Where we are in the Cadence campaign
You have Cadence's positioning from b2 and its content engine from b3-b4. Now we spend money to get attention. Paid is where AI's "generate a hundred options" superpower is most tempting and most dangerous: variety is cheap, but a pile of untested variants is just noise. This session is about the discipline that turns volume into learning - and keeps every claim honest enough to survive a click.
AI's real job in paid: variety, structured for testing 6 min live
The value of AI in paid is not "better copy". It is cheap variety - many concepts and variants in minutes. But variety only pays off if you can tell what won. The whole craft is generating widely, then testing narrowly: change one thing at a time, or you learn nothing from the spend.
LiveWhy "more variants" is not a strategy3 min▶
AI makes it trivial to ship twenty ads. The problem is not making them - it is learning from them. If two ads differ in headline and image and audience, and one wins, you have no idea why, and you cannot repeat the win. Variety without structure is just spending faster.
- Variety is the input, not the goal. Use AI to explore the space of angles and phrasings cheaply, then narrow to a real test.
- One variable per test. Hold everything constant except the one thing you want to learn about. That is what makes the result readable.
- The platform will "optimize" for you - carefully. Auto-optimization is fine for scaling a known winner; it is not a substitute for a deliberate test when you are trying to learn something.
Self-studyWhat AI is good and bad at in paid2 min read▶
| Trust it to... | You still own... |
|---|---|
| Generate many angles, headlines, and hooks fast | Which angles are on-brand and true |
| Write copy variants for a chosen variable | The test design and the one variable that changes |
| Draft creative concept briefs for designers | The budget, the metric, and the decision rule |
| Suggest audience/targeting hypotheses | Validating them in-platform with real data |
Everything on the left is what you will build in the demos. Everything on the right is why AI drafts and you decide.
Copy, concepts, and audience drafts 5 min live
A paid campaign needs three kinds of asset, and AI can draft all three - if you feed it the positioning brief and forbid it from inventing proof. Copy variants per angle, creative concept briefs a designer can build from, and audience hypotheses you validate in-platform. None of these are final until you have checked every claim.
LiveThe three assets AI drafts for you3 min▶
Anchor everything in the b2 positioning: the angle is why this audience cares, and the copy is that angle in headline, primary text, and CTA form. Never let a variant carry a number you did not supply.
| Asset | Your discipline / tip |
|---|---|
| Headline variants per angle | Same promise, different phrasing - so a test compares wording, not claims |
| Primary text (body) variants | Only benefits you can back; mark any stat "[need source]" not invented |
| CTA variants | Match the offer stage: free solo plan vs Teams demo are different asks |
| Creative concept brief | Describe the idea for a designer; AI does not make your final image (b8 covers rights) |
| Audience / targeting hypotheses | Treat as guesses to validate in-platform, never as facts about your buyers |
The variant that shipped a fake number. A team let AI "make the headline punchier" and it added "saves 5 hours a week" - a figure no one had measured. The ad ran, a prospect asked where the number came from, and the account got a warning. Rail one: no claim ships that you cannot back.
Self-studyAngles vs variants - keep them straight2 min read▶
- An angle is a reason to care: "stop losing decisions between calls", "give your team one source of truth", "notes write themselves". These come from positioning.
- A variant is one angle expressed different ways: three headlines for the same angle, two CTAs, one image concept.
- Test angles against each other first (which reason resonates), then test variants within the winning angle (which phrasing performs).
- Keep a variant log. A simple sheet of angle -> variant -> what changed makes your results repeatable instead of lucky.
Test design that gives a clean read 4 min live
A test you cannot read is money spent to learn nothing. Four things make a test clean, and AI can draft the matrix but it cannot make these calls for you - because they involve your budget, your patience, and your risk.
LiveThe four things a clean test needs2 min▶
- One variable. Change exactly one thing between A and B - headline, or image, or CTA. Everything else identical.
- Enough budget and time. A test that ends before it gathers real signal is a coin flip. Decide up front how long and how much.
- One clear metric. Pick the metric that matches the goal: click-through for attention, sign-ups or demo-requests for conversion. Not five metrics - one.
- A decision rule set in advance. "If B beats A on sign-up rate by a meaningful margin after the run, we scale B." Write it before you look, so you are not tempted to read the tea leaves.
Self-studyCommon test mistakes to avoid2 min read▶
- Two variables at once. New headline and new image together - you will never know which one moved the number.
- Calling it early. Peeking after a day and declaring a winner on tiny numbers. Set the run length and honour it.
- Chasing a vanity metric. A high click rate that never converts is not a win. Match the metric to the actual goal.
- No decision rule. Without a rule set in advance, every result looks like whatever you hoped for.
Generate the Cadence ad set - 3 angles, real copy ★ 13 min · everyone builds
Turn the b2 positioning brief into a working ad set: three angles, each with headline, primary-text, and CTA variants - and not a single invented statistic. This is rail one applied to paid.
Open any chat AI. Paste your Cadence positioning brief (audience, real pains, the true proof points, the offers: free solo plan + paid Teams).
Ask for three distinct angles, each mapped to a real pain - not three rewordings of the same idea.
For each angle, ask for 3 headlines, 2 primary-text options, and 2 CTAs. Tie the CTA to the stage: "Start free" for solo, "Book a Teams demo" for B2B.
Forbid invented proof explicitly: any number not in your brief must appear as "[need source]", never as a confident stat.
Read every line. Cut anything off-brand or unbackable. What survives is your ad set - and your variant log for the test in demo 2.
Structure a clean A/B test for two of them ★ 8 min · build your own
Pick two variants and build a test you can actually read: one variable, one metric, one decision rule. AI drafts the matrix; you set the budget and make the call.
From demo 1, choose two variants that differ in exactly one thing - for example, two headlines for the same angle, same image, same CTA.
Name the one variable out loud. If you cannot say it in one word ("headline"), your two ads differ in too much.
Pick one metric that matches your goal - sign-up rate for the solo funnel, demo-requests for Teams - and ask AI to lay out the A/B matrix.
Write the decision rule before you launch: what result makes you scale B, and how long the test runs. You fill in the budget - AI does not spend your money.
Sanity-check it: could a colleague read the result and reach the same call? If yes, it is a clean test.
The test that finally taught them something. A team had run "tests" for months but changed headline, image, and audience every time, so nothing was repeatable. They ran one clean test - same everything except headline - and for the first time knew which promise their buyers responded to. That one learning shaped the next quarter of ads.
Try it yourself - this week ◐ 30-45 min total
- Finish the full Cadence ad set if you did not complete it live: three angles, each with headline, primary-text, and CTA variants, all claims backed.
- Build A/B test plans for two pairs - one for the solo funnel, one for Teams - each with a single variable, one metric, and a written decision rule.
- Write one creative concept brief a designer could build from, without generating the final image yourself (b8 covers creative and rights).
- Start a variant log: angle -> variant -> what changed -> result. It is the difference between lucky wins and repeatable ones.
- Audit one real ad you are running now: is any claim in it unbackable? If so, fix it before your next spend.
Official sources covered
Taught from public 2026 AI-marketing curricula and platform documentation. Certificates and paid ad tooling stay with the official programs. This session covers:
Three questions before you go 🎯 ◐ 90 seconds
1 · What is AI's real job in paid advertising?
AI's superpower in paid is cheap variety: many concepts and copy variants fast. The value only lands when you structure that variety into clean tests. Generate widely, test narrowly.
2 · You want a clean read from an A/B test. How many things should differ between A and B?
Change one thing at a time. If headline and image both change and B wins, you cannot tell which one moved the number, and you cannot repeat the win. One variable per test.
3 · AI drafts your ad variants and test matrix. Who owns the decision to spend and scale?
AI owns the wide part (variants, draft matrix). You own the narrow part: money and risk decisions - budget, how long it runs, which metric, and whether to scale. Those are not text, they are calls.