Practising and Reviewing Talks
🗒️ Description
The loop around a talk: how I rehearse before it, how I get feedback that does not mislead me, what I do in the hours and days after, and what I do with the talk once it exists. Eleven techniques, the third note of the second research wave behind Conference Talk Craft. It also holds the two “meta” items that sit before the first rehearsal: the abstract, which is a promise to the room, and the language layer for talks given in Polish about a field that speaks English.
The shape of the evidence is different from the other notes. For practice there is a large literature, but it is about word lists, motor skills and classroom speeches, not about 25-minute technical talks. Nobody has tested a rehearsal schedule or a debrief on conference speakers. So most of this is a sensible transfer with the seams shown, and one item (speech tempo) is deliberately left open.
Siblings: Talk Narrative and Structure, Slides for Technical Talks, Delivery and Stage Operations.
🧩 The techniques
Before the talk: rehearsal and feedback
| Technique | What it is | Priority |
|---|---|---|
| Spaced rehearsal | Spend the same total rehearsal time (about 3–4 hours for a 25-minute talk) over 4–6 sessions across 5–10 days instead of one block the night before. Each session starts from memory, with no slides, and only then checks against the deck. | High, free |
| Self-recording and review | Record a full run, then watch it in passes that each look at one thing. Write a prediction of the problems first. Use at most five criteria, and leave the review with one change to make. | High |
| Peer dry run | One run for a listener who is allowed to interrupt, structured to avoid the usual failures: they explain the talk back first, mark where they lost the thread by slide number, answer written review questions, run a premortem (“it is a year later and this talk flopped — why?”), and only then give opinions. | High before a new talk; medium for a repeat |
| Speaking clubs | Regular low-stakes practice in front of strangers. A club is a source of volume: roughly 50–100 hours a year of speaking and being timed. A self-run drill gives most of the drilling value: a random prompt, a timer with signals at 0:30, 0:50 and 1:15, and a one-sentence review. | Low as a way to improve one talk; medium as a long-term habit |
Around the talk: contract, language and review
| Technique | What it is | Priority |
|---|---|---|
| Abstract as a binding contract | Title as verb + object + outcome. Prerequisites in place of a level label (“you have used an agent framework”). Three concrete things the room will see. One line on what is out of scope. At 14 days before the talk, diff the abstract against the actual deck and fix whichever is wrong. | High |
| Polish and English language layer | Decide the language per layer, not per talk: slides, speech, demo output, glossary. Never two languages on one slide. Default for a Polish room is Polish slides and Polish speech. Keep a standing glossary. Put a Polish carrier noun before an English term and decline the noun, not the term (“protokół MCP”, “w narzędziu Claude Code”); where the term itself must take an ending, use an apostrophe or hyphen (LLM-a, Claude’a). | High if the room is Polish |
| Post-talk retrospective with a rubric | Two passes. Hot, within two hours: a form of 3–5 questions plus a ten-minute voice memo answering the four After Action Review questions (what was supposed to happen, what did, why the difference, what do I do next time). Cold, two to seven days later: score the recording against a rubric whose anchors are written down, so talk five can be compared with talk one. At most two action items come out. | High |
After the talk: what the artefact becomes
| Technique | What it is | Priority |
|---|---|---|
| Signature talk | One talk with a stable spine, delivered 5–10 times to non-overlapping audiences. Swap the stale examples, keep the argument, keep it versioned in git, and write down per delivery what changed. By the second delivery the cost per delivery is lower than a fresh talk. | Medium; high if there is a talk worth repeating |
| Talk-to-content reuse | A fixed set released over the following weeks, one a week: a post the same evening, a carousel rebuilt for reading (not the stage PDF), three to five clips cut from blocks designed as stand-alone units, a newsletter piece, a vault note. | Medium |
| The talk as a YouTube artefact | Agree rights with the organiser before the talk. If the video is for online viewers, record it separately for the camera instead of reusing the room recording. Chapters follow YouTube’s rules (the first at 00:00, at least three, each at least ten seconds). Corrected captions. | Medium |
| Ethics and affiliation | Say who employs me in the first minute. One contact slide, not one per section. Use only images and memes I can publish. Get permission before naming a client case, even anonymised. Cut jokes that exclude part of the room. | High, and cheap |
⚖️ What the evidence actually supports
Sound practice, wrong justification
| Practice | The usual reason | What the research says |
|---|---|---|
| Spaced rehearsal | ”Spacing is proven to improve memory, so it will improve a talk” | Proven for facts and simple motor skills: the optimal gap is about 20–40% of how long I need to remember (Cepeda 2008), and a meta-analysis of practice-distribution studies found d = 0.46 overall, d = 0.97 for simple tasks but only about 0.1 for complex ones (Donovan & Radosevich). No study uses a full talk. The retention interval here is days, so the gap is days. Learners do not trust it: in one study 72% judged massed practice at least as good although spacing worked better (Kornell 2009). Retrieval beats rereading (Roediger & Karpicke 2006), so rehearse aloud from memory, not by flipping slides. Whether the benefit comes from sleep consolidation is disputed. |
| Self-recording | ”Seeing yourself is painful but it works” | Video feedback improves communication skills with an effect near d = 0.40, larger when a standard form guides the viewing (Fukkink 2011; an earlier synthesis, Bourhis & Allen 1998, was weaker). Watching alone barely helps self-assessment (r rose from 0.48 to 0.49 against experts), watching against a written benchmark does more (Hawkins 2012, small study). Self-ratings are lenient on content (De Grez 2012) and harsh on appearance: 98% of 47 socially anxious patients came across better than they had predicted (Warnock-Parkes 2017). Hence the written prediction first. |
| Peer dry run | ”Ask a friend what they think” | Feedback can hurt: in a review of 607 effects over a third of the interventions made performance worse (Kluger & DeNisi 1996), usually because they pointed at the person, not the task. Peers and experts notice different things: peers mark delivery and slides, faculty mark content (Taylor & Swanberg 2020), and teacher feedback beat peer feedback on presentation behaviour (van Ginkel 2016, 2017). The speaker cannot predict the listener (the tapper study, whose numbers are shaky — see Talk Narrative and Structure). The premortem’s “30% better” figure is repeated far more than it is tested. |
| Speaking clubs | ”A club makes you a good speaker” | No controlled study of membership outcomes exists; the testimonials come from people who stayed. What a club does reliably is give hours, a timer and a forgiving room. The audience is also general, so it trains pleasing a mixed room, not a room of engineers. |
| Retrospective | ”Review it right away while it is fresh” | Next-day, memory-based ratings of lecturers were more accurate than immediate ones, though with more halo (Murphy & Balzer 1986), and a clerkship study found no difference with delay. The defensible reasons for a hot pass are response rate and capturing detail that decays: which block lost them, which question they did not ask. Debriefs themselves work (d ≈ 0.67 across team debriefs, Tannenbaum & Cerasoli). Satisfaction forms say little about learning (Alliger 1997), so ask what the audience did or understood, not whether they liked it. |
| Signature talk | ”Practice makes perfect” | Practice follows a power law: each repetition helps less (Newell & Rosenbloom 1981). Past the fifth to tenth delivery the gain is small and the costs are staleness and a recital quality. The advice to rehearse a signature talk until it is flawless does not follow from the curve. |
| Reuse | ”Spaced release helps the audience remember” | Spacing is supported by a large meta-analysis (Cepeda 2006, 317 experiments), but for learners of facts. Treat weekly releases as a distribution plan with a plausible side benefit, not a memory intervention. |
| YouTube chapters and captions | ”Chapters raise ranking” | No published evidence. Captions do help comprehension (Gernsbacher 2015), which is the reason to correct them. |
| Disclosure | ”Telling people my employer will cost me their trust” | A meta-analysis of sponsorship disclosure found it does not reliably reduce persuasion (Eisend 2020). Disclosure is a duty, and the feared cost is smaller than it feels. |
| Abstract | ”A good abstract attracts attendees” | It also sets the expectation the talk is judged against (expectation-disconfirmation, Oliver 1980). Presentation Patterns names the audience member who holds you to the printed text the Abstract Attorney. Cutting content is fine; silently switching topic is a broken contract with people who chose the session. |
Facts with dates (check before quoting)
- Toastmasters in Poland: first Polish club on 1 July 1994 in Warsaw; about 60 clubs in 2017; Poland became its own District 108 in July 2018. Dues went from 60 to 72 USD per six months on 3 August 2026. All as read from the organisation’s own pages on 2026-11-06.
- Image and quotation rights (Polish copyright law, not legal advice): the right of quotation (arts. 29 and 34, with the source and author named), the limit on use that harms the normal exploitation of the work (art. 35), and consent to publish a person’s likeness (art. 81). The “Distracted Boyfriend” meme is a licensed stock photo, not a free image; before publishing a recording, clear the memes.
- Chapters: YouTube needs the first at 00:00, at least three, each at least ten seconds.
🧱 What stays open
- Polish speech tempo for me: OPEN. I have no usable recording of my own talks, and a tempo number measured from a cohort meeting transcript came out at 29–43 words per minute, which is implausible and was discarded. I do not import the English 130–150 words per minute, and the Polish “4–5 syllables per second” figure circulating online is second-hand. The first move is to record one full run and measure syllables per second with pauses removed. See Delivery and Stage Operations for the voice controls.
- No evidence about Polish talks. The language-layer advice is expert practice. No study tests English slides with Polish narration.
- Data-retention terms of LLM tools used as a rehearsal partner were not verified. Do not paste a client’s material into one without checking the plan’s terms.
🔧 Cheapest first moves
- Calendar the rehearsals: five 40-minute slots over the week before, each starting from memory.
- Prediction card: before watching a recording, write three things I expect to see, then compare.
- Premortem line: ask the dry-run listener one question first, “it is a year later and this talk failed: why?”, and write the answer before anything else is said.
- Abstract diff at T-14: paste the abstract and the slide headlines side by side.
- Glossary file: one list of the 20 terms I keep switching on, with the form I use in Polish.
- After Action Review memo: ten minutes, four questions, recorded on the phone within two hours.
📖 Further reading
- Conference Talk Craft — wave one: rehearsal ladder, timed dry run, reuse and Q&A
- Public Speaking — hub note
- Talk Narrative and Structure · Slides for Technical Talks · Delivery and Stage Operations
- What Carries Over Between Talks — my own record of what survives from one talk to the next
- The Art of Public Speaking — a classic on preparing and delivering a speech
- TED Talks — Anderson on preparing a talk
Template: knowledge_note_info