Where we are in the Cadence campaign
In b1 you learned what AI can and cannot be trusted with. In b2 you nailed Cadence's positioning - who it is for, the real pain, the value prop. Tonight that positioning becomes content. This is the session where AI starts to feel like leverage: you will draft a full pillar article section in minutes. It is also the session where AI most tempts you into slop, so we build the guardrails at the same time as the speed. Same two rails as always: you own the brand and the claim, you protect the data.
The content engine: brief in, on-brand draft out 6 min live
AI can multiply your output overnight. But volume without a brief, without your voice, and without a fact-check is just faster sludge - and sludge erodes the brand you are trying to build. The trick is not "prompt harder". It is a pipeline where AI does one job well and you own the rest.
LiveWhy "just generate more" backfires3 min▶
Point AI at "write me ten blog posts" and you get ten posts that read like everyone else's ten posts - because the model reaches for the most average phrasing every time. Average is the enemy of a brand. The pipeline exists to move each piece from average-and-fast to distinctive-and-still-fast.
- Brief first, always. The single biggest quality lever is the brief - audience, angle, the one true proof point, the offer. No brief, no engine, just a blender.
- AI drafts, you don't outsource judgement. The draft is a starting block, not a finish line. It gets you to 70% in minutes so your time goes to the 30% that carries the brand.
- Voice and facts are non-negotiable stops. Skipping the voice edit gives you generic copy; skipping the fact-check gives you a fabricated claim. Neither ships.
Self-studyWhat each step actually produces2 min read▶
| Step | What comes out of it |
|---|---|
| Brief (you) | Audience, angle, one true proof point, offer, must-avoid claims |
| Draft (AI) | A fast 70% first draft grounded only in the brief you gave |
| Voice edit (you + AI) | The draft rewritten to sound like your brand, not the internet |
| Fact-check (you) | Every claim sourced or cut - nothing invented survives |
| Publish (you) | Copy you would put your name on, structured for the reader |
Notice AI touches exactly one row. That is deliberate. The moment you let it own the brief or the fact-check, you are back to sludge.
Brand voice + the anti-slop edit 6 min live
Out of the box, AI writes in "content marketing beige" - fluent, confident, and instantly forgettable. You fix that two ways: teach it your voice with real samples, and run an edit pass that hunts down the tells that scream "a robot wrote this". Both are quick once you know the moves.
LiveTeach voice with 2-3 real samples (few-shot)3 min▶
You cannot describe your voice into existence with adjectives - "make it punchy and warm" gets you nothing. You show it. Paste two or three short pieces of real, on-brand writing and tell the model to match their rhythm, vocabulary, and attitude. This is called few-shot prompting, and it is the difference between generic and unmistakably you.
- Pick samples that already sound right. A great past email, a landing section, a founder note. Two or three is enough; more just dilutes.
- Name what makes them yours. Short sentences? Dry humour? No jargon? Say it, so the model knows what to copy versus what was incidental.
- Save the winning prompt. Once it matches your voice, keep that prompt block. It becomes your reusable brand-voice header for every future content task - you will build one tonight.
LiveThe tells that scream "AI wrote this" - and how to cut them3 min▶
Readers have gotten good at spotting AI copy, and nothing tanks trust faster. The tells are predictable, which means they are fixable. Run this table as a checklist on every draft before it goes near a publish button.
| The tell | The fix |
|---|---|
| Generic throat-clearing intro ("In today's fast-paced world...") | Cut the first paragraph. Open on the reader's actual problem or a concrete moment |
| Listicle sameness - every section the same shape and length | Vary rhythm: a one-line punch, then a longer point. Break the pattern on purpose |
| Hedging ("it is important to note", "can potentially help") | Say it straight. Delete the qualifiers or make a real claim you can back |
| Empty adjectives ("robust", "seamless", "cutting-edge") | Replace with a specific, true detail. Show the thing instead of praising it |
| Any stat or quote with no source you supplied | Source it or cut it. This is the fact-check pass - no unsupported claim ships |
The blog that got quietly unpublished. A brand shipped fifteen AI-drafted posts in a week to "win at SEO". Readers noticed the sameness, a competitor screenshotted the "in today's fast-paced world" openers, and the team pulled the lot. The anti-slop pass takes ninety seconds per piece. Skipping it cost them a month.
Structure for the reader (and the algorithm) 4 min live
Great copy that is a wall of text still fails. How you structure a piece serves two audiences at once - the human skimming on a phone, and the search and answer engines deciding whether to surface you. The good news: what helps one helps the other.
LiveOne idea per section, skimmable, clear hierarchy2 min▶
Structure is not decoration - it is how meaning survives a busy reader. AI is genuinely good at restructuring, so once you have a voice-checked draft, ask it to organise for skimming.
- Clear heading hierarchy. One H1, descriptive H2s a skimmer could read alone and still get the gist. Each H2 owns one idea.
- One idea per section. If a section argues two things, split it. Tight sections are easier to read and easier for answer engines to quote.
- Skimmable by design. Short paragraphs, the occasional list, a bolded takeaway. Front-load the point; do not bury it in paragraph four.
Draft a Cadence pillar-article section ★ 13 min · everyone builds
Take the b2 positioning brief and turn it into a real section of a Cadence pillar article - grounded, on-brand, and with every unsupported number flagged instead of faked. This is the whole pipeline in one build.
Open any chat AI. Paste your Cadence positioning brief from b2: who it is for (consultants and team leads in back-to-back meetings), the pain (decisions and action items lost between calls), the true proof point (notes drafted the moment a call ends), the offer (free solo plan, paid Teams plan).
Add the brand-voice header from Part 2 - your two or three real samples plus the voice description. If you have no Cadence samples, use two short pieces you like the sound of and adapt.
Ask for one section of a pillar article - say, "How teams lose decisions between meetings". Enforce the rule: use only the facts in the brief, and mark any stat or claim you were not given as [need source] rather than inventing it.
Read the draft against Part 2. Does it open on the problem, not "in today's world"? One idea per section? Any beige adjectives? You have a 70% draft; the next demo is the 30%.
Note every [need source] tag. Those are your homework - real numbers you will go find or claims you will cut. Nothing ships with an unfilled tag.
Run an anti-slop edit pass ★ 8 min · everyone builds
Take the draft from demo one (or any AI draft you have) and run the edit pass: strip the tells, tighten the rhythm, and fact-check every claim. This is the 30% that turns 70% into shippable.
Paste the draft back in with the tells table from Part 2. Ask the AI to flag - not fix yet - every generic intro, hedge, empty adjective, listicle-sameness section, and any stat without a source.
Now ask it to rewrite, cutting the throat-clearing opener, varying sentence rhythm, and replacing empty adjectives with specific true details from the brief. Keep your brand-voice header in play.
Fact-check by hand. Go claim by claim. Every number, comparison, or "customers say" either gets a real source you own or gets cut. A [need source] tag with no source does not become a footnote - it becomes deleted text.
Read it out loud. If a sentence sounds like a brochure, rewrite it. If it sounds like your best colleague explaining the thing, it is done.
The two-pass habit that saved a launch. A team ran flag-then-rewrite on every launch-week asset. The flag pass caught three invented stats and a competitor comparison nobody could back. Ten minutes of editing kept a fabricated claim off the homepage the week the press was watching. The rails are not bureaucracy - they are what lets you move fast without the headline.
Try it yourself - this week ◐ 30-45 min total
- Finish the Cadence pillar section from demo one, then run the anti-slop pass on it. Fill or cut every [need source] tag before you call it done.
- Build your real brand-voice header: pick two or three genuinely on-brand samples from your own work and write the voice description. Save it where you can paste it every time.
- Run the tells table on one existing piece of your live content. Count the tells you find - that number is why this session exists.
- Take one AI draft from your week and do a by-hand fact-check. Note how many claims had no real source. Bring that number to b4.
- Optional: outline the full Cadence pillar article (one H1, five or six H2s, one idea each). You will turn those H2s into a topic cluster in b4.
Official sources covered
Taught from public AI-marketing curricula and vendor documentation. Certificates and paid tooling stay with the official programs. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · Your team can now produce ten times the content. Why is "just generate more" still a bad plan?
AI multiplies whatever you point it at. Without a brief, voice, and fact-check, it multiplies average, generic, sometimes-fabricated copy - which damages the brand and can ship false claims.
2 · What is the reliable way to make AI write in your brand voice?
Adjectives do not transfer a voice; examples do. Showing the model two or three real samples (few-shot) teaches it your actual rhythm and vocabulary far better than any description.
3 · Your AI draft has a confident statistic you did not supply. What is the fact-check rule?
Rail one. Every claim is sourced or cut - never invented, never dressed up with a fake citation. A confident stat you did not supply is a fabricated claim until you prove otherwise.