AI can generate content. It cannot create conviction. This course teaches you to tell the difference, in what you receive and in what you send.

It is a practice course. Every module ends with an exercise using your real communication: real inbox, real drafts, real stakes. You finish with a tested framework, a pressure-testing routine for AI drafts, and one concrete habit installed on your calendar. Use Download my workbook in the sidebar to keep everything you write here.

The problem flipped

For most of professional history, producing communication was the hard part. Writing a solid memo took hours. Building a deck took days. Scarcity of output was the constraint, so effort was a reliable signal: if someone sent you a polished document, someone had thought hard about it.

AI removed the production constraint. Anyone can now generate a polished-looking memo, summary, or deck in minutes. The volume of professional communication is rising while the effort behind any given piece of it has become unknowable. Polish no longer proves thinking.

That flip changes the core skill. When production was scarce, the valuable skill was producing. Now that production is abundant, the valuable skill is filtering: deciding what deserves attention, and making your own messages worth the attention you are asking for.

The evidence: noise has a price tag

This is measurable, not a vibe.

Research from BetterUp Labs and the Stanford Social Media Lab surveyed 1,150 U.S. desk workers and coined a name for the incoming problem: workslop, AI-generated work that looks polished but lacks the substance to move a task forward. About 40 percent of workers reported receiving it within the past month, and each instance took nearly two hours to sort out, an invisible tax estimated at roughly $186 per employee per month.

The social cost ran deeper than the hours. Roughly half of recipients viewed workslop senders as less capable and less reliable, and 42 percent said they trusted those colleagues less. Sending noise does not just waste the reader's time. It spends the sender's credibility.

On the receiving side, the 2026 Workplace Communication Survey from Korbyt and Reworked found that 81 percent of workers say they can usually tell when a message was created with AI, 45 percent question a message's accuracy when they suspect AI was involved, and 44 percent report tuning out workplace communication even when the volume feels manageable. When asked what they actually want AI to do, 92 percent agreed: reduce information overload rather than add to it.

Read those numbers together and the picture is clear. Your audience is already filtering, already detecting generic output, and already discounting it. The question is which side of their filter your messages land on.

Why this raises the value of your voice

Here is the counterintuitive part: an AI-saturated environment makes distinctly human communication more valuable, because it is scarcer relative to everything else in the inbox. A message with a clear point of view, specific facts, and a recognizable voice now stands out the way a handwritten note stands out in a stack of form letters. The skill this course builds is deliberate scarcity: putting the thinking into your messages that the generic flood conspicuously lacks.

You need one framework that evaluates communication whether you wrote it, an AI wrote it, or a colleague sent it. The Signal Test has four parts. A message that passes all four is worth someone's time. Each failed part tells you exactly what to fix.

Part 1: Stakes

Signal names its consequence. Noise circulates because circulating is easy: FYI messages with no F worth Y-ing, updates that update nothing.

Noise: "Sharing an update on the vendor evaluation process. We continue to make progress across multiple workstreams and will keep the group posted as things develop."

Signal: "Vendor decision is due Friday. We are down to two options, and the open question is whether we accept a 6-week implementation delay for a 15 percent lower cost. I need your call on that trade-off by Thursday."

Part 2: Specificity

Signal contains claims concrete enough to be checked and, in principle, wrong: names, numbers, dates, positions. Noise is unfalsifiable, and it would read the same in any company, any quarter. Unfalsifiable content is one of the most reliable markers of low-effort AI output, because generating safely vague text is what an unbriefed model does best.

Noise: "The initiative is well positioned to drive meaningful improvements in operational efficiency across the organization."

Signal: "The new intake form cut duplicate data entry for admissions staff from 4 systems to 1. Early result: about 20 minutes saved per admission, based on the first 30 admissions at two communities."

You can argue with the second version. That is the point. Arguable claims are the ones that carry information.

Part 3: Stance

Signal takes a position and absorbs the risk of being wrong. Noise is symmetrical: considerations on the one hand and the other, ending at "it depends," transferring all the thinking to the reader. Balance reads as thoroughness but often functions as hiding.

Noise: "There are several factors to weigh. Option A offers cost advantages while Option B provides greater flexibility. Both approaches have merit depending on organizational priorities."

Signal: "Recommend Option A. The cost difference funds the training budget we cut in March, and the flexibility Option B offers solves a problem we do not currently have. The main risk is vendor lock-in, and I think the 3-year contract term makes it acceptable."

A reader can act on the second version in one reply. The first version schedules a meeting.

Part 4: Signature

Signal sounds like a specific person who knows the reader, shares history with them, and writes the way they talk. Since 81 percent of your audience believes they can detect AI-generated messages, generic register does not read as neutral. It reads as "the sender did not consider this worth their own time," and the message gets discounted before its content is weighed.

Signature lives in small things: referencing the conversation you actually had last week, using the shorthand your team uses, acknowledging the objection this specific reader raised last time, sounding the way you sound.

Scoring and using the test

Score any message 0 to 4, one point per part. As a reader, a 0 or 1 gets minimal attention whatever its polish. As a writer, any failed part is your revision instruction. As an editor of AI drafts, the test is your quality gate. Note what the test does not measure: length, formatting, or vocabulary. A two-line message can score 4. A ten-page deck can score 0. Most workslop scores 0 to 1 while looking like a 4.

As reader: the triage protocol

Your attention is the resource everyone else's noise is spending. Protect it with a standing protocol:

1. Score before you study. A 20-second scan against the Signal Test comes before any deep read. Stakes and stance are usually visible in the first four lines. If they are missing there, they are usually missing everywhere.
2. Route by score. 3 to 4: read fully. 2: read for the signal part, skim the rest. 0 to 1: do not absorb the cost silently. Reply with the question that forces the missing part: "What do you need from me on this?" (missing stakes) or "What is your recommendation?" (missing stance).
3. Train your senders. The workslop cost lands on recipients because recipients absorb it quietly. Consistent, polite score-forcing questions change what people send you within weeks. You are not being difficult. You are declining to complete other people's unfinished work.

For suspected workslop, the sharpened version: "Before I dig in, can you give me the three-line version of what this concludes and what you want me to do with it?" Anyone who did real thinking answers easily.

Use AI on the incoming side too. Where volume is the problem, AI is a legitimate filter:

Read this and answer three questions only:
1. What does it conclude?
2. What does it want from me, specifically, and by when?
3. What verifiable claims does it make (numbers, dates,
   names, commitments)?
If any answer is missing from the document, say
"not stated" rather than inferring one.

As writer: the pre-send pass

Now flip the filter outward. Before sending anything with stakes, run this pass. With practice it takes three minutes.

1. State the stakes in line one. The reader should know within four lines why this message exists and what it affects.
2. Replace one abstraction with one specific. Find your vaguest claim and make it checkable.
3. Commit to a stance. If the message contains options, it also contains your recommendation and your reason. If you genuinely cannot recommend, say precisely what information would change that, which is itself a stance.
4. Add one line only you could write. One authentic line shifts the register of an entire message.
5. Cut a fifth. Then read it aloud once. Anything you would not say to the person's face gets rewritten in words you would.

Worked example

The scenario: you must tell a leadership group that a system rollout is slipping six weeks.

Before the pass (scores 1 of 4):

Subject: Rollout Update

Hi all, sharing a status update on the rollout. The team has been working through several implementation challenges, and while significant progress has been made, we are adjusting the timeline to ensure a successful outcome. We remain committed to delivering a high-quality result and will continue to keep everyone informed as the plan evolves. Please reach out with any questions.

The right relationship with the tool

AI is a strong editor, summarizer, challenger, and drafting accelerant. It is a poor replacement for your judgment, your stance, and your knowledge of the reader, and audiences penalize output where the replacement shows. The failure pattern behind most workslop is sequence: generate first, think later, and later never comes. The fix is to reverse the sequence. Thinking first, generation second, pressure-testing third.

Step 1: Brief before you generate

Never open with "write an email about X." Brief the AI the way you would brief a capable new chief of staff:

Draft [message type] to [audience].

The stakes: [why this matters to them, what it affects]
My position: [what I recommend or have decided, and why]
Must include: [the specific facts: numbers, dates, names]
This reader: [what they care about, what they pushed
back on before, how formal we are with each other]
My voice: [how you write. Or paste your identity file
if you built one in the Second Brain course.]

Notice what the brief forces: you cannot fill it in without having done the thinking. The brief is the work. The generation is the typing.

Step 2: Pressure-test the draft

Score this draft against four criteria: Does it state
consequences for the reader? Does it make specific,
verifiable claims? Does it take a clear position? Does
it sound like a specific person rather than anyone?
Identify the weakest of the four and show me exactly
which sentences fail it.

You are the most skeptical person in the room reading
this. What would you push back on, what would you not
believe, and what question would you ask that this
draft cannot answer?

Cut this by a third without losing any specific claim,
any part of the recommendation, or the reasoning.
Flag every sentence you cut because it carried no
information.

The compressor's flag list deserves attention. Sentences that cut without loss are the generic filler your readers have learned to skim past. Seeing your own filler listed is the fastest way to stop producing it.

Step 3: The ownership pass

Last step, no AI involved. Read the draft aloud and change every phrase you would not say. Add the one line only you could write. Confirm the recommendation is one you will defend when someone pushes back, because someone will, and "the AI suggested it" is not a stance.

A useful ownership standard: send nothing you could not summarize from memory, without the document, in three sentences. If you cannot, you have not finished thinking, and your reader will be the one to discover that.

The red-flag list

Markers of the generic register, in AI drafts and human ones. One instance is style. A cluster is a message nobody thought about.

• Filler openers: "I hope this finds you well," "I wanted to reach out," "In today's fast-paced environment"
• Importance claims without content: "significant," "meaningful," "robust," "impactful" with no number or example attached
• Symmetrical hedging: every point paired with its counterpoint, no resolution
• Suspiciously tidy structure: three perfectly parallel points for everything
• Praise inflation: "great question," enthusiasm disconnected from anything specific
• Conclusions that conclude nothing: "moving forward, continued collaboration will be key"

Across the research and across organizations, what breaks through is remarkably consistent. Every winning pattern below is a Signal Test part doing its work in the wild.

• Landing: short messages with the decision up front. What was decided or what is needed in the subject line and first sentence, reasoning below for those who want it. It wins because it respects the 20-second triage everyone now runs. Getting tuned out: the long update where the point is buried in paragraph four, which readers correctly treat as optional.
• Landing: AI used to reduce, disclosed when it matters. The survey data is blunt: 92 percent of workers want AI used to cut information volume, and trust tracks the sender's identity and consistency. Getting tuned out: AI used to multiply output, and the broadcast update sent to everyone because sending to everyone became free. Recipients read untargeted distribution as "the sender did not think about who needs this," and they tune out the channel, not just the message.
• Landing: named trade-offs and owned misses. Messages that state what was given up, and updates that lead with the miss and the correction, consistently outperform success-only communication on the measure that matters: whether people believe the next message. Getting tuned out: the all-wins update, which readers discount as marketing regardless of accuracy.
• Landing: the human line. In change communication especially, one sentence of genuine, specific acknowledgment ("I know this is the second system change this year, and the last one cost you two weekends") does more than paragraphs of framing. It cannot be faked by register alone, because it requires actually knowing what the reader has been through.

The meta-lesson: none of the winning patterns require writing talent. They require the thing AI made scarce: evidence of thought. Stakes, specifics, a stance, and a signature are thinking made visible.

The habit, precisely defined

Install it now: add the five steps as a saved note, a text snippet, or a checklist wherever you write. The habit survives if the checklist is one keystroke away and dies if it is not.

What you built in this course

• A noise audit of my inbox, with a triage protocol to manage it
• The Signal Test: Stakes, Specificity, Stance, Signature
• A scored, revised, ready-to-send version of a real message
• A briefing template and pressure-test prompts for every future AI draft
• The red-flag list for detecting the generic register
• The three-minute signal pass, saved where I write

The next 30 days

• Days 1 to 7: install the pass. Run it on every message that matters. Slow at first, automatic by Friday. Send the message you built in this course.
• Days 8 to 14: fix your gap. Every message this week gets checked against the Signal Test part you fail most.
• Days 15 to 21: train your inbox. Use the score-forcing questions on incoming noise. Politely, consistently. Watch what people send you change.
• Days 22 to 30: adopt your pattern. Implement the landing pattern you chose in Module 5. Then pull three messages you sent this month and three from before the course, score all six, and keep the difference as your evidence.

The one-month test

One question at day 30: do the people you communicate with respond faster, push back more usefully, and ask fewer clarifying questions? Those three outcomes are what "cutting through" actually measures.

Every prompt is tool-agnostic and copy-ready. Replace anything in [brackets].

Filtering incoming

Assess this document for substance. Identify:
1. Claims that are specific and checkable
2. Sentences that carry no information and could be
   deleted without loss
3. Whether it contains an actual position or
   recommendation, and if so, what it is
Give a one-line verdict: is this worth a full read,
a skim, or a reply asking for the point?

From this thread, extract only:
1. Decisions made, with who made them
2. Open questions and who owns each
3. Action items with owners and dates
4. Anything directed at me specifically
Ignore pleasantries and repetition. Flag any point
where participants contradict each other.

Extract from these notes in three short bullets each:
decisions made, action items with owners, and open
questions. Nothing else.

Briefing

Draft a short [email / chat message] to [person].
Point: [the one thing it must land]
Ask: [what I need from them and by when]
Tone: [how we talk to each other]
Four sentences maximum. Ask first, context after.

Draft a message delivering this news: [the news, stated
plainly, with the real numbers and dates].
Structure: the news in line one, the cause in one
sentence, my recommendation, then the one decision I
need from the reader and by when.
No softening preamble. Acknowledge this specific
impact on the reader: [what it actually costs them].

Revising

This draft hedges. Rewrite it to commit: pick the
strongest supportable version of each claim, state my
recommendation plainly, and move every "on the other
hand" into a single short risks section with my
response to each risk.

Remove all of the following from this draft: filler
openers, importance words with no number or example
attached ("significant," "robust," "impactful"),
praise inflation, and closing sentences that conclude
nothing. Show me what was removed.

Here are two examples of how I actually write:
[paste two real messages you sent and liked].
Rewrite this draft to match that voice: same sentence
length, same level of formality, same kind of
directness. Do not change any facts or the
recommendation.

You are [role/name], who cares most about [their
priority] and previously raised [their objection].
You have 20 seconds. React to this message: what do
you take away, what do you skip, and what do you
reply?

Final gates

List every specific claim in this draft: numbers,
dates, names, quotes, and statements about what other
people said or decided. Present them as a checklist
for me to verify. Mark any that you generated or
inferred rather than took from my input.

Summarize this message in three sentences: the point,
the reasoning, the ask. If any of the three is
unclear from the draft, tell me which and why.

Standing instructions worth saving

When drafting communication for me:
- Put the point and the ask in the first four lines
- Never invent facts, numbers, or quotes; mark gaps
  as [NEEDED: description]
- No filler openers, no importance words without
  content, no conclusions that conclude nothing
- If I have not stated a position, ask me for one
  before drafting; do not write around the absence
- Default to shorter; flag any sentence that exists
  for tone rather than information