Lesson 7 of 7 · 7 min read

Critique, then rehearse

Review the rendered slides against an explicit checklist, own every claim on them, then rehearse aloud with a five-minute version.

People miss slide flaws

Your deck now exists in your own template. This lesson checks it twice. First comes a critic pass against an explicit checklist. Then you rehearse it aloud.

The case for a checklist comes from Stephen Kosslyn and three colleagues, who published three studies in 2012. In the first, academic and business slideshows broke eight principles of perception and memory at similar rates. In the second, audiences reported that these flaws irritated them.

The third study matters most here. Observers often could not spot a flaw on a slide. When they did spot one, they often could not say what was wrong. So reviewers need an explicit list, not instinct, to find slide problems.

This is experimental evidence for reviewing with a checklist. The studies tested general slide principles, not AI-made decks, and they did not test this course’s checklist. They show that a reviewer without a list misses things.

The critic pass

The checklist is the four rules. Does the deck open with the answer and the ask? Does each slide make one claim in an action title? Does evidence with a source sit under every claim? Is the deck free of the AI tells from Lesson 5?

An assistant can run that checklist as a critic. The first prompt below asks it to review the deck as a skeptical executive. Emeri’s research team assembled that prompt from the published lists of tells. No source has tested it, so treat what it returns as a list to check, not a verdict.

Give the critic your list of tells from Lesson 5 as well. It names the problems your assistant produced before, so the critic can look for them first. Run the critic in a new conversation, with only the PDF and your sources attached. It then reads the deck without your earlier instructions, as a reader would.

Read each finding against the checklist. Fix the ones the checklist supports. Drop suggestions that add decoration or restate taste, even when the critic sounds sure.

Check the rendered slides, not the outline

Some problems only show on the finished slide. Text runs past the edge. A chart label is cut off. Two boxes overlap. Body text is too small to read across a meeting room. An outline or a list of slide text shows none of these.

One Hacker News user reports that Claude checks its own decks by converting them to images and looking for overflowing text. A critic that works only from written instructions cannot render slides in every product. So do the rendering yourself.

Check the deck in the form each reader meets it. A projected deck needs text large enough to read from the back of the room. A deck that someone reads alone has to make sense without you there to explain it. Lesson 1 made that choice for your deck, so check the file against it.

Export the deck as a PDF, or as one image per slide. Give that file to the critic with your sources. Then page through it yourself at full size, and look at every slide you changed since the last check.

You own every claim and every word

The presenter owns every claim and every word on the slides, whoever drafted them. Lesson 5 described presenters whose polished AI decks pleased management. They could not explain terms on their own slides, because the model had chosen those terms.

Go through each slide and ask three questions. Can you name the source of every number? Can you explain every term and abbreviation in your own words? Would you defend this claim if the most skeptical person in the room asked about it? Rewrite or delete anything that fails.

Check the speaker notes the same way. Notes that an assistant drafted may use words you would never say aloud. Replace every sentence you would not say yourself.

Rehearse aloud

Rehearse aloud, not in your head. Start with the titles-only test from Lesson 3, said aloud this time. If the titles do not make the argument when you say them, the deck does not either.

Then give the five-minute version. Nancy Duarte’s advice for presenting to senior executives is to imagine your slot was cut to five minutes. You would lead with the findings, the recommendation and the ask, and bring in the supporting data after. That is expert consensus. Time yourself, and cut until it fits.

Last, do not read your slide text aloud word for word. Mayer, Heiser and Lonn found in a 2001 study that on-screen text repeating the narration lowered retention, compared with no on-screen text. The experiments tested narrated lessons, not meetings.

A 2012 meta-analysis by Adesope and Nesbit, described in a later paper, found that a few key words on screen did better than full text. So short titles and labels beside the evidence are fine. That is experimental evidence. Lesson 1 put your full script in the notes so you could rehearse from it. Once you know the talk, cut the notes to key words, so you speak instead of read.

Give the talk once to one person who has not seen the deck. At the end, ask them to state the ask in one sentence. If it differs from your one sentence from Lesson 2, fix the deck before the meeting.

Worked example · Larkfield Supply

Morgan’s critic pass and one rehearsal

Morgan has seven slides and an appendix slide from Lesson 6. The leadership meeting is in three days, and Priya receives the memo the day before.

Morgan exports the deck to PDF. In a new conversation, Morgan attaches the PDF and the sources and runs the critic pass.
The critic lists four problems. The third title, “Overtime last peak”, names a topic, because the assistant shortened it during the build. The 96% figure lost its “Morgan’s estimate” label when the chart was rebuilt. The legend on the slide 2 chart is cut off. The cost slide has an icon beside each line that carries no information.
What would you do next?The critic also calls slide 7, the decision slide, “too plain” and suggests a photo of a delivery truck. Should Morgan add it?Think it through, then show the answer
No. The photo would be decoration unrelated to the claim, which is a surface tell from Lesson 5. The critic pass is untested, so Morgan keeps only the findings the checklist supports.
Morgan reads every word on every slide. Slide 2 uses the term “OTIF rate”, which the assistant introduced. Larkfield tracks on-time shipments, not that measure, and Morgan could not explain the difference in the room. Morgan changes it to “on-time shipments”.That answers the question you guessed at: Morgan’s assistant put the term “OTIF rate” on a slide. Larkfield tracks a different measure. Guess: what should Morgan do with the term?
Morgan restores the action title from Lesson 3: “Overtime ran 410 hours but did not hold the on-time rate.” Morgan restores the estimate label, fixes the legend and removes the icons. Morgan exports again and checks the changed slides in the PDF.
Morgan reads the titles aloud to a shift supervisor, then gives the five-minute version with a timer. It runs seven minutes.
What would you do next?What should Morgan cut to reach five minutes?Think it through, then show the answer
The walk through overtime week by week, and the detail of the shift hours. Both are in Priya’s memo and the appendix. The ask, the fall from 97% to 88%, the cost against last peak’s overtime and the labeled estimate stay.
On the second run, Morgan notices reading the cost slide’s notes aloud word for word. Morgan cuts the notes for that slide to four key words.

The second run takes four minutes and forty seconds. Morgan can name the source of every number and explain every term. The deck goes into the leadership pack with the memo. Whether the team approves the shift is its decision. The deck’s job is to make the case clear and true.

Starter prompts

Critic pass (untested)

I have attached my deck as a PDF and my sources.

[attach the PDF or slide images, and your sources]

Review this deck as a skeptical executive. List slides with topic-label titles, equal-weight layouts, unsourced claims, or decoration that carries no information. Also list any text that runs past the slide edge, overlaps other content, or is cut off. For each problem, give the slide number and what you saw. Do not rewrite the slides. Do not add facts. If you cannot see the rendered slides, say so and stop.

Questions I must be able to answer

Here is the text of my slides and speaker notes:

[paste your slide text and notes]

List every technical term, abbreviation, number and named fact. For each one, write the question someone in the room could ask about it. Do not answer the questions for me. Do not suggest new content.

Titles and the five-minute version

Here are my slide titles in order:

[paste your titles]

Read them as one paragraph. Tell me where the argument has a gap, where two titles make the same claim, and where a title names a topic instead of stating a claim. Then list which slides I could leave out of a five-minute version without losing the ask. Do not add claims of your own.

When not to use AI

Deciding whether a claim is true. The critic can flag a number with no source, but only you can check the source.
The final look before the deck goes out. Page through every rendered slide yourself, because a critic that cannot see the rendered pages misses overflow.

Recall from memoryName the three questions you ask of every slide before you present it.Answer it without looking back, then show the answer

Can you name the source of every number? Can you explain every term and abbreviation in your own words? Would you defend this claim if the most skeptical person in the room asked about it?

Exercise · your project

Critique your deck against the checklist, then rehearse it aloud.

Export your deck as a PDF or as slide images.
Run the critic pass with your sources, and fix only the findings the four rules support.
Run the second prompt and answer each question aloud. Replace any term you cannot explain.
Say the titles aloud in order, then give the five-minute version with a timer and cut until it fits.
Once you know the talk, cut your speaker notes to key words, then check every changed slide in a fresh export.

Mark this lesson as done