AI Guides › Workbench
By Nigel Guy · 9 min read
Most people prepare a talk the same way. They open a slide template, fill it with everything they know, rehearse in front of colleagues who already know the work, and finish on "Thank you — any questions?" That feels thorough. But Patrick Winston spent an hour at MIT taking that routine apart. The lecture is free, and asking an AI to "make my presentation better" won't get you any of it. You'll get a tidier version of the same mistakes.
The rule: use the lecture as a checklist and Claude as the person who applies it. Run one prompt for one job at a time, on your real material, and keep the judgement calls for yourself.
Winston was a long-serving MIT professor who directed its AI Laboratory from 1972 to 1997. He died in July 2019. His How to Speak talk, recorded for MIT's January IAP term in 2018, is on MIT OpenCourseWare as a free 63-minute video with a full transcript. MIT describes the course as a tradition going back more than 40 years.
There is no Claude feature built on Winston's framework. If a post says Claude "can now" do this, it means someone wrote prompts. The seven below are my own and paraphrase the lecture's heuristics. They work in Claude's free tier or any capable chatbot. Watch the lecture once before you use them. Each prompt is a reminder of what he teaches, not a substitute for it.
| Prompt | What it does | Cost at time of writing | Best for | Catch |
|---|---|---|---|---|
| 1. Promise opener | Writes an empowerment promise to replace the opening joke | £0 on Claude Free | Any talk | Only as good as the outcome you give it |
| 2. Slide crime audit | Checks a deck against Winston's slide crimes | £0; a big deck may hit Free limits | Decks you've already drafted | It can't see the room or how you deliver |
| 3. Winston's star | Builds symbol, slogan, surprise, salient idea, story | £0 | Research, products, pitches | Can invent a surprise that isn't true |
| 4. Room recce | Turns time and place into questions for your host | £0 | External or high-stakes talks | You still have to visit the room |
| 5. Inform and inspire | Plans a teaching talk with cycling, fences, signposts | £0 | Lectures, training, workshops | Passion has to be yours |
| 6. Five-minute case | Gets vision and steps on the table by minute five | £0 | Job talks, pitches, approvals | Overclaims unless you mark what's finished |
| 7. The stop | Contributions slide and final words | £0 | Every talk | Won't stop you saying "thank you" on the day |
Claude Pro was around £18 a month including VAT at time of writing. Anthropic lists US prices, so check the sterling figure at checkout. You don't need Pro for any of this.
Winston advises against opening with a joke, because people are still settling in. Open instead with what they'll know at the end that they don't know now.
You are a speaking coach who follows Patrick Winston's advice: open with an empowerment promise, not a joke.
Talk topic: [TOPIC]
Audience: [WHO, AND WHAT THEY ALREADY KNOW]
Length: [MINUTES]
What they should be able to do or decide afterwards: [OUTCOME]
Steps: 1) If the outcome is vague, ask me one question to sharpen it before writing. 2) Write three opening promises of no more than two sentences each, naming something specific the audience will gain. 3) For each one, add a single line of passion: why I find this worth caring about.
Constraints: no jokes, no "thank you for having me", no claims about results I haven't given you.
Before answering, check that each promise could be proved true or false by the end of the talk.
Fill in topic, audience, length and the outcome you want.
The crimes, as Winston lists them: too many slides, too many words, reading your slides aloud, background clutter, logos and titles you don't need, small fonts, laser pointers that turn you away from the room, and a deck that's "too heavy" when you print it and lay it out.
Act as a blunt slide reviewer using Patrick Winston's list of slide crimes.
My deck (slide by slide text, or attached): [DECK]
Talk length: [MINUTES]
For each slide, flag: more words than I'd say aloud in a sentence; any line I'd likely read verbatim; decoration, logos or titles that add nothing; text likely below roughly 40pt; anything I'd need a laser pointer for (suggest a numbered arrow instead).
Then judge the whole deck: is it too heavy (too much text, too little white space or imagery)? Which slides could go entirely? Is there more than one deliberately overwhelming slide (allow one at most)?
Output a table: slide number | crime | fix. Then a cut list.
If you can't see fonts or images, say so; don't guess.
Paste or attach the deck and give the talk length.
What makes ideas memorable, in his account, is five things that start with S: a symbol, a slogan, a surprise, a salient idea (one that stands out, not necessarily the most important) and a story about how it works and why it matters.
You are helping me package one piece of work so it's remembered, using Winston's star: symbol, slogan, surprise, salient idea, story.
The work: [WHAT YOU DID, IN PLAIN WORDS]
Evidence I can stand behind: [RESULTS, FIGURES, EXAMPLES]
Audience: [WHO]
Give me two options for each point of the star. The surprise must be something my evidence actually supports; if nothing does, say so rather than inventing one. Pick the single salient idea and explain why the others should take a back seat.
Ask me for evidence first if what I've given is too thin.
Fill in the work, your evidence and the audience.
Claude can't see your venue. What it can do is turn Winston's points into a list of things to ask: a sensible time (he likes 11am, after people are awake and before lunch), a well-lit room, a room you've "cased" in advance, and a room that's more than half full.
I'm giving a talk and need to check the time and place before the day.
Event details: [DATE, TIME, VENUE, EXPECTED NUMBERS, ROOM CAPACITY, IN PERSON OR ONLINE]
Write: 1) a short email to my host asking about lighting (can it stay full up?), room size versus attendance, a board or flip chart, and a visit beforehand; 2) a ten-point checklist for when I visit; 3) any risk in the timing (straight after lunch, end of day, competing sessions) and one change I could ask for.
Don't assume facts about the venue I haven't given you; list them as questions.
Fill in the event details you have and leave the rest blank.
For teaching talks, Winston uses four heuristics. He cycles the key idea three times, because a share of any audience has drifted off at any given moment. He builds a fence around it by saying what it is not. He uses verbal punctuation, numbered signposts where people can rejoin. And he asks a real question, then waits about seven seconds for an answer. He also prefers a board or a prop to slides when the aim is to teach.
Plan a [MINUTES]-minute teaching session on [TOPIC] for [AUDIENCE].
Key idea they must leave with: [IDEA]
Ideas it's commonly confused with: [NEIGHBOURING IDEAS, OR "unsure"]
Produce an outline that: states the key idea three times in different words at spaced points; includes one "this is not that" contrast; marks numbered signposts; includes two questions that aren't obvious but are answerable, with where to pause; and suggests one physical prop or board drawing instead of a slide.
Mark where I should show genuine enthusiasm, and leave those lines for me to write.
Fill in length, topic, audience, the key idea and anything it gets confused with.
Winston's account of hiring panels is blunt. A candidate has about five minutes to show a vision, meaning a problem people care about plus something new in the approach, and to show they've done something, by listing the steps towards it. The talk ends by listing contributions, which mirror those steps.
You are preparing me for a room that will decide something: [JOB TALK / PITCH / FUNDING OR APPROVAL MEETING].
The problem I address: [PROBLEM AND WHO CARES]
What's new about my approach: [APPROACH]
Steps to the solution, marked done / in progress / planned: [STEPS]
Write the first five minutes as speaking notes: problem, new approach, then the steps, with what's done stated plainly. Then write the matching contributions list for the end.
Constraints: never present a planned step as complete. Flag any step where my evidence is weak.
If my approach doesn't sound different from existing work, tell me and ask what separates it.
Fill in the decision, the problem, your approach and your steps with their status.
Winston calls a final slide of "Questions?", "The End" or "Thank you" wasted space. A long list of collaborators belongs on the first slide, not the last. A slide headed "conclusions" invites a shrug. The slide that stays up during questions should be contributions. For the final words, he'd use a joke (by the end it can land) or a salute to the audience rather than "thank you".
Help me end my talk on [TOPIC] for [AUDIENCE].
My contributions (what I actually did): [LIST]
Collaborators to credit: [NAMES, OR "none"]
Write: 1) a contributions slide of no more than five short lines; 2) where collaborators go on the opening slide; 3) two closing lines that salute this specific audience, plus one optional light closing line if the setting suits humour; 4) one sentence that tells the audience I've finished, without "thank you" or "any questions?".
Ask me what's genuinely mine if the list blurs my work with others'.
Fill in the topic, audience, your contributions and anyone to credit.