Social ads: interrupting well

Nobody asked to see your ad

Search ads answer a question someone just typed. Social ads interrupt someone who was doing something else entirely — watching a friend's holiday, half-reading an argument. That single difference drives everything: the maths (colder attention converts at a fraction of search's rate — Lesson 4 prices this), the creative (you must earn attention you weren't given), and the job (social ads create demand; search collects it).

It also sets the price mechanism. Lesson 1's auction runs here too: the platform predicts how likely each person is to respond to your ad, and charges you accordingly. An ad people scroll past without a flicker gets expensive. An ad that stops the right people gets cheap. On social, more than anywhere, the ad is the product — the platform is only the courier.

Targeting in a post-signal world

A few years ago, the craft was targeting: stacking interests, behaviours and demographics into ever-finer audience definitions. That world is mostly gone. Apple's App Tracking Transparency prompt (2021) cut off much of the behavioural signal from iPhones; EU consent rules mean a large share of visitors never enter any audience at all (Lesson 5 deals with this properly); and the platforms responded by rebuilding delivery around a different engine: watching who responds to the ad itself, and finding more people like them.

The practical consequences, honestly sized for a small budget:

• Broad-ish targeting plus strong creative is the modern default. Give the delivery system room to learn.
• Interest stacks still earn their keep only for genuinely niche audiences — Storkflow's accountancy-firm owners, say — not as a fourteen-checkbox comfort blanket.
• Lookalikes need a quality source list, such as the Module 8 customer list — with a lawful basis for that use, as M8.L2 taught you to ask.
• Retargeting is real and cheap, and it waits for Lesson 5, where its dishonesty problem gets dealt with too.
• Micro-slicing kills small budgets. Split €500 across six narrow audiences and no slice ever accumulates enough conversions for the system — or you — to learn anything. One audience, few concepts, clean tests.
• Special categories are restricted. Housing, credit, employment, politics and health carry platform-enforced targeting limits. A physio clinic advertising "back pain relief" is brushing against health rules — check the current policy before building, not after rejection.

Creative is the targeting

Here is the sentence to keep: the hook decides who stops, and who stops teaches the algorithm who to find. "If you run an accountancy practice…" as an opening line targets accountants more precisely than any interest checkbox, because everyone else keeps scrolling and the system notices.

This means your Module 7 hook craft transfers directly, with a harder edge: organic distribution merely ignores a weak hook, but paid distribution fines it, impression by impression. It also means your proven organic winners are the cheapest ad research you will ever get — a post that already stopped scrolls for free has pre-paid its creative test.

The working anatomy: hook (stop) → tension (relate) → proof (believe) → CTA (act). Formats: single image, carousel, short subtitled video — most feeds are watched with sound off — and UGC-style phone footage, which routinely outperforms polished brand assets because it doesn't look like an ad.

One variable at a time

Run ad A against ad B with different hooks, different images and different offers, and the winner teaches you nothing — three things changed, so no single element can take credit. The discipline: 2–3 concepts, 2 hook variants each, one variable under test. Kill losers on a pre-committed rule; feed winners. Pre-committed matters — decided before any data arrives, so you can't romance a favourite that's losing.

And watch frequency: impressions ÷ reach. A small audience sees the same ad again and again; when frequency creeps up while clickthrough falls, the audience is bored, not undecided. The fix is new creative or a wider audience — never a higher bid. You cannot pay people to be un-bored.

Ana's test grid, worked

Ostara Ceramics, built from the canonical Module 2–3 research: the considered gift-buyer, and the three message pillars — not on every table, arrives safely, ready to give, a gift that shows you tried.

Concept A — "Not on every table." Repurposes her proven organic studio-process video: the Instagram-sameness pillar, aimed at taste anxiety.

• Hook A1: "You've seen these plates on every feed. Not these."
• Hook A2: "If their table shouldn't look like everyone's Instagram, don't buy the plates everyone buys."

Concept B — "Arrives safely, ready to give." UGC-style phone unboxing: insured parcel, plastic-free gift wrap, the handwritten card. Aimed at breakage worry.

• Hook B1: "1,375 parcels shipped last year. 9 broke. Every one replaced free."
• Hook B2: "The gift that survives the post — or the replacement is free."

Concept C — the control. Flat, beautiful product photography, "Handmade in Porto". The ad everyone expects to win. It rarely does — which is exactly why it's in the grid: if C beats A and B, that's a finding, not an embarrassment.

Audience: one broad EU audience, 28–55, gift-adjacent. No interest stack. Test variable: the hook — concept held constant within each pair. Landing page: the Module 4 gift page, not the homepage. Kill rule, pre-committed: after €50 of delivery per concept, kill anything with CTR under 1% or link-click CPC over €0.60. Where does €0.60 come from? Lesson 4's arithmetic — her economics put a ceiling near €0.45 on what a click may cost, so €0.60 is that ceiling plus room for early noise. For now, take it as given; tomorrow you'll compute your own.

Note what this grid is: a plan. Ana happens to have a real €300/month already running, so hers could go live inside existing spend — but the deliverable, for her and for you, is the grid itself, with its rules written down before any euro moves. That's what the module project grades.

---