three to five bullets — the workhorse of decks
- Strength
- reliable scaffolding when the items are parallel and short
- Common failure
- becomes slide-fill — six bullets, none of them earning their line
§ 6.b · text
A bullet is a sentence that gave up.
The corpus runs 4.3 lists for every callout across 475,051 visual text atoms. Bullets default. Emphasis is rare.
List 210k
Callout 49k
4.3× more lists than callouts
- List 34.2% three to five bullets — the workhorse of decks
- Paragraph 36.8% connected prose — when the argument needs sentences, not bullets
- Callout 12.7% one sentence in a frame — what the eye must catch
- Quote 3.2% attributed voice — credibility imported from a source
Text is four kinds, all measured by their bounding boxes: list (210k atoms), paragraph (200k), callout (49k), quote (16k). The 65k callouts and 15k quotes WITHOUT bounding boxes — LLM-inferred mentions in body text — sit outside the visual layer. Those are textual mentions, not visual atoms.
Four kinds. One job each.
A list scaffolds. A paragraph argues. A callout commits. A quote imports voice. The corpus uses lists for everything — including the jobs the other three do better.
connected prose — when the argument needs sentences, not bullets
- Strength
- argues with cohesion — clauses, transitions, qualifiers
- Common failure
- wall of body copy — slide becomes a memo with margins
one sentence in a frame — what the eye must catch
- Strength
- does the prioritisation FOR the reader; the strongest editorial move
- Common failure
- overuse — three callouts compete; none lands
attributed voice — credibility imported from a source
- Strength
- human voice cuts through analytical prose; one quote can carry a slide
- Common failure
- unattributed or self-attributed — reads as vanity, not evidence
82% of lists ship two bullets or fewer.
The bullet wall is the loud failure — but rare. Across 210,152 lists with measured bounding boxes, the distribution skews tight: the corpus self-disciplines at ≤4 bullets 92% of the time. The 3% that goes 7+ does the most damage.
- 1-2 172k 82.1%short list — usually a definition pair or a contrast
- 3-4 21k 9.8%classic list — the storymakers safe-zone
- 5-6 10k 5%over the line — eye starts to skim
- 7-9 4k 2.1%wall territory — most points lose their slot
- 10+ 2k 1.1%pure bullet-soup — no editorial work happened
A quote is twice a callout. A list trails wider.
Character-count quartiles per kind. The interquartile range (p25-p75) captures where the typical atom lives; the marker is the median. Each kind has its own length-shape — and the editorial role follows the geometry, not the other way round.
- List 85 138 221three to five bullets — the workhorse of decks
- Paragraph 84 162 254connected prose — when the argument needs sentences, not bullets
- Callout 57 91 145one sentence in a frame — what the eye must catch
- Quote 125 184 259attributed voice — credibility imported from a source
0 80 160 240 320 characters per atom
Same content. Two slides.
Six bullets that all matter equally land as none mattering at all. One callout backed by a sentence lands as a position. Read each one cold; what do you walk away with?
Q3 results were strong across the board
- Revenue grew 23% year-over-year, beating consensus
- Net income reached a new record at $4.2B
- Operating margin expanded by 180 basis points
- Free cash flow generation continued to be robust
- Customer acquisition costs declined for the third quarter
- International markets contributed 41% of total revenue
- Form: 6-bullet list, all sub-clauses parallel
- Comparison made: none — every line is "and another thing"
- Anti-pattern: bullet-wall · no editorial hierarchy
Reader leaves with: "Q3 was good." They cannot recall a single number.
International markets carried Q3
International revenue grew +41% — twice the domestic
growth rate, for the third consecutive quarter.
The other five line items — net income, margin, cash flow, CAC, consensus beat — all moved in the right direction by smaller magnitudes. The story is geographic.
- Form: action title + callout + supporting paragraph
- Comparison made: international vs domestic growth, three quarters running
- Primitive: highlighted callout · subordinated body · single number
Reader leaves with: "International grew 2× domestic for three quarters."
A callout is six pieces, in this order.
Frame, lead, claim, number, source, anchor. The corpus's strongest callouts ship all six. The forgettable ones drop one or two — usually the source or the anchor.
Risk
International revenue grew +41% — twice the domestic rate, for the third consecutive quarter.
Source: Q3 results · audited 1 2 3 4 5 6Slide-mockup — a callout doing its full job, with each piece labelled.
- 1
Frame
visual demarcation — colour, border, fill. Says "stop here".
- 2
Lead
two-word chip or eyebrow. The genre tag. "Risk", "Opportunity", "Insight".
- 3
Claim
one sentence. The position the callout commits to.
- 4
Number
the receipt — the % / $ / × that anchors the claim. Optional but powerful.
- 5
Source
attribution — who counted, when. Inline or in chrome. Rarely both.
- 6
Anchor
connector to the rest of the slide — arrow, caret, or bold word in the title.
Four corpus callouts. One reads strong.
Read each one cold, before the verdict appears. The strong ones anchor a verb to a number; the medium ones fight themselves; the weak ones label without arguing. The pattern is editorial, not typographic.
-
Valero is a ~3MMbbl/d pure play refining system with a $49bn TEV. If Phillips' ~2MMbbl/d refining system was valued at an equivalent $/bbl it would be valued at $34bn.
Strong — arguestwo anchored numbers, one comparison, one valuation gap — the reader leaves with $34bn and the maths.
-
In fewer than 2 years under 3G's ownership, Burger King experienced 5x+ of EBITDA multiple expansion
Strong — arguesone verb (expanded), one multiple (5x+), one timeframe — anchored, sourceable, repeatable.
-
100% of Advantage's top 100 clients in 2021 were clients in 2022, with these clients representing only ~55% of total 2022 revenues, highlighting lack of concentration
Medium — fights with itselftwo stats fight for the lead — retention OR concentration? Which is the takeaway? Reader has to choose.
-
Hovde's High Performers Class of 2022, 2021; 2021 Bank & Thrift Sm-All Stars
Weak — labels, doesn't arguean award name dressed as a callout. No verb, no comparison, no claim — it is a credential, not a position.
A quote runs longer than a callout — but ships five times less.
The two strongest editorial atoms — quote and callout — sit at opposite ends of the corpus. Quotes are 11× longer than callouts on average, but ship one-third as often. The corpus prefers the editorial frame to the human voice.
A quote is the corpus's rarest text atom — three out of every hundred slides. Callouts ship four times more often (12.7%). Lists ship ten times more (34.2%).
The geometry tells you why. A callout is a sentence in a frame the writer designed; a quote is a longer artefact someone else said. The first is cheap to ship, the second is expensive to source. The frequency reflects the cost, not the editorial value.
The strongest atom in the corpus is also the rarest. That is not coincidence — it is unfinished editorial work.
- Paragraph 36.8%
- List 34.2%
- Callout 12.7%
- Quote 3.2%
Text crowds the upper-left quadrant.
Across 475,051 text atoms with measured bounding boxes, the canvas usage is asymmetric — text starts top-left and rolls down. Callouts break the pattern by sitting narrower and lower, where the eye lands second.
modal text block · 68×32%
left center right top middle bottom i
How to read this
- Heatmap
- Surface coverage. Each bbox in the corpus contributes to every 40×22 cell it overlaps; opacity is power-scaled (γ=0.6) so subtle bands stay visible. Darker = more atoms cover that point.
- Sample box
- The modal cluster. Atoms grouped by centre into 8×8 bins; the densest bin is plotted at its centroid, sized by the average width × height of atoms in it. Reads as: "the typical atom of this family looks roughly this big, here."
- Source
corpus.components.bbox— atoms with measured bounding boxes only. Inline mentions extracted from body text are excluded from this analysis.
Text peaks where the slide argues.
Text presence per slide-type tells a story the kind-frequency hides. The 'expose contradiction' slide ships text 96% of the time — the corpus's most text-heavy genre. Section dividers run lean (21%); cover slides under-text (60% — image dominates). Text follows the slide's argumentative load, not its space.
96% peak Expose contradiction
70% median Market sizing
21% floor Section divider
slide-type 18 most-frequent · n = 189,105
Text
Images
Charts
Diagrams
Tables
Metric
Chrome
Expose contradiction 11,726
96%
24%
21%
4%
33%
1%
77%
Key messages 13,445
91%
50%
8%
13%
3%
8%
29%
Executive summary 6,818
90%
35%
12%
6%
10%
17%
45%
Key takeaways 10,052
89%
34%
24%
6%
4%
17%
44%
Propose solution 5,667
87%
51%
14%
33%
6%
9%
53%
Solution 8,470
83%
63%
11%
38%
4%
6%
23%
Industry trends 8,214
78%
19%
68%
6%
7%
7%
71%
Team bio 5,245
78%
91%
3%
2%
4%
5%
14%
Appendix disclosure 9,347
72%
15%
1%
0%
22%
0%
46%
Market sizing 5,135
70%
24%
57%
14%
12%
33%
78%
Traction 10,066
65%
36%
57%
6%
11%
28%
54%
Cover 15,202
60%
89%
0%
0%
0%
0%
15%
Two-thirds of text slides ship without a source.
What companies text on slides shows what they hold the slide accountable for. Title is universal (95.1%). Image and source-note tie in second place (38.4% each). Two thirds of text-bearing slides cite no source at all — the corpus's quietest failure mode.
- 01 Title 95.1%universal — almost every text slide is titled
- 02 Image 38.4%mood / proof — text + visual is the corpus default
- 03 Source-note 37.7%attribution — text WITHOUT a source is the silent failure
- 04 Chart 24.6%argument-pair — text frames, chart proves
- 05 Table 14.3%reference-pair — text narrates, table holds the lookup
- 06 Disclaimer 8.5%caveat layer — usually finance and regulated
- 07 Metric 8.4%hero stat — text orbits the number
- 08 Diagram 7.8%structure-pair — text labels the diagram
Five failure modes. Named.
Naming a failure mode is half of fixing it. The corpus repeats five distinct text errors at scale. Three are severe (audience leaves with nothing); two are medium (audience leaves with the wrong thing).
-
▤ The bullet wall
Seven or more bullets. None earn their slot.
3.2% of lists- Bad
- A list of 9 product features, every line a noun phrase, no hierarchy.
- Fix
- Pick the three that argue. Group the rest under a heading or move to appendix.
-
∅ Verb-less bullets
Bullets that label, not bullets that argue.
28% of lists- Bad
- "Strong customer adoption · Growing margins · Operational leverage" — three labels in search of a verb.
- Fix
- Each bullet must move something — "Customer adoption tripled in EMEA". Verb in every line.
-
" Self-attributed quote
A quote with no source — or attributed to "us". Imports nothing.
41% of quotes- Bad
- "This is the future of work" — [the company itself]
- Fix
- Cut. Or replace with a customer / analyst / partner quote that imports outside credibility.
-
¶ The paragraph in disguise
A bullet that is actually two sentences.
6% of lists- Bad
- "Strong revenue growth in Q3 driven by APAC expansion. The team continues to invest in product."
- Fix
- If a bullet has a full stop in the middle, it is a paragraph. Write it as one.
-
◳ Callout soup
Three or more callouts. Each one cancels the others.
12% of slides- Bad
- Slide with four boxes around a chart, each one bolded — eye picks at random.
- Fix
- One callout. The other points become bullets or move to the next slide.
Four rules. One sequence.
Text is cheap to produce and expensive to ignore. These four rules catch the failures the corpus repeats by default — applied in order, they recover most of the editorial work text already wants to do.
- 01
No bullet without a verb
A noun-phrase bullet is a label, not an argument. Verbs commit. The corpus average list has 4 bullets; trim until each one moves something.
67% of slides ship lists - 02
One callout per slide. Maximum.
Two callouts split attention; three cancel. The corpus runs four lists for every callout — exactly the wrong ratio. Invert it.
4× lists vs callouts - 03
Quote only when the source is the argument
An attributed quote imports credibility. Self-attributed (or unattributed) quotes import nothing. Cut to zero before adding a vanity one.
3% of slides ship quotes - 04
Paragraph beats list when the order matters
Bullets imply parallelism. Sentences carry transitions, qualifiers, and causal chains. If your bullets need to be read in order, write a paragraph.
37% of slides use paragraphs
01 / 13