Gather your own evidence first: a year of figures, recent reviews, staff views and a few competitors' websites. Paste it into ChatGPT, Claude or Gemini and ask it to sort the evidence into strengths, weaknesses, opportunities and threats, citing what supports each point. Research the outside world with sources, challenge the draft, then turn it into actions.
Done this way, a SWOT takes three to four hours spread over a week, and it tells you something you didn't already know. Done the quick way, typing "do a SWOT for a podiatry clinic", you get a list that would fit any clinic anywhere, and it isn't worth the paper. The six stages below are designed to stop that happening.
Which AI tool to use at each stage
You don't need a special SWOT tool. The general assistants do this well, but they suit different stages.
- The internal pass works in any chat assistant that accepts a long document: ChatGPT, Claude or Gemini. If your evidence pack includes anything commercially sensitive, use a business plan, which doesn't train on your content by default, or check the model-training switch in your privacy settings.
- The external scan needs a research mode that searches the web and cites sources. These take several minutes and produce a long report. Read the sources, not just the summary.
- The stress test works best in a different assistant from the one that wrote the draft. A model asked to criticise its own earlier answer tends to defend it; a second model has no stake in it. If you have ChatGPT, paste the draft into the free tier of Claude or Gemini for this step, with anything confidential removed.
Stage 1: Build the evidence pack (60-90 minutes)
AI can only be as specific as what you give it. Collect these into one document, numbering each item so the AI can cite it:
- Figures for the last 12 months: revenue by service, new versus returning customers, average spend, cancellations and no-shows, and anything that changed sharply.
- 30 to 50 recent reviews, good and bad, copied as text. Remove customer names.
- Complaints and compliments from the last year, summarised in a line each.
- Staff views: ask everyone three questions. What do customers praise? What do they complain about? What would you fix first?
- Where customers come from: referrals, search, word of mouth, repeat.
- Three to five competitors: their websites' service pages and prices, copied as text, with the date you copied them.
If you don't have a year of clean figures, three months exported from your booking or till system is enough to start; say so in the pack, so the AI doesn't treat a quiet quarter as a trend. Keep personal data out. Totals and anonymised comments are all the AI needs. If you're unsure which details count, the tutorial on giving AI your business context shows what to include and what to leave out.
A business that hasn't opened yet has no figures at all, and the stages still work with a different pack. A pottery studio planning to open with evening classes might number these instead: the prices and waiting lists of the three nearest studios (one beginners' course was full for eight weeks), notes from 25 conversations at a craft fair, the costings for the kiln, wheels and rent, and the owner's own teaching experience. Strengths then describe capabilities rather than a track record ("can run beginners' courses capped at six people"), so add one rule to the Stage 2 prompt: "Mark any strength not yet proven with paying customers as UNTESTED." Those untested strengths become the first things the business measures once it opens.
A pack doesn't need to be elegant. Here are the first lines of the one used by the podiatry clinic in the worked example further down (figures illustrative):
EVIDENCE PACK - podiatry clinic - Oct 2025 to Sep 2026
E1 Revenue by service: routine nail care 41%, orthotics 17%,
nail surgery 19%, biomechanical assessment 12%, other 11%
E2 Prices: 30-min routine appointment $55; 60-min biomechanical
assessment $70
E3 Nail surgery: 96 procedures, 58 of them referred by a doctor
or physiotherapist
E4 No-shows: routine nail care 14% of bookings; all other services 4%
E6 Orthotic lab: price per pair up at both of the last two renewals
E14 40 reviews from the booking site and search listing, full text,
names removed (copied 2 Sep 2026)
E18 Receptionist: "During the morning clinic the phone just rings.
I'd guess we miss eight or ten calls."Two habits in that excerpt pay off later. Every figure has its period attached, so the AI can't mistake a bad month for a bad year. And the staff view is quoted in the receptionist's own words rather than tidied into "phone handling could improve", which is the kind of soft phrasing that lets a weakness slip out of the final SWOT.
Stage 2: The internal pass (20 minutes)
Strengths and weaknesses are internal: things about your business that you control. Ask for these first, from your evidence only.
Below is an evidence pack for my business, a [type of business] with
[number] staff. Each item is numbered E1, E2 and so on.
Using ONLY this evidence, list up to 5 strengths and up to 5 weaknesses.
Rules:
- Strengths and weaknesses must be internal: things we control.
- After each point, cite the evidence numbers that support it.
- A point with no supporting evidence goes in a separate list called
"Unsupported", not in the SWOT.
- Be specific. "Good customer service" is not acceptable; say what
customers praise, how often, and in which evidence.
- Include at least one weakness I might not want to hear.
[paste evidence pack]The "Unsupported" list is the useful trick here. It catches the flattering generalities AI likes to add, and it also shows you where your evidence is thin.
Read the output against the evidence yourself. Pick two points at random and look up the cited items: do they really say that? If a strength rests on one enthusiastic review, it's an anecdote, not a strength. If the unsupported list is long, don't argue with the AI; go and find the missing evidence, or accept that you don't know. "We think our prices are competitive" is a common one. Either you compare them properly in Stage 3, or it stays off the SWOT.
To see what that check looks like, take a three-chair barbershop that ran the prompt on a pack of 45 reviews, a year of till totals and answers from its three barbers. Part of the reply read (illustrative):
STRENGTHS
1. Weekday walk-ins are served quickly: 31 of 45 reviews mention
"no wait" or "got straight in" (E12).
2. Loyal customer base (E3, E7).
WEAKNESSES
1. Saturday waits of 40+ minutes, raised in 9 reviews (E12) and by
two barbers (E7).
UNSUPPORTED
- Competitive pricing (no competitor prices in the pack)Strength 1 and the weakness hold up; the owner found the reviews. Strength 2 doesn't. E3 is revenue by service and E7 is the barbers' answers, and neither measures loyalty. The follow-up that fixed it was one line: "For strength 2, quote the exact line from E3 and E7 that shows loyalty. If there isn't one, move it to Unsupported." It moved. The owner then pulled the share of card payments from returning customers out of the till system, which turned a vague claim into something checkable.
Stage 3: The external scan, with sources (45 minutes)
Opportunities and threats are external: changes in customers, competitors, suppliers, rules and technology that you don't control. This is where a research mode helps, such as ChatGPT's deep research, Gemini Deep Research or Perplexity, because they search the web and cite what they found.
I run a [type of business] serving [who your customers are] in a
[town / city / rural] area. Research external factors likely to affect
a business like mine over the next 2 years.
Give up to 6 opportunities and up to 6 threats. For each:
- one sentence describing it
- the source (link), and the date of the source
- how confident you are, and why
Only include items with a source published in the last 2 years.
Do not include anything about my own business's strengths or weaknesses.
Also, from the competitor pages below, list any service, price or
offer they have that I don't. Quote the page text you relied on.
[paste competitor pages]Then check it. Open every link. Delete anything whose source doesn't say what the AI claims, anything older than two years, and anything that applies to the whole economy rather than to businesses like yours. Expect to delete a third of the list; that's normal. Where two sources disagree, keep the item but note the disagreement, and give it less weight when you choose actions. A threat you can only find in one opinion piece is a question to watch, not a fact to plan around.
The deletions are rarely obvious lies. More often a real source has been stretched. Consider a four-person garden-maintenance firm whose research report listed "rising labour costs squeezing margins across landscaping firms" as a threat, linked to a genuine trade-body survey. Opening the survey showed it was of contractors with 50 or more staff bidding for commercial grounds contracts, a market the firm never enters. The item was true for someone else. The fix was to rewrite it from the firm's own records instead: two of its four staff had asked for a pay review, and a local rival was advertising for crew at a higher hourly rate. That is a threat the owner could act on, and it came from evidence, not from the report.
For a comparison of the research tools, see Perplexity vs ChatGPT for business research, and for a fuller competitor process, competitor research with AI in one afternoon.
Stage 4: Stress-test the draft (20 minutes)
A SWOT you wrote about your own business will be kinder than one an outsider would write. Ask the AI to be the outsider:
Here is my draft SWOT. Act as a sceptical adviser who has seen many
businesses like this one.
1. Which "strengths" are really just what every competent competitor
also does? Move them out.
2. What weaknesses are probably missing, given the evidence?
3. Which threats are overstated, and which real ones are missing?
4. Is anything in the wrong box (an action listed as an opportunity,
or an internal issue listed as a threat)?
Explain each change in one line.
[paste SWOT]A bakery-café that sells a third of its output wholesale to three restaurants (E1) got this back from a second assistant (illustrative, trimmed):
MOVED OUT OF STRENGTHS
- "Fresh bread baked daily": all three competitor pages say the
same (E15, E16, E17).
MISSING WEAKNESS
- One restaurant is 38% of wholesale revenue (E2). Losing it would
cut total revenue by roughly an eighth.
WRONG BOX
- "Start a loyalty card" is listed as an opportunity. It is an action.
OVERSTATED THREAT
- "Rising flour prices": your supplier invoices (E9) show a 3% rise
over 12 months, well below the figure in the linked article.Three of the four changes were right, and the concentration risk was the most valuable line in the whole exercise: the owner had filed that restaurant under strengths as "a loyal wholesale partner". The flour point was wrong for a reason worth knowing. E9 covered only the white flour used in the café, while the rye and spelt flours in the wholesale loaves had risen faster and weren't in the pack. The model reasoned correctly from incomplete evidence. When the stress test downgrades something, check that your pack gave it the whole picture before you accept the change.
Then show the revised version to one member of staff who deals with customers every day, and ask what's wrong with it. The AI can spot a generic list; only your team can spot a false one.
Stage 5: Turn the SWOT into actions (30 minutes)
A SWOT that ends as four lists changes nothing. The standard way to turn it into actions is a TOWS matrix, which pairs the boxes:
| Pairing | The question |
|---|---|
| Strengths + Opportunities | Which strength lets us take which opportunity? |
| Weaknesses + Opportunities | Which weakness stops us taking an opportunity, and how do we fix it? |
| Strengths + Threats | Which strength protects us from which threat? |
| Weaknesses + Threats | Where does a weakness make a threat dangerous, and what do we do first? |
Using this SWOT, build a TOWS matrix with 2 actions in each of the four
pairings. Each action must name the specific strength, weakness,
opportunity or threat it uses. Then recommend the 3 actions with the
best payoff for the effort, for a business with [staff number] staff
and roughly [hours] a week of owner time to spare. Give a first step
for each that could be done within 2 weeks.
[paste final SWOT]For the barbershop above, one filled-in row of its matrix looked like this:
| Pairing | Action | First step (within 2 weeks) |
|---|---|---|
| Weakness + Opportunity: Saturday waits (W1) + a new office block opening two streets away (O2) | Move Saturday demand to a bookable weekday early slot aimed at office workers | Open 8-9am bookings on Tuesdays and Thursdays; hand cards to the first two weeks' Saturday walk-ins |
Notice what makes it usable: it names W1 and O2, so anyone reading it later can see why it exists, and the first step is small enough to start on Monday. An AI draft that says "improve capacity management" in that cell hasn't done the job yet; ask it to rewrite the action as something one person could begin this week.
Choose three actions, not eight. Give each an owner and a date.
When more than three candidates look good, a rough sum settles it faster than a debate. Take the podiatry clinic from the worked example below, and assume around 50 routine bookings a week (illustrative). A 14% no-show rate is 7 empty slots, or $385 a week at $55 each. Getting it down to 8% brings back 3 of them: about $165 a week, just under $8,000 over a 48-week year, for the cost of a reminder service and an afternoon setting up deposits. A gait-assessment evening might produce four $70 bookings. Both earned a place on the list, but the sum showed which to start first.
Stage 6: The one-page result
SWOT: [business name] Date: [date] Next review: [date + 6 months]
STRENGTHS (evidence) WEAKNESSES (evidence)
1. 1.
2. 2.
3. 3.
OPPORTUNITIES (source, date) THREATS (source, date)
1. 1.
2. 2.
3. 3.
THREE ACTIONS
1. [action] - uses [S2 + O1] - owner - first step - by [date]
2.
3.
UNSUPPORTED CLAIMS TO CHECK
- Worked example: a two-podiatrist clinic
Say a podiatry clinic with two podiatrists and a receptionist runs these stages. The clinic is hypothetical. The AI's first quick draft, without evidence, included "experienced team", "strong reputation" and "growing health awareness". None survived Stage 2.
The evidence-based version looked like this:
- Strengths: nail surgery done in-house, around a fifth of revenue and the main reason for medical referrals (E3, E9); reviews repeatedly mention explaining things clearly (E14, 22 of 40 reviews).
- Weaknesses: a 14% no-show rate on routine nail-care appointments (E4); the phone goes unanswered during clinic sessions, mentioned in 6 reviews (E14) and by the receptionist (E18); hour-long biomechanical assessments priced barely above a 30-minute appointment (E2).
- Opportunities: two running clubs nearby with no podiatry partner (from the clinic's own enquiries, E11); rising demand for gait assessments from people starting running, which the research pass found sources for and the owner checked.
- Threats: a larger foot-care chain opening a branch nearby, offering cheaper routine nail care (competitor page, E20); orthotic lab prices up at the last two renewals (E6).
The stress test moved "we offer online booking" out of strengths, since both competitors do too. It also flagged that the associate podiatrist leaving would be a serious risk, which the owner hadn't written down.
The three actions: offer running clubs a gait-assessment evening (strength plus opportunity); cut no-shows with reminders and a small deposit before the chain opens (weakness plus threat); and reprice biomechanical assessments to reflect the hour they take. The clinic tested the pricing change on new bookings first, the approach described in testing a new service idea with AI before you launch it.
Each action had one number attached so the clinic would know in three months whether it had worked: no-shows on routine appointments down from 14% to under 8%, at least four new gait-assessment bookings from the first club evening, and no drop in biomechanical bookings after the price change. The targets are examples, but the habit matters. An action without a measure tends to be quietly forgotten by the next busy month.
How AI-written SWOTs go wrong, and how to spot it
- Every point fits every business. Test: could a competitor copy your list unchanged? If yes, it's generic.
- Actions in the opportunities box. "Launch a newsletter" is something you'd do, not something happening in the world. Opportunities are external changes.
- Invented competitor facts. AI will sometimes say a competitor offers a service it doesn't. Only trust claims you can see on their current page.
- Stale market claims. Without a dated source, a "trend" may be several years old. Your prompt asked for dates; hold it to that.
- Your framing, flattered back. If your evidence pack is all praise, the SWOT will be too. Include the complaints.
- Too many points. Twelve strengths means none of them is special. Cap each box at five.
Before the final version goes to a partner or a lender, run it through a five-minute fact-check routine. Any figure in it should be traceable to your own records or a source you've opened.
When to run it again
A SWOT is a snapshot, so set a date to redo it: every six months for most small businesses, and straight away when something big changes, such as a new competitor opening, a key member of staff handing in notice, or a supplier raising prices sharply. Keep the evidence pack and the prompts from this round. Next time, update the pack, rerun the same prompts, and ask the AI one extra question: "What has moved between the last SWOT and this one, and what does that suggest?" The comparison is often more useful than either SWOT on its own, because it shows whether last round's actions actually shifted a weakness or simply sat in a folder.
For the podiatry clinic, the second round's answer (illustrative) opened with three lines: no-shows on routine appointments down from 14% to 9% since deposits started, short of the 8% target; missed calls during clinics unchanged; and the chain's branch open for four months, with routine nail-care bookings down 6% since. The unchanged line is the one to act on. Missed calls weren't among the first three actions, and six months on they matter more, because a patient who can't get through now has a cheaper option down the road that answers its phone. That moved the phone problem into the next round's top three, with the no-show action kept running rather than restarted.
Further reads
- How to Use AI for Scenario Planning: Best, Worst and Likely Cases — Take the threats box into best, worst and likely cases.
- How to Write a One-Page AI Strategy for Your Business — Where the SWOT actions can feed your AI plans.
- How to Use ChatGPT as a Business Adviser Without Being Misled — Keep the AI honest in strategy conversations.
- How to Write a Business Plan With AI, and What to Check Yourself — A SWOT is often the first section of a plan.
- How to Track Competitors' Prices and Offers With AI — Keep an eye on the threats you found.
- Can AI Tell You What to Charge? The Limits of AI Pricing Research — If pricing turns up as a weakness.
- How to Build an Investor Pitch Deck With AI, and What to Check — Which parts of an investor deck to hand to AI and which to keep, slide by slide, plus the checks that catch invented market figures and mismatched numbers.
- How Consultants Use AI for Client Research Before Discovery Calls — A research routine for discovery calls: scale the prep to the deal, make AI cite everything, check what matters and turn findings into sharper questions.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.