How this track works
Ten sessions, all hands-on, all building one real thing: the launch campaign for Cadence, a sample AI note-taker with a free solo plan and a paid Teams plan, so you practise both B2C and B2B motions on one artifact. Tonight you learn what the tool is; then b2 nails positioning with a live simulator, and b3-b10 build the content, SEO, ads, email, social, creative, and measurement of a full campaign. No code, ever. Two rails run through all ten, starting now.
What an LLM actually is - for a marketer 6 min live
A large language model is a very good next-word predictor trained on a huge slice of text. That one fact explains its magic and its danger for marketing: it is a brilliant first-draft, ideation, and rewriting engine, and an untrustworthy source of facts, stats, and claims. Knowing which is which is the whole game.
LiveWhy it invents stats - and why that is a brand risk, not a quirk3 min▶
The model generates the most likely-sounding continuation, not the true one. Ask it for "three statistics proving marketers love our category" and it will produce three confident, well-cited, completely invented numbers - because plausible stats are a likely continuation. In marketing that is not a curiosity; it is a fabricated claim that can reach thousands and land you in front of a regulator.
- It is probabilistic, not factual. No lookup happens unless you supply the source. Fluency tells you nothing about truth.
- Invented stats and fake sources are the classic trap. "73% of buyers say..." with a real-looking citation that does not exist. Never publish a number the model produced from memory.
- The fix is grounding, not scolding. "Be accurate" does nothing. Giving it your real research, brief, and proof points does almost everything.
The campaign built on a stat that never existed. A team asked a chatbot for "a punchy market stat" and dropped the confident figure straight into an ad and a landing page. It was invented. The ad ran, a prospect asked for the source, and the whole campaign had to be pulled and corrected. An ungrounded marketing claim is a liability with good formatting.
Self-studyThe can / cannot table - pin this to your monitor2 min read▶
| Trust it to... | Never trust it to (without a source + check) |
|---|---|
| Draft and rewrite copy from your brief | Produce stats, market data, or citations from memory |
| Generate angles, hooks, and variants to choose from | Make claims about your product's performance |
| Summarize research and restructure it | Quote a customer or a review that does not exist |
| Tailor one message to different segments | Know anything after its training cutoff |
| Draft outlines, briefs, and calendars | Be the source of truth for a legal or comparative claim |
Everything on the left is a build-along later in this track. Everything on the right is why every session pairs AI drafting with your sources and your review.
The marketing AI tool landscape, 2026 5 min live
"Which AI tool should we use?" is the wrong first question. The right one is: is this tool safe for the data I am about to put in it, and does it protect our brand? There are three layers, and the line between them is about data and control, not raw quality.
LiveGeneral vs marketing-native vs in-suite3 min▶
Each layer has a job. Do not pay for a specialist tool to do what a general assistant does well, and do not push confidential data through a consumer tool to save a click.
- General assistants (ChatGPT, Claude, Gemini, Copilot) are the flexible workhorses for drafting, ideation, and reasoning over a brief you paste in.
- Marketing-native tools (Jasper, Copy.ai, HubSpot AI, Canva) add brand kits, templates, and workflow - useful when consistency and speed at scale matter.
- In-suite copilots live in your CMS, email platform, ad manager, and CRM - convenient, and governed by that vendor's data terms. Read them.
Self-studyThe 2026 landscape at a glance2 min read▶
- General assistants: ChatGPT, Claude, Gemini, Microsoft Copilot - drafting, ideation, analysis-over-supplied-material.
- Marketing-native: Jasper, Copy.ai, Writer, HubSpot AI, Canva Magic - brand-kit consistency and templated production.
- In-suite copilots: in HubSpot, Klaviyo, Meta/Google Ads, Salesforce - AI where campaigns already run.
- The trend: 2026 curricula treat AI as a co-pilot across the whole funnel - research, positioning, content, SEO, ads, email, social, analytics - not a single magic button.
The two rails: own the brand, protect the data 4 min live
Everything else in this course is technique. These two rails are non-negotiable, and they are what separate a marketer who gets leverage from AI from one who becomes a cautionary headline.
LiveRail 1 - you own the brand and the claim2 min▶
AI drafts; you own every word, claim, and image that goes out under your brand. That means no fabricated stats, no fake reviews or testimonials, no unverifiable comparative claims - and clear disclosure where the law requires it. The FTC (US) and ASA (UK) both require that AI-generated endorsements and material connections be disclosed and that ads not deceive.
The fake-review crackdown. Regulators have fined brands for AI-generated and incentivized reviews presented as genuine. The marketing equivalent of the finance scandal: it is fast, it is cheap, and it is exactly the kind of shortcut AI makes tempting. Rail one exists so your credibility - the one thing AI cannot manufacture for you - stays intact.
LiveRail 2 - protect the data (privacy + consent)2 min▶
Personalization is powerful and legally loaded. Use consented first-party data, respect GDPR and PDPA, and never paste customer PII or unreleased plans into a consumer AI tool. Great targeting that breaks privacy law is not a win.
- Consent is the foundation. Personalize with data people agreed to share, for the purpose they agreed to.
- Keep confidential data in safe tools. Customer lists, unreleased launches, and contracts go in enterprise or approved tools only (rail one of the data world).
- Design for it. Every workflow you build in this track has a human review step and a "would I be comfortable if this were public" check.
Slop vs grounded - watch the fabrication die ★ 12 min · everyone builds
The single most important habit in the course, in one experiment. Ask for the same marketing asset twice - once from the model's memory, once grounded in a real brief - and watch generic, invented copy become something you would actually ship.
Open any chat AI. Pick a safe subject: a public product, or our sample product Cadence. Keep it non-confidential (rail two).
Ungrounded ask: "write a landing-page hero with a strong stat" - no brief. Note how generic the copy is and how the "stat" is invented.
Grounded ask: paste a short brief - who it is for, the real pain, one true proof point, the offer - and ask again, adding "use only the facts in this brief; if you need a stat I did not give you, say [need source] instead of inventing one".
Compare. The grounded version speaks to a real person, uses only claims you can back, and flags gaps instead of faking them. That delta - slop to shippable - is the value of everything that follows.
Now stress it: remove the proof point and re-run. A good prompt makes it write "[need source]" rather than invent. That is the behaviour you will engineer on purpose in b2 and b3.
Write your team's AI marketing ground rules ★ 10 min · build your own
A reusable artifact you will actually keep: a short "how our marketing team uses AI" card. You draft it with AI, then edit it - which is rail one in miniature.
Give the AI the two rails and your context: team, the tools you have (or have not) approved, and the data you handle (customer lists, unreleased launches, brand assets).
Ask for a one-page card with three lists: Green light (safe uses + which tool), Red light (never: fabricated stats, fake reviews, PII in consumer tools), and Always (verify claims, disclose where required, human sign-off, stay on brand voice).
Edit it hard. Add your actual brand-voice rules, tool names, and approvers. The AI draft is 70%; your judgement is the 30% that makes it real.
Share it with the team. You will refine it again in the leader session a3 (brand + disclosure) and the builder finale b10 (governance at scale).
The card that unblocked a team. A content team was split - half afraid of AI, half quietly pasting client data into free tools. A one-page card agreed in half an hour replaced both fear and free-for-all with a shared green light. Output went up and risk went down, because "what's allowed" stopped being a guess.
Try it yourself - this week ◐ 30-45 min total
- Finish both demos if you did not complete them live - especially slop-vs-grounded. Feeling the fabrication yourself beats being told about it.
- Run the ungrounded prompt on three real asks from your week. Screenshot the invented stats - that is your reminder to always bring your own facts.
- Finalise your one-page AI marketing ground rules and share it with a colleague for a sanity check.
- Confirm which tools you are cleared to use for real customer data. If the answer is "none yet", that is a great thing to raise - the leader session a4 helps.
- Optional: skim your brand-voice guide (or note that you need one). Bring one voice rule to session b2.
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 · You ask a chatbot for "a strong market statistic" with no source. It gives a precise, cited number. What do you know?
An LLM predicts likely text, not true facts. A confident stat with no supplied source is a fabricated claim - a brand and legal risk. Bring your own facts.
2 · You need to personalize an email using your customer list. Which tool is acceptable?
Rail two. Customer PII goes only in approved tools, and personalization uses consented data under GDPR/PDPA. Great targeting that breaks privacy law is not a win.
3 · What is the reliable fix for AI that produces generic, invented copy?
Scolding does nothing; a real brief plus a "no invented claims" rule does almost everything. Ground it, forbid fabrication, and keep a human edit. That is the whole track in one line.