How to Use AI for SWOT Analysis

SWOT without evidence is a workshop game

SWOT is a familiar grid: strengths, weaknesses, opportunities, threats. Familiarity is why people dump the company name into ChatGPT and ask for a SWOT. The model will fill four boxes with fluent, generic claims. Some will be true of many firms. That is a workshop game. It is not a strategy. Use AI for SWOT only on evidence you provide: last year's results, customer comments you are allowed to share, a competitor pack you already collected, a risk log. Strengths and threats that the model invents are unowned. Unowned bullets get into decks. Decks get into board packs. Board packs become the story of the firm. Stories without sources waste the hour that should have gone to the missing fact.

The thinking-partner essay still applies. Use the model to challenge assumptions, not to supply the market. Ask what is missing from the pack. Ask which strength is only a slogan. Ask which threat is a duplicate of a weakness. Disagreement between the grid and your evidence is useful. Agreement between the grid and the model's training data is not. Training data is an average of other people's public narratives. Your strategy is supposed to be specific. Specificity comes from the pack. If you have no pack, you do not have a SWOT task. You have a research task. Do the research. Then sort.

This essay is a method for an evidence-bound SWOT and a short strategy memo that does not launder guesses as findings. You will see how to brief the grid, how to keep inferences labelled, and how to fact-check the tempting bullet that would make the slide look balanced. Balanced slides are a design preference. Strategy is not a design preference. An empty threats cell is more honest than four invented competitors. Honesty is the operating standard. Balance can wait until you have facts. If the pack is thin, the grid should look thin. That thinness is the finding.

Strengths and threats must come from the pack you brought

Treat each cell as a claim that needs a pointer. Strengths come from results, capabilities you can show, and customer evidence. Weaknesses come from the same, including complaints and failed projects. Opportunities are changes in the pack: a tender, a regulation date, a capacity you already funded. Threats are named competitors, named risks, named dependencies, with dates. If the pack is silent, the cell says not found. The model will try to be helpful and fill the silence with a trend. Trends without a source are decoration. Decoration in a strategy memo is how a guess becomes an investment story.

Competitive analysis belongs in the pack before the SWOT, not as a surprise the model retrieved from memory. If you need competitor facts, collect them in the allowed way, then paste. Do not ask ChatGPT what a rival charges if you have no source. It will produce a number. You will be tempted to keep the number because the grid looks empty without it. Empty is correct. A fake number is a fact-checking failure you chose. The fact-check essay is the neighbouring skill. Use it on every figure that would change a decision if it were wrong.

A SWOT prompt that refuses invention

Assemble the pack. Label each source. Brief the model as an analyst who may not add facts. Ask for a four-cell table with a pointer on every bullet and a separate list of inferences. Ask for empty cells where the pack is silent. Generate once. You then sit with the grid as a thinking partner: which inference will we own, which empty cell is a research task, which bullet is a slogan. Write the strategy memo from the owned bullets only. The memo should be shorter than the SWOT. SWOT is a sort. The memo is a decision. If the memo restates all sixteen bullets, you have not decided. You have formatted.

Run a pre-mortem on the memo, still on the pack: if this choice fails in a year, which threat in the evidence did we downplay. Do not let the model invent a new threat in that pass. Invention in a pre-mortem feels insightful and is still fiction. Insight is allowed as a question. Fiction is not allowed as a finding. Keep questions in a list. Keep findings on pointers. That split is the whole quality system for AI-assisted strategy notes. A question you cannot source is homework. A finding you cannot source is a problem you chose to keep.

Role: You are an analyst preparing a SWOT I will take to a strategy meeting.
Task: Fill a SWOT table using only the evidence pack I paste.
Context: I will name the decision the meeting must take and the date of the pack.
Constraints:
- Every bullet needs a source pointer. If none exists, write not found in that cell.
- Do not invent competitors, prices, market sizes, or trends.
- Label inferences separately from stated facts.
- Do not balance the grid by adding decorative bullets.
Output: SWOT table, inference list, research gaps, and a half-page memo from owned bullets only.
Quality checks: List bullets a sceptical director would refuse without a file.

If the quality list is long, you are not ready for the memo. You are ready for research or for a thinner grid. A thin honest SWOT beats a thick invented one. Thickness is not rigour. Pointers are rigour. Take the gaps to the meeting as work, not as shame. If the meeting wants sixteen bullets, give them the empty cells in writing. Empty cells are how you stop a guessing contest from calling itself strategy.

Strategy memos that quote the model's imagination

The failure mode is a board slide with four even columns and no footnotes. It looks like analysis. It is a mood board. People argue about wording instead of evidence. A second failure is mixing public model knowledge with internal numbers without labels, so nobody knows which sentence is confidential and which is a guess. Label the class. A third is using SWOT as a substitute for a decision. The grid becomes the deliverable. The meeting ends with we should be more innovative, which is not an action. Force the memo to name one choice and one owner. If the model cannot find a choice in the pack, say so.

Watch for threats that flatter. The model may invent a competitor that makes your story dramatic. Drama is not diligence. Watch for strengths that repeat the About page. If the only source is marketing copy, it is not a strength for this exercise. It is a claim that still needs evidence. Cut it. Cutting is the craft. Generation is cheap. Cutting is the job the meeting is paying you for. If a bullet only exists to fill a column, it is decoration. Decoration does not survive a sceptical director, and it should not survive you either.

Related reading on StudyGrid

Read next: AI as a Thinking Partner AI for Competitive Analysis How to Fact-Check ChatGPT. Those essays sit beside this one. Use them when you need the neighbouring skill, not as a substitute for the check you still have to make.

What to do this week

Take one real decision and a pack you already have. Run an evidence-only SWOT. Leave cells empty where the pack is silent. Write a half-page memo from owned bullets only. Fact-check every figure that would change the choice if it were wrong. Bring the empty cells to the meeting as listed research, not as failure. That is how to use AI for SWOT analysis without turning the grid into a game.