Core message extraction

Extraction is what happens before the title is written; the title is what's left when extraction is finished. Look at the chart, the analysis, the interview transcript — refuse to ship the slide until you can answer one question: if the audience reads one sentence on this page, which sentence? That answer is the core message. The action title is its public face; this canon is the private discipline that produces it.

Where it comes from

Three parents — a journalist, a soldier, and the brothers who tied them together.

The journalist's parent is the nut graf: the paragraph that tells the reader what this piece is about and why they should care. William Blundell's The Art and Craft of Feature Writing (Plume, 1988) is the canonical text — the Wall Street Journal's feature-writing seminars transcribed across two decades. Roy Peter Clark at Poynter later refined the term, arguing the nut graf is less a paragraph than the kernel — the sentence that crystallises the story's reason for existing. A reader who skips the lede should still hit the nut graf and know what the story is for. That is the slide-title problem one century older.

The soldier's parent is the U.S. Army's commander's intent, codified in FM 6-0 (2003) and carried into the current FM 3-0 Operations. Battle plans don't survive contact with the enemy, so every order is topped with "a clear, concise statement of what the force must do and the conditions the force must establish" — the one sentence subordinates execute against when the radio dies. "Capture Hill 4305" travels through the fog; the fifty-page annex does not.

The synthesis is Chip and Dan Heath's Made to Stick (Random House, 2007), chapter 1: "Simple — Find the Core". They open with commander's intent and close with Southwest Airlines' "the low-fare airline" — the same artefact in fatigues and in a 737. The rest of their model — the SUCCESs mnemonic (Simple, Unexpected, Concrete, Credible, Emotional, Stories) — only earns its place once the core is named. SUCCESs tests; extraction finds.

The distinction this canon lives on, beyond the three sources: finding the core is not summarising, it is prioritising. A summary shortens; extraction chooses.

What it actually is

Extraction is a process applied to raw content — analysis, data, interview notes, a chart someone built — that ends in one sentence. The process is four questions asked in order. Each either lets the sentence through or sends the author back to climb.

The four extraction questions, in order:

1. What's the one thing? Not "the chart shows revenue by quarter" — that is the topic. The answer has to be a sentence with subject, verb, and a defensible position: "Revenue growth has reversed; Q3 turned negative for the first time since 2019."
2. So what? Read it as a senior reader. If the honest reply is "so what?", the sentence is a fact, not a message. "Q3 revenue fell 4%" is a chart caption. "Q3 revenue fell 4%, breaking the eighteen-quarter growth streak that underwrites the current valuation" is a core message — the clause after the comma is what the question forced into existence.
3. What's the move? Complete "…so we should…". If nothing follows, it's trivia. The move can be "keep doing what we're doing" — that counts — but it must be writable.
4. Is this the highest defensible claim? Not the safest — the highest the evidence will carry. First drafts cluster at the timid end ("Customer satisfaction shows mixed signals"). Climb until the evidence pushes back: "Satisfaction has bifurcated — enterprise up twelve points, SMB down nine. The product is splitting in two." Extraction is finished when one more rung would force a hedge.

The output is one declarative sentence, ≤20 words. It then becomes the slide's action title (action-titles) — but the title is the artefact. Skipping the process and writing the title directly produces what looks like an action title and reads like a topic.

Why it works

The mechanism is cognitive economics: extraction pre-digests interpretation so the audience doesn't have to. The author has the data, the context and the time the readers don't — every minute spent climbing is a minute forty readers don't climb themselves.

Two failure modes extraction prevents:

• The topic title. "Q3 revenue performance" + a chart. The audience does the extraction itself; three people reach three conclusions. The slide has outsourced communication.
• The multi-thesis stapler. "Revenue grew, satisfaction declined, and we're launching in Q4." Three messages spliced; the audience holds none. Extraction forces which one is the one.

The deeper reason is WYSIATI — What You See Is All There Is (Kahneman, Thinking, Fast and Slow, ch. 7). System 1 builds a story from whatever fragment is loudest and treats it as the whole truth. Loudest fragment is a chart → chart-shaped impression. Loudest fragment is a clear sentence → they leave with the sentence. Extraction is the chance to be the loudest on purpose.

Extraction finds the core; the Heaths' SUCCESs mnemonic tests it (Simple, Unexpected, Concrete, Credible, Emotional, Stories). A message that passes the four questions but is abstract or stakes-empty will be correct and forgotten by Tuesday. The two steps are sequential, not alternatives.

When it's the right tool

Extraction is the right tool for any slide that carries an argument — recommendations, findings, charts, summaries, openings, closings. Anywhere the audience is meant to leave with a specific conclusion, extraction is the entry price.

It is the wrong tool for pure reference material (appendices, bibliographies, disclosures — looked up, not read) and genuinely exploratory decks where the answer hasn't crystallised yet. Faking a core message in that context is dishonest; mark the slide "open question" and move on.

Diagnostic: if the slide has a chart on it, it needs a core message. Charts are evidence; evidence requires a claim.

Worked example — a margin chart, walked through the four questions

The chart: a stacked-area decomposition of gross margin over eight quarters into mix, price, and cost. The shape — a 180-bp dip two years ago, a 220-bp recovery since; mix carries the recovery, price is flat, cost is a slow headwind. Placeholder title: "Gross margin trend, FY22–FY24". A topic. Extraction begins.

Q1 — one thing? "Gross margin recovered." The chart already says so; no stakes. Climb.

Q2 — so what? The recovery isn't the news; what drove it is — that decides whether the recovery survives FY25.

"Margin recovered 220 bps over two years, driven almost entirely by mix shift toward premium SKUs."

Still a chart with a longer caption. Climb.

Q3 — the move? Mix-led, price flat, cost a headwind. …so we should… defend mix and pre-empt cost.

"Margin recovery has been mix-led, not price-led; once mix stabilises, cost dominates."

Implies a move. Once more.

Q4 — highest defensible claim? Two more facts: premium mix sits at 41% against a 45–50% ceiling, and cost is accelerating, not linear. The next two years will not look like the last two.

"The 220-bp margin recovery is mix-led and almost spent: premium mix sits at 41% against a 45–50% ceiling, and cost is accelerating. FY25 margin will compress unless we act on cost now."

Verb ("will compress"), defensible (bands in the appendix), so-what (FY25 guidance), move ("act on cost"). The action title compresses to one sentence:

Slide title: "FY25 margin will compress: the mix engine is nearly tapped and cost is accelerating — act now."

Before extraction, the chart was the slide. After, the chart's job is to prove the title — same chart, inverted relationship.

Common pitfalls

Chart-caption titles. "Revenue fell 4% in Q3" — the chart already says this. Captions restate; messages interpret.

Multi-thesis "X and Y" sentences. "Revenue grew and satisfaction declined." Two messages spliced; the audience holds the louder one. Pick one.

Hedging adverbs that leak conviction. "Satisfaction may be showing some signs of possible softening." The hedges (may, some, possible) are negotiation with the data in public. Drop them, or the slide isn't ready.

Meta-titles. "This slide explores Q3 performance." The title is the conclusion, not the agenda. Explores, discusses, provides an overview of are the author writing about the slide instead of writing it.

Stopping at description before implication. "Premium mix rose from 28% to 41%." True, specific, fails the so-what. The implication ("and the engine is almost out of runway") is what turns description into a message.

Treating extraction as summarising. A summary shortens; extraction chooses. The summary of a twelve-column spreadsheet is a six-column spreadsheet. The core message is one of those columns matters and here is why.

Related canon

• action-titles — the artefact. Extraction is the process; the action title is what it outputs. Strong action title → extraction happened. Topic title → it didn't.
• so-what-test — question 2, isolated. Use as a spot-check on finished slides; full extraction when authoring.
• headline-test — the post-hoc validator. Asks whether a reader who saw only the title would walk away with the right point. Fails → re-run extraction.
• governing-thought — Minto's deck-apex. Slide-level core messages ladder up to it.
• big-idea-formula — Duarte's deck-level cousin to the governing thought.

Canonical references

• Chip Heath & Dan Heath — Made to Stick — Random House, January 2007. Chapter 1 ("Simple — Find the Core") is the primary source; SUCCESs runs through chapters 2–7. The companion HBR essay The Curse of Knowledge (December 2006) explains why experts cannot find their own core.
• William E. Blundell — The Art and Craft of Feature Writing — Plume, November 1988 (ISBN 978-0452261587). The WSJ feature-writing seminars transcribed; canonical primary source on the nut graf.
• Roy Peter Clark — The Nut Graf and Breaking News — Poynter Institute, 2003. A kernel, not a paragraph.
• U.S. Army — FM 3-0 Operations — HQDA, October 2022. Doctrinal home of commander's intent.
• Daniel Kahneman — Thinking, Fast and Slow — FSG, 2011. Chapter 7: WYSIATI.
• Barbara Minto — The Pyramid Principle — Pearson, 1973 (revised 2009). Deck-level governing thought.
• Nancy Duarte — Resonate — Wiley, 2010. Big idea as the deck-level analogue.