything.
Typical output: a structured tree with categories such as Donor income, Earned income, Reach and audience, Operating efficiency, Partnerships. Each idea sits in only one category. The board can now compare like with like and pick a focus.
What changed: the MECE constraint forced the AI to think in categories before listing ideas, which is how decision-makers want to receive options.
Use SMART to turn vague goals into specific, measurable, realistic, relevant and time-bound objectives. SMART is the oldest goal-setting framework in business, and the most reliable.
Scenario: you are setting quarterly goals for a six-person sales operations team for July to September 2026.
Typical output: ‘Improve productivity. Communicate better with sales. Deliver on time. Develop skills.’ Generic platitudes you could have written yourself in 30 seconds. Nothing measurable, nothing time-bound, no way to know at the end of September whether anything was achieved.
Typical output: ‘Reduce average lead response time from 18 hours to 8 hours by 30 September 2026. Onboard 40 sales staff onto the new CRM dashboard by 15 August 2026. Cut weekly reporting time per rep from 2.5 hours to 1 hour by end of September.’ Each goal can be tracked. Each has a date.
What changed: the SMART criteria forced the AI to ask for, and use, the missing context (team size, period, budget) and to produce goals you can actually measure on the last day of the quarter.
Use Jobs to Be Done to understand what the audience wants to achieve in a real situation. The framework, made famous by Clayton Christensen, asks not ‘who is the customer?’ but ‘what job is the customer hiring this product to do?’.
Scenario: your charity is redesigning its donation page and you want to make it more useful for first-time donors.
Typical output: generic UX tips. Make the button bigger. Add testimonials. Reduce form fields. Useful but design-led, not user-led. Nothing about what the donor was trying to do when they arrived.
Typical output: three concrete user situations with the specific feelings and goals attached. The redesign brief writes itself: each job needs a different entry path on the page.
What changed: the prompt shifted from features to outcomes. The AI now thinks like a service designer rather than a graphic designer.
Use 5 Whys to move from a surface problem to the real root cause. Originated at Toyota, the 5 Whys keeps asking ‘why?’ until you reach a root cause rather than a symptom.
Scenario: customer churn at your subscription service has risen by 12% in the last quarter.
Typical output: ‘Improve onboarding. Run a win-back campaign. Survey lost customers. Review pricing.’ A familiar checklist of churn responses. None of them address the actual underlying cause because the prompt did not ask for it.
Typical output: ‘Why did churn rise? Because more customers cancelled in month 2. Why? Because they did not use the product in week 3. Why? Because the second-week email did not land. Why? Because we removed it during the redesign. Why? Because we assumed onboarding ended at week 1. Root cause: removal of week 3 nudge. Diagnostic test: re-enable the email for a cohort and compare 60-day retention.’
What changed: the AI was forced to trace a causal chain rather than throw tactics at the symptom. The diagnostic test at the end gives you something to act on.
Use MoSCoW to prioritise what must, should, could and will not be included. MoSCoW stands for Must have, Should have, Could have, Won’t have (this time). It forces a list to be prioritised rather than just sorted.
Scenario: you are planning the first release of a community app for a residents’ association with a tight three-month build.
Typical output: a wishlist of every feature anyone might want: events, messaging, voting, classifieds, alerts, maps, payments, photos, surveys. No priority. A team trying to deliver in three months would still have to argue every line.
Typical output: Must: events listing, broadcast alerts, member directory. Should: discussion threads, photo uploads. Could: classifieds, polls. Will not: payments and integrations, deferred to v2 because of compliance scope.
What changed: MoSCoW made the AI commit. The ‘will not’ list is the part that matters most: it puts the decisions back in the project’s pocket.
Use SWOT to assess strengths, weaknesses, opportunities and threats in a balanced way. SWOT is the oldest balanced assessment framework, still the fastest way to get an honest read on a situation.
Scenario: you run an independent coffee shop and are weighing whether to renew the lease for another five years.
Typical output: a generic list of pros and cons. ‘Pros: established location. Cons: rent might go up.’ The AI cannot say anything specific because you did not give it the structure to think.
Typical output: a focused SWOT with internal vs external clearly distinguished, and a final read such as ‘Strengths and customer loyalty likely outweigh the rent rise, provided you can grow lunch trade by 15% in the next 12 months. Without that, the renewal is marginal.’
What changed: the SWOT structure forced the AI to separate things you can change (S and W) from things you cannot (O and T), and to land on a decision.
Use PESTLE to analyse external political, economic, social, technological, legal and environmental factors. PESTLE is the standard scan of the outside world that any plan should pass through before being signed off.
Scenario: a UK SME is considering opening a small office in Lisbon in 2027.
Typical output: a tourism-style summary. ‘Portugal has a strong startup scene. Lisbon is popular with remote workers.’ Useful but not decision-quality. You would miss tax, visa, employment law and energy cost factors that change the case.
Typical output: structured factors covering UK-EU trade arrangements (Political), Portugal’s corporate tax and IFICI regime (Economic), local hiring norms and language (Social), broadband and EU AI Act applicability (Technological), employment law and visa routes (Legal), energy costs and carbon reporting requirements (Environmental). Each with a specific implication.
What changed: PESTLE made the AI cover the categories you would have forgotten on your own. It is the prompt equivalent of a checklist on a clipboard.
Use RACI to clarify who is responsible, accountable, consulted and informed. RACI is the simplest way to stop two people doing the same job and no one doing the other.
Scenario: you are project-managing the renovation of a community hall and the volunteer team keeps stepping on each other.
Typical output: a list of suggested job titles. ‘Project lead, fundraiser, comms volunteer.’ Pleasant but not actionable. The volunteers will still argue about who actually decides what.
Typical output: a clean matrix where every task has exactly one A, at least one R, and Cs and Is filled in. The committee can read it in two minutes and stop arguing.
What changed: RACI’s hard rule of one A per task forces a decision the prompt would otherwise dodge. The matrix removes the ambiguity that creates volunteer friction.
Use the Balanced Scorecard to review performance across financial, customer, process and learning areas. Developed by Kaplan and Norton, the Balanced Scorecard stops organisations from being dragged into pure financial reporting.
Scenario: you are preparing an annual review for the trustees of a medium-sized UK secondary school.
Typical output: a summary heavy on attainment and budget, light on everything else. Important things (staff turnover, parent satisfaction, pupil wellbeing trends) get a paragraph at the end. Trustees ask the same questions they always ask.
Typical output: a structured review the trustees can scan in five minutes, with the non-financial pillars given equal weight to the financial one.
What changed: the Balanced Scorecard kept the non-financial reality of the school visible. The board now asks broader questions.
Use McKinsey 7S to check whether strategy, structure, systems, people, skills, style and values are aligned. The 7S is an alignment scan: it asks whether the parts of an organisation are pointing in the same direction.
Scenario: two charities have merged six months ago and morale is unexpectedly low.
Typical output: a list of post-merger HR tips. Run an all-hands. Survey staff. Hold listening sessions. Helpful but generic, and unlikely to find the actual misalignment causing the problem.
Typical output: ‘Likely root causes: Style mismatch between the two former cultures has not been worked through (Shared values), and the combined structure has duplicate functions still doing parallel work (Structure). Diagnostic question: have any of the original org charts actually been retired?’
What changed: 7S looks at the whole system rather than just the human-resources layer. It exposes structural and cultural roots that an HR-only diagnosis misses.
Use Porter’s Five Forces to analyse market competition and pressure from buyers, suppliers, rivals, entrants and substitutes. Porter’s framework is the standard tool for working out whether a market is worth entering or staying in.
Scenario: you are weighing whether to open an independent gym in a market town of 25,000 people.
Typical output: a market-research preamble. ‘Demand for fitness is strong. Membership has grown. Consider your unique selling point.’ Reassuring noise rather than a clear read on the competitive structure.
Typical output: ‘Rivalry: high (3 incumbents in a small catchment). Buyer power: high (low switching cost). Supplier power: low (equipment is commodity). New entrants: medium (chains can roll out fast). Substitutes: high (home fitness apps). Verdict: structurally tough; the only attractive entry is a differentiated boutique offer the existing operators do not have.’
What changed: Five Forces moved the AI from cheerleading to structural analysis. The verdict is now grounded in the shape of the market, not a general ‘be unique’ line.
Use the Ansoff Matrix to compare growth options across existing and new products and markets. The Ansoff Matrix gives you four boxes: market penetration, market development, product development and diversification. It sorts growth ideas by how much risk they carry.
Scenario: you run a small organic bakery selling to local cafes and weighing where to grow next.
Typical output: a brainstorm of ideas of mixed riskiness, all listed at the same level. Open a second shop. Sell online. Add a sourdough class. Expand into hotels. No sense of which carry more risk than others.
Typical output: Market penetration: increase order frequency from existing cafes, win the cafes we lost last year (lowest risk). Market development: extend to nearby town cafes, sell to local hotels (moderate). Product development: add savoury items, launch sourdough kits to existing cafes (moderate). Diversification: open consumer-direct online shop, run paid baking workshops (highest).
What changed: the matrix gave the AI a way to compare ideas on the same axis (risk), which turns a brainstorm into a roadmap.
Use a Risk Impact Matrix to rank risks by likelihood, impact, priority and response. The Risk Impact Matrix sorts risks into a 2×2 (or 3×3) grid so that attention goes to the right places.
Scenario: you are organising a 1,000-person community festival in three months and the trustees want to see a risk register.
Typical output: a long flat list of every conceivable issue, from minor (wet ground) to existential (terror threat) presented at the same weight. The trustees cannot tell what matters.
Typical output: a ranked table where weather and crowd control sit as high-likelihood high-impact priority 1 risks with specific mitigations (marquee booking, steward ratios), and low-likelihood low-impact risks sit as priority 4 with ‘accept’ as the response. The trustees see at a glance where the attention should go.
What changed: the matrix turned a worry list into a managed register. Every risk has a score and a planned response, which is what governance actually needs to see.
The frameworks are most powerful when they layer. A few combinations worth knowing.
- SWOT plus PESTLE. Use PESTLE to gather the external context, then feed it into the Opportunities and Threats columns of a SWOT. Stops the SWOT from being lazy on the outside-world side.
- MoSCoW plus SMART. Prioritise with MoSCoW, then turn each Must and Should into a SMART objective. Now you have not just a priority list but a list you can manage by date.
- 5 Whys plus RACI. When the root cause is process failure (the 5 Whys ends at ‘no one was responsible’), build a RACI to plug the gap. The diagnostic feeds the fix.
- MECE plus Ansoff. Use MECE to make sure you are covering all growth options, then use Ansoff to score them by risk. Breadth, then judgement.
- Porter’s Five Forces plus Balanced Scorecard. Use Porter’s to understand the market you are in, then use the Balanced Scorecard to measure whether you are responding to it across all four pillars rather than just chasing financials.
There is one habit that lifts the quality of AI output more than any framework, and it is the habit of telling the model what success looks like before you ask the question. Industry research in 2026 consistently identifies this as the single biggest determinant of prompt quality. State the audience, the format, the length, the tone, the constraints. Then attach the framework. Then ask.
The frameworks in this guide work because they all do that job in a compressed way: they tell the model what shape the answer should take. The reason a SMART prompt produces a SMART answer is not that the model loves the acronym. It is that the acronym is a shorthand contract for what counts as a good answer. The same is true of MECE, of PESTLE, of RACI, of all 13. They are contracts the model can fulfil.
The next time you open an AI tool and the answer that comes back feels fine but forgettable, do not blame the model. Pick the framework that fits the job, attach it to the question, and try again. Most of the time, the answer that comes back the second time will not be forgettable at all.
