slug: action-titles view: skill layer: slide agent: architect companion: corpus/storymakers/frameworks/slide/action-titles.md audience: llm

Action titles — operational reference

Operational chuleta for an LLM writing or auditing slide titles. The showcase (action-titles.md) carries the argument; this file carries the rules, recipe, anti-patterns, and the action-title test.

One sentence. Every slide title is a declarative sentence with a verb and an insight; the titles in sequence reconstruct the executive summary.

Use this when

Deck is a decision artefact — recommendation, board update, investment memo, M&A review, post-mortem.
Reader is being asked to agree, decide, or act.
Deck has a single governing thought (a pyramid sits beneath).
Slides will be read at speed by senior readers scanning the title bar.

Use something else when

Divider, agenda, section-break slides — function is navigation; force a verb and you get nonsense ("Section 2: we now turn to operations").
Training / educational decks — the title is the topic.
Workshop discussion decks — slides provoke, don't conclude.
Reveal-driven keynote / pitch decks — the deck withholds the punch line on purpose.

Decision tree

Q1. Does the title contain a finite verb?
    NO  → It's a label. Rewrite. STOP and re-enter at Q1.
    YES → Q2

Q2. Does the title state a claim, not just a fact?
    NO  → It's a caption. Add the so-what. Re-enter at Q2.
    YES → Q3

Q3. Is the claim defensible from this slide's chart / body?
    NO  → Title overreaches. Soften, or fix the analysis.
    YES → Q4

Q4. Does the title ladder up to the deck's governing thought?
    NO  → Slide is off-pyramid. Cut the slide or fix the structure.
    YES → Q5

Q5. Action-title test — read this title in sequence with its
    neighbours. Does the sequence still read as a story?
    NO  → Order or content is wrong. Reorder or rewrite neighbours.
    YES → Ship.

Recipe (per deck — ~60 min for a 12-slide deck)

Governing thought (5 min) — one declarative sentence with a verb.
Supports (10 min) — three to five claims that prove it; each becomes a section.
Title-only outline (the ghost deck) (15 min) — every slide gets a one-sentence title before any body is built.
Action-title test (5 min) — if the title sequence doesn't reconstruct the executive summary, fix titles, not bodies.
Build the bodies (20 min) — each chart, table, list is evidence for the title above it, never the reverse.
Final pass (5 min) — verb, voice, one claim per title; no multi-clause and / but.

Anti-patterns (reject on sight)

Pattern	Why it's wrong
"Volume by quarter"	Topic label. No verb, no claim.
"Key findings" / "Summary of analysis"	Meta-label. The findings are the title.
"Margins have been impacted"	Passive evasion. Name the agent and the magnitude.
"Revenue grew 12%" (no driver)	Caption, not a claim. Add the so what.
"Pricing strategy"	One-noun title. Always wrong.
Title overreaches the chart	Worse than topic — broadcasts unsupported conclusion.
"X grew, but Y fell, and Z held"	Three claims welded. Split the slide.
"X drove Y" twelve times in a row	Mechanical. Vary verbs.
Slide reusable verbatim on a different company	Title is generic. Rewrite for the specific finding.

Verb bank (defensible, ships in real decks)

magnitude — grew, declined, fell, climbed, rose
attribution — drove, caused, contributed, accounted for
value — delivered, generated, captured, returned
gaps — narrowed, widened, closed, opened
trajectories — recovered, eroded, compressed, expanded
guidance — exceeded, missed, met, beat
distribution — concentrates in, splits across, falls between
call-to-action — recommends, requires, warrants

Output checklist (the action-title test)

Every title is a declarative sentence with a finite verb.
Every title states a claim, not a topic.
Every claim is defensible from the slide's evidence.
Active voice; one claim per title; verbs vary across the deck.
Titles in sequence reconstruct the executive summary.
Every title ladders up to the governing thought.
The reader can extract the argument from the title bar alone.

Canonical signals (for matchers)

Component kind: title — the discriminator is sentence vs label, not slot.
Keywords: action title, key takeaway, in summary, drove, delivered, grew, declined, recovered, recommendation.
Structural signal: title is ≥6 words, contains a finite verb, is unique within the deck (not a recurring section header).
Anti-signal: title is one or two nouns, ends without a predicate, or repeats verbatim across multiple slides.

References

Minto, The Pyramid Principle, 1973 — the underlying logic.
Zelazny, Say It with Charts, 1985 — slide-craft canon.
Zelazny, Say It with Presentations, 2000 — deck-level extension.
Rasiel, The McKinsey Way, 1999 — house-style restatement.
Figliuolo, The Elegant Pitch, 2016 — modern playbook.