Last spring a neighbour asked me to help organise a street party for 60 people. I had no supplier list, no budget spreadsheet, and no clue how many sausages 60 humans actually eat. What I had was a ChatGPT account and a free afternoon.
By six o’clock I had a run sheet, a shopping list, a rain plan, and a one-page brief for eight volunteers who had never done this before. The party happened. Seventy-four people came, we spent £648 of a £700 budget, and the rain plan got used for exactly twenty minutes.
Here is the whole process, step by step. Swap the street party for a product launch, a dissertation, a wedding, or a client report and it works the same way.
Step 1: Write a brief, not a question
Most people type “help me plan a party” and then wonder why the answer sounds like a greeting card. The output is generic because the input is generic.
Spend four minutes writing a brief instead. Four sentences is enough.
“You’re helping me plan a street party for 60 neighbours in a residential cul-de-sac on 12 July, 1pm to 7pm. Budget is £700. I have eight volunteers who’ve never run anything like this. I need a plan I can hand to them.”
That one paragraph changes everything downstream. Role, audience, limits, deliverable, all stated up front. If you want a deeper version of this framing habit, there’s a solid breakdown of how to brief an AI chat like a new colleague that’s worth ten minutes of your time.
Step 2: Ask for options before you ask for a plan
Divergent first, convergent second
If you ask for a plan immediately, you get the first plan that pops into the model’s head, dressed up with headers. Instead, make it generate a spread.
My prompt: “Give me ten different formats this party could take, from cheapest to most ambitious. One line each. Don’t explain them yet.”
Ten options took maybe fifteen seconds to produce. A bring-and-share picnic, a hired hog roast, a pizza van, a sports-day format, a film-night format. Three of them I’d never have considered, and one of those became the actual plan: a bring-and-share with a paid pizza van as backup.
Then, and only then, pick two or three and go deep.
Step 3: Make it interview you instead
This is the step almost everyone skips. Before it plans anything, have it hunt for the information it’s missing.
“Before you write anything, ask me the ten questions you’d need answered to plan this properly.”
Mine came back with questions I hadn’t thought about: Is there a noise curfew? Do we need a road closure permit? Who has allergies? Is there a toilet within walking distance? Where does waste go? Two of those questions turned into real tasks and one of them (the permit) would have derailed the whole thing if we’d discovered it three days before.
A chatbot will happily plan around gaps it can’t see. Making it ask first is the cheapest insurance you’ll ever buy.
Step 4: Hand it a scoring rule
Vague decisions produce vague recommendations. Give it criteria and it becomes genuinely useful.
“Score the three food options on cost per head, prep time, and how well they work outdoors with no power. Show a table. Recommend one, then tell me what would change your mind.”
That last clause matters. Asking what would change its mind forces the chatbot to show you the assumptions under the recommendation, so you can spot the weak one rather than discovering it at the till.
Step 5: Build the deliverable in chunks
Do not ask for the entire plan in one message. You’ll get something shallow and padded.
- Message one: the hour-by-hour run sheet. Start times, who does what, when.
- Message two: the shopping list, grouped by shop, with quantities for 60 people.
- Message three: volunteer roles, one paragraph each, written as instructions a first-timer can follow.
- Message four: the rain plan. A separate, shorter run sheet for wet weather.
Each chunk gets full attention, and each one builds on the last because it’s all in the same conversation. If your chat keeps drifting or forgetting earlier instructions, the eight-step AI chatbot how-to covers the fixes for that in more detail.
Step 6: Stress test it before you trust it
A plan that only works in perfect conditions isn’t a plan. Ask directly:
“Poke holes in this. What breaks if it rains, if three volunteers drop out, or if 90 people turn up instead of 60?”
This produced the single most valuable output of the afternoon: a sentence telling me that the bring-and-share model collapses if attendance is more than 30% above estimate, because food contributions scale with households, not guests. We set a hard cap on the guest list because of that line.
Step 7: Check anything numeric or legal yourself
A ChatGPT AI chatbot is excellent at structure and terrible at being a source of record. It will produce a confident cost-per-head figure that is off by a third, or cite a food-hygiene rule that varies by country.
My rule for the afternoon: use it for shape, verify for fact. Anything with a number, a regulation, or a name attached got checked against a real supplier quote or a council website before it went in the plan. There’s a clear-eyed breakdown of where ChatGPT gets it right and where it gets it wrong that’s worth reading before you rely on it for anything consequential.
Step 8: Turn the whole thing into a reusable recipe
The last thing I did cost thirty seconds and saved hours. I asked it to write a master prompt: a single block of text containing the brief, the constraints, the interview questions, the scoring criteria, and the order of the deliverables.
Now when I need to plan something, I paste that block, change four details, and start from step three instead of step one. Event planning, client onboarding, a kitchen renovation, a six-week content calendar. Same skeleton, different filling.
Worth knowing: the underlying models keep shifting, and the features available to you (memory, file uploads, custom instructions) depend on which version you’re on. If you want the current lay of the land, this explainer on OpenAI’s models in 2025 and what you can actually build is a good check-in point.
What those four hours actually produced
Not a magic plan. A first draft that was about 70% right, which is exactly what you want from a tool like this. The remaining 30% was local knowledge, supplier phone calls, and one very patient neighbour with a van.
The deliverables, in case you want to aim for the same shape:
- An hour-by-hour run sheet from 1pm to 7pm, with named owners for each block.
- A shopping list of 23 items grouped by shop, quantities scaled to 60 with a stated buffer.
- Eight one-paragraph role briefs, short enough to read on a phone.
- A one-page rain plan that we used for twenty minutes and were glad to have.
- A reusable master prompt for the next project.
Total cost of the planning itself: nothing. The part that mattered wasn’t the tool. It was refusing to accept the first generic answer, insisting on the interview questions, and writing down the numbers so a human could check them. Do those three things and a chatbot stops being a novelty and starts being the most useful unpaid assistant you have.