Paid Media Strategy · eBra · United States
A new campaign built around a single idea: Discover and the YouTube feed are content feeds, so the ad unit is a content card — and a content card is the only format that reads as continuous with an article-style advertorial. Native card → editorial article → offer, with no mode switch anywhere in the chain. Every other Demand Gen surface breaks it, so every other surface is switched off.
An advertorial is an article. It only works on someone who clicked expecting to read something — a person in reading posture, browsing content, with a minute to give. That posture is not a property of the audience; it is a property of the placement. Get the placement wrong and the best advertorial in the world bounces, because the person arriving was never in the mood to read.
Discover and the YouTube feed are the only two Demand Gen surfaces that produce that posture. Both are content feeds: the ad unit is a card with an image and a headline, sitting between the organic articles and videos the person came to browse. The ad looks like the content around it, and the content around it is exactly what the advertorial looks like. That is the whole campaign in one line — native card → editorial article → offer, with no mode switch anywhere.
A Discover card is an image plus a headline — structurally identical to the article link next to it. The advertorial's own hero is an image plus a headline. The click is a continuation, not a jump.
Nobody opens Discover to be sold to; they open it to scroll through things worth reading. That is the only mindset in which "I tested 7 wireless bras" earns two minutes instead of a bounce.
Discover bills on the click, not the impression. A card that genuinely reads as content earns a high CTR — and therefore earns its impressions for free. Native isn't only better creative here; it's cheaper media.
Demand Gen serves across six surfaces by default, sharing one budget and one tCPA. Four of them break the chain:
· YouTube in-stream — interruptive. It rolls over the video the person chose to watch. Someone interrupted mid-video is not going to read 2,000 words.
· Shorts — high-velocity swiping, 6–15 seconds of attention. The opposite of reading posture.
· Gmail — an inbox is a task environment, not a browsing one. It converts, but it converts on offers, not on articles.
· Display — banner blindness; the unit reads as an ad no matter what is in it.
Left blended, the algorithm buys wherever delivery is cheapest — and cheapest is almost never the surface where the advertorial works. Restricting the channels is what makes the creative, the placement and the landing say the same thing.
If the campaign optimises for View-Through Conversions (VTC), Google flips it to impression billing and the free-impression maths disappears. VTC optimisation is off by default — we keep it off and optimise to clicks + standard conversions only. It is the single most important lever in the whole build.
Through Channel controls in the campaign settings (Google Ads UI): keep Discover ON and YouTube in-feed ON; switch off Gmail, YouTube in-stream, Shorts and Display. This is a UI control — the API does not fully expose it yet, so the campaign is created via API and the channel selection is set by hand before enabling. Verify in the channel report every Monday: if spend shows up on Gmail or in-stream, the control has reverted and the coherence is broken.
This is not a speculative proposal. Shapermint already runs Demand Gen at scale in account 103-730-5439 — $371,798 over 90 days — and since August 1 there has been a live Discover campaign pointing at exactly these two landing pages. That test is the best available benchmark, and it says three very specific things.
Workspace rule: the CDP (Redash q2635) is the source of truth for revenue and orders. Comparing platform against CDP over the same window reveals which lane is inflating. The gap between the two image lanes is stark.
| Lane | Spend | Orders (platform) | Orders (CDP) | Inflation | CPA platform | CPA CDP | ROAS CDP |
|---|---|---|---|---|---|---|---|
| Discover + advertorial YT_US_DG_DISCOVER…26W31_IMG_ADV_VIRAL · Aug 1–6 |
$1,310 | 16.5 | 16.0 | +3% | $79.29 | $81.84 | 0.79 |
| Gmail / image YT_US_DG_GMAIL…25W48_IMG_META_TEST · 30 days |
$54,125 | 929.6 | 552.0 | +68% | $58.22 | $98.05 | 0.70 |
The Gmail lane looks like the best in the account at a $58 platform CPA — but the CDP says $98. Sixty-eight percent of the orders it reports do not exist. The Discover lane reports 16.5 against 16.0 real: a 3% gap. Which means that on Discover, the tCPA you hand Google is the CPA you actually pay. That makes the campaign governable — and it is the technical reason it is worth separating instead of leaving it blended with Gmail.
| CDP source | Spend | Revenue | Orders | ROAS | CPA | AOV | % acq. revenue | ATC% | CVR% |
|---|---|---|---|---|---|---|---|---|---|
| Google Discovery | $55,435 | $38,989 | 568 | 0.70 | $97.60 | $68.64 | 74% | 5.2% | 1.12% |
| YouTube | $57,296 | $41,449 | 613 | 0.72 | $93.47 | $67.62 | 66% | 10.4% | 1.78% |
| Google Display | $2,510 | $719 | 9 | 0.29 | $278.86 | $79.89 | 73% | 1.2% | 0.29% |
Three readings: (a) Discovery's AOV ($68.64) sits 24% above the account AOV ($55.35) — whoever enters through the advertorial buys bigger; (b) 74% of the revenue is net-new customers, so judging this lane against a brand-blended ROAS makes no sense; (c) the bottleneck is not the click, it's the add-to-cart: 5.2% against YouTube's 10.4%. The click is already cheap; what's missing is the page. Section 04 identifies a very concrete reason why.
Two Discover campaigns run today: YT_US_DG_DISCOVER…26W31_BROAD_IMG_ADV_VIRAL and YT_US_DG_DISCOVER…26W32_BROAD_IMG_QUIZ_FUNNEL (both tCPA $65, $200/day each). They are exploratory tests, and they read like it: 13 ad groups of which 11 are removed, six different final URLs blended into one campaign, quiz funnels and advertorials and PDPs sharing a budget. They proved the format works. They cannot tell you which part of it works.
Rebuilding on top of that history means inheriting a learning phase trained on six mixed destinations and a tCPA that was never reachable. A clean campaign with one thesis, two landing pages and one audience type per ad group is cheaper to read and faster to optimise. The existing tests keep running as-is — see the coexistence rule in section 07 so the two don't bid against each other.
A single US campaign. Inside it, one audience type per ad group, so CPA is readable per type and budget can move to the winner without guessing. Four ad groups is the maximum $400/day can sustain without starving signal density — the maths is below.
DGEN_US_NB_SHM_EBRA_Discovery_YTFeed_ADV
Channels: Discover ON · YouTube in-feed ON
Gmail / in-stream / Shorts / Display OFF
Budget: $400/day · Bidding: tCPA $85 → $75 → $68
Goal: Sales · standard Shopify purchase conversion
VTC optimisation: OFF
Geo: United States · Language: English
Routing: cold → advertorial · warm → offer page
Status on creation: PAUSED, reviewed in UI before enabling
| Ad group | Audience type | Concrete signal | Opt. Targeting | % budget | $/day | Route |
|---|---|---|---|---|---|---|
| AG1 · Custom Intent | Custom segment (search + URL) | Searchers of wireless bra, wire free bras, comfortable wireless bras, minimizer bra, best no wire support bra, comfortable bras for large breasts + competitor wireless-bra URLs | OFF | 35% | $140 | Advertorial |
| AG2 · In-Market | In-market segment | Women's Intimates & Sleepwear + Bras + Women's Apparel | OFF | 25% | $100 | Advertorial |
| AG3 · Affinity + Lookalike | Affinity + Lookalike (1P seed) | Shoppers > Value Shoppers + Beauty Mavens; Lookalike seeded on Truekind Wireless Shaping Bra purchasers (Customer Match) | ON | 25% | $100 | Advertorial |
| AG4 · Remarketing | Your data | Site visitors 30–90d (eBra PDP + advertorial readers) + cart abandoners + Customer Match. Exclude 90-day purchasers. | OFF | 15% | $60 | Offer page (already educated) |
A Demand Gen ad group needs ≥30 conversions/month to exit learning and genuinely optimise. At $400/day, with the expected CPA of each audience type, four ad groups clear the threshold; a fifth breaks it.
| Ad group | $/day | Expected CPA | Conv./month | ≥30? |
|---|---|---|---|---|
| AG1 · Custom Intent | $140 | $80 | ~52 | Yes |
| AG2 · In-Market | $100 | $85 | ~35 | Yes |
| AG3 · Affinity + Lookalike | $100 | $90 | ~33 | Yes |
| AG4 · Remarketing | $60 | $45 | ~40 | Yes |
| Campaign | $400 | ~$75 | ~160 | — |
ON only where we want Google to expand beyond the signal (AG3, Affinity/Lookalike). OFF on Custom Intent, In-Market and Remarketing — leave it ON for remarketing and Google spends the retargeting budget on cold prospects disguised as warm, at which point the ad group's CPA stops meaning anything. It is the single most-missed setting.
| Dimension | Setting | Why |
|---|---|---|
| Gender | Female · exclude Male · include "Unknown" | The product is for women, but excluding "Unknown" wipes out all signed-out inventory — which on Discover is a lot. |
| Age | 35–44, 45–54, 55–64, 65+ · include "Unknown" · exclude 18–24 | The advertorial is written literally for "after 40". 25–34 stays only if the bracket's CPA holds up at 14 days. |
| Device | Mobile prioritised (no desktop exclusion at launch) | ~90% of advertorial traffic is mobile. Measure and adjust in week 3, not before. |
| Exclusions | Past-90-day purchasers off AG1–AG3 | Pure prospecting: without the exclusion, repeat purchases flatter the CPA and the acquisition read is contaminated. |
The live Discover campaigns run at tCPA $65 against a real CPA of $79.29 on platform / $81.84 on CDP. A target set 20% below what is achievable does not lower the CPA — it makes Google throttle delivery and buy the cheapest, worst inventory in the pool. The right move is to enter above the real CPA and tighten with evidence.
| Step | tCPA | Implied CDP ROAS | Condition to advance |
|---|---|---|---|
| Launch | $85 | 0.81 | Starts above the real CPA ($80) so learning isn't starved. No changes for 14 days or 50 conversions. |
| Step 2 | $75 (−12%) | 0.92 | Only if the 14-day CDP CPA ≤ $85 and the advertorial's ATC% ≥ 7%. |
| Target | $68 (−9%) | 1.00 | $68 ≈ Discovery's CDP AOV ($68.64) → ROAS 1.0 on a channel that is 74% new customers. |
No tCPA move greater than ±15% at a time, and never during the learning period. No budget increase greater than +20% every 3–5 days. Both steps of the ladder sit inside those limits by design.
$400/day is 1.2% of the account's monthly spend ($1,020,614 over 30 days) and moves total Demand Gen from roughly 12% to 13% of the account — still inside the 10–20% band this channel should occupy for a DTC business.
It is not a round number: it is the floor at which four ad groups each clear 30 conversions/month. Below it, the structure has to shrink. If $400/day of new budget isn't available, the correct move is not to run the same four ad groups on less — it is to drop to three ad groups at $300/day, folding In-Market into Custom Intent. Same campaign, one less read, still healthy learning.
If it needs to be funded from inside Google rather than added, the natural source is the Gmail/image lane: $1,804/day at a 0.70 CDP ROAS with 68% attribution inflation. Trimming it 22% covers this campaign entirely.
The two URLs are not alternatives: they are two rungs of the same funnel. The advertorial educates cold traffic; the offer page closes traffic that is already educated. That defines the routing per ad group. But reading both pages live surfaced a problem that undermines a good share of the current spend.
The advertorial says "SEE TODAY'S 40% OFF DEAL" (and in the body, "the current 40% OFF Spring Sale"). Verified live today, both the offer page and the destination PDP show 30% OFF / "Save $17.00". That is ten points of discount promised and not delivered, at the highest-intent click in the entire funnel. On top of that, the copy says "Spring Sale" in August, on a page whose own slug declares it evergreen (-evg).
This is consistent with the 5.2% ATC rate against YouTube's 10.4%: it isn't that people don't want the product — it's that they land and the price isn't the one they were promised. Fixing this is cheaper and faster than any bid change, and there is no case for scaling budget onto a funnel that contradicts itself.
try.shapermint.com/lp1-mbra-adv-evg
Format: long-form review with a byline — "A Bra Expert's Perspective: Is this Truekind Bra ACTUALLY Worth the Hype?" by Jené Sena, bra expert for women 40+.
Arc: personal story (the shop-window reflection) → the problem after 40 → the insight ("it was never the underwire that lifted you — it was the architecture") → a 7-bra wear test → head-to-head across 3 categories (support, silhouette, price-per-wear) → social proof (sells out at Walmart, 5M+ sold) → offer + 60-day guarantee.
Serves: AG1 · AG2 · AG3
It educates, compares and closes in one session. It is the only asset that can carry a $0.61 CPC against an audience that doesn't know the brand.
shapermint.com/pages/truekind-supportive-comfort-wireless-shaping-bra-offer-page-2?variant=40278561292422
Format: extended PDP-style offer landing — benefits, "Other Bras" comparison, testimonials, colour picker, bundle.
US offer verified today: 30% OFF · Save $17.00 · free shipping over $80 · 635 on-page reviews · 1M+ sold.
Serves: AG4 (remarketing), and it is the destination of the advertorial's CTA.
It opens by selling: no narrative, no byline, no third-party authority. Sending cold Discover traffic here asks people to buy before they understand the problem.
1. Align the offer (blocking). Either the advertorial drops to 30% OFF, or the offer page goes to 40%. Either works; what does not work is the two disagreeing. The campaign does not go live without this.
2. Strip "Spring Sale" from the evergreen advertorial. Replace with season-free language ("today's deal", "current offer") so the page stops ageing out every quarter.
3. Consolidate destinations. The current Discover campaign has 6 different final URLs mixed into one campaign, which makes it impossible to know which landing converts. The new campaign uses exactly two: advertorial for cold, offer page for warm.
lp1-mbra-adv-evg — editorial hero with byline and dateline
On Discover the creative buys the reach: at a 4.55% CTR and a $0.61 CPC the implied CPM is $27.86 — but you don't pay it, you earn it. Which is why the creative is not "the pretty part": it is the economic lever. The card has to look like feed content, not like an ad.
| Asset | Spec | Minimum | Note |
|---|---|---|---|
| Landscape image | 1.91:1 · 1200×628 · ≤5 MB · JPG/PNG | 3 | The primary Discover card format |
| Square image | 1:1 · 1200×1200 · ≤5 MB | 3 | YouTube in-feed prioritises this ratio |
| Portrait image | 4:5 · 960×1200 · ≤5 MB | 2 | Best mobile real estate |
| Square logo | 1:1 · 1200×1200 | 1 | Required |
| Landscape logo | 4:1 · 1200×300 | 1 | Optional but contributes to Ad Strength |
| Headlines | ≤ 40 characters | 5 | Editorial tone, not promotional |
| Long headline | ≤ 90 characters | 1 | The one shown most often on Discover |
| Descriptions | ≤ 90 characters | 5 | — |
| Business name | ≤ 25 characters | 1 | "Shapermint" |
| Video (optional) | 3–6s scroll-stopper · 1:1 and 9:16 | opt. | Only adds against YouTube in-feed inventory |
No packshot on white, no logo in the first 2 seconds of video, no centred ad-style typography. The card has to look like an article photo. The winners in this account's Discover lane are the ones that don't look like Shapermint.
| # | Angle | Art direction | Headline (≤40) | Description (≤90) | Cluster it activates |
|---|---|---|---|---|---|
| 1 | The architecture reveal | Detail shot of the wide band + contour cup on fabric; natural light, magazine-editorial framing | It was never the underwire | The wire was a shortcut. The lift always came from the band and the cup. | Wireless (LEAD) |
| 2 | The turtleneck, two bras | Same black sweater, same woman, two back shots side by side — one with visible roll lines, one smooth | Same sweater. Two different bras. | The wide back panel is why one lies flat and the other doesn't. | Back bulge / minimizer |
| 3 | The shoulder marks | Close-up of a shoulder at end of day, no marks; real bathroom light, not studio | 12 hours. Zero shoulder marks. | Cushioned wide straps spread the weight instead of carving a line. | All-day comfort |
| 4 | The empty shelf | Empty in-store bra rack, phone photo, slightly imperfect | The shelf was empty. Again. | The #1 wireless bra in the world sells out the week it lands. | Best-of / social proof |
| 5 | The authority | Editorial portrait of the expert with visible byline, magazine-column styling | I tested 7 wireless bras. One worked. | A decade fitting women 40+. Here's the only one that kept the structure. | Best-of / review |
| 6 | After 30 washes | Two bras side by side on marble: one holding shape, one stretched and misshapen | After 30 washes. Side by side. | Cheap wireless stretches out in 8 weeks. Do the price-per-wear math. | Support & lift |
“After 40, most wireless bras stop working. A bra expert tested 7 — one quietly won.” (84 characters)
3 angles per ad group at launch; refresh the weakest every 2–3 weeks or whenever CTR drops >30% from peak. On the Discover lane fatigue shows up in CTR first, not CPA — and because CTR is what buys the impressions, a fatigued creative makes the click more expensive before the CPA ever reveals it.
Direct pull from KeywordPlanIdeaService (Aug 7, 2026, geo United States) using 10 seeds from the wireless-bra problem space. It is not used to bid on keywords — Demand Gen has no keywords. It is used to build AG1's Custom Segment ("people who searched for X") and to decide which cluster each card writes against.
From the wireless-bra problem space · geo US · English
546,760 monthly searches across the 7 clusters
Median Search CPC $2.61 · interquartile range $1.96–$3.33
$0.61 real Discover CPC vs. $2.61 median Search CPC for the same interest
| Cluster (advertorial theme) | Keywords | Searches/mo | Comp. | CPC range | Top terms |
|---|---|---|---|---|---|
| Wireless / no-underwire LEAD | 440 | 313,920 | HIGH | $0.09–12.46 | wireless bra (110,000) · wire free bras (33,100) · comfortable wireless bras (14,800) · comfy wireless bras (14,800) |
| Back bulge / smoothing / minimizer | 141 | 116,800 | HIGH | $0.26–3.05 | minimizer bra (40,500) · best minimizer bras (9,900) · top rated minimizer bra (9,900) |
| Long-tail / spelling variants | 71 | 82,760 | HIGH | $0.30–3.82 | minimizerbras (40,500) · under wire (22,200) · no under wire bra (3,600) |
| Support & lift for larger busts | 133 | 12,510 | HIGH | $0.22–6.09 | best supportive bras for large breasts (1,600) · best no wire support bra (1,600) · supportive no wire bra (1,300) |
| All-day comfort / pain relief | 41 | 7,910 | HIGH | $0.21–5.56 | comfortable bras for large breasts (1,300) · most comfortable bras for large breasts (720) |
| Best-of / review / comparison (advertorial intent) | 28 | 6,420 | HIGH | $0.25–4.90 | best no wire bra (1,600) · best non underwire bra (1,600) · best bras without wires (1,600) |
| Coverage / t-shirt / everyday styles | 16 | 1,150 | HIGH | $0.32–4.31 | padded no wire bra (480) · non underwire padded bra (210) |
| Total actionable | 870 | 541,470 | — | — | — |
| [EXCLUDED] brand & competitor | 20 | 5,290 | — | — | Kept out of the prospecting Custom Segment — covered by Search Brand |
Load the ~25 highest-volume terms from the Wireless + Minimizer + Best-of clusters as "people who searched for these terms" (7–30 day window), plus 4–6 competitor wireless-bra URLs as a site signal.
The three largest clusters (Wireless 313,920 · Minimizer 116,800 · Long-tail 82,760) hold 94% of the demand — they are what cards 1, 2 and 6 should be written against.
Never include brand terms (Shapermint, Truekind, Empetua) in the prospecting segment: that turns an acquisition CPA into a brand CPA in disguise.
| Metric | Source | Target | Alert | Note |
|---|---|---|---|---|
| ROAS | CDP · revenue_1dc / spend | ≥ 0.85 at 30d | < 0.60 | Never platform conversions_value |
| CPA | CDP · spend / conversions_1dc | ≤ $85 → $68 | > $110 | Ladder from section 03 |
| ATC% (the page KPI) | CDP · add_to_cart_1d / users_1d | ≥ 7% | < 4% | Discovery baseline today 5.2% · YouTube 10.4% |
| CTR (the creative KPI) | Google Ads | ≥ 4.0% | < 2.5% | The current test runs at 4.55% |
| CPC | Google Ads | ≤ $0.75 | > $1.10 | Above $1.10 the channel mix is leaking |
| % acquisition revenue | CDP · revenue_1dc_acq | ≥ 70% | < 50% | Discovery baseline 74%. If it falls, it's remarketing in disguise |
| Channel split | Google Ads · channel report | 100% Discover + YT feed | any Gmail/in-stream | Mandatory weekly check |
| Conv./month per ad group | Google Ads | ≥ 30 | < 20 | Below 20, merge the ad group into another |
The tCPA is set against platform conversions; the CPA we report comes from the CDP. On this lane the two agree (3% gap), which is what makes the campaign governable — but that has to be re-verified every 30 days. The day the gap exceeds 15%, the tCPA stops being a real target and must be recalibrated against CDP.
| Day | Action |
|---|---|
| Monday | Pull 7-day CDP + platform. Check the channel split (did Gmail or in-stream leak in?). Compare CDP CPA vs tCPA. |
| Tuesday | Pause creatives with CTR <2.5% or >30% below peak. Pause ad groups that violate the Bleeding Rule (>$50 with zero conversions in 30 days). Do not touch the winners. |
| Wednesday | Scale the winning ad group by +20% maximum. Move tCPA only if the ladder condition is met. |
| Thu–Sun | Monitor. No structural changes. |
The new campaign does not replace anything on day one, so overlap has to be managed deliberately — two campaigns chasing the same audience on the same surface bid up each other's CPC and split the conversion signal.
| Running campaign | Overlap with the new one | Action |
|---|---|---|
…26W32_BROAD_IMG_QUIZ_FUNNEL$200/day · quiz landing |
None — different funnel entirely (lp1-find-your-perfect-bra-2 quiz, not the advertorial) |
Keep running. Genuinely a separate test. |
…26W31_BROAD_IMG_ADV_VIRAL$200/day · same advertorial + offer page |
Direct — same surfaces, same audiences, same two landing pages | Wind down once the new campaign exits learning (50 conversions or 14 days) — not before, so nothing is lost while the new one is still learning. |
Advertorials are allowed, but the page must be transparent about its commercial nature. Concretely: (1) a visible "Advertisement" / "Advertorial" label at the very top; (2) disclosure near the CTA that the page belongs to the brand; (3) no unsupportable health or results claims; (4) no impersonation of a real news outlet and no press logos without an agreement — "As seen on" only where the coverage genuinely exists; (5) real, verifiable testimonials; (6) the advertised offer must match the landing page — the 40%/30% mismatch in section 04 is exactly the kind of discrepancy Google enforces against under "Misrepresentation".
tracking_url_template set with UTMs so the CDP classifies the channel correctly…26W31_ADV_VIRAL winds down at learning exit; the quiz-funnel test keeps runningCDP ROAS ≥ 0.85 with CDP CPA ≤ $85 and ≥70% acquisition revenue, at $400/day. That means roughly $12,000 of spend generating about $10,200 of revenue that is almost entirely net-new customers, at an AOV 24% above the account average — at an acquisition cost that other lanes currently pay more for.
The real test of the thesis is narrower than the ROAS, though: does ATC beat 7%? That is the number that tells us the native-card → article chain is holding. The click is already proven cheap ($0.61); if people arrive in reading posture and the article does its job, ATC moves. If at 30 days ATC is still below 4% with the offer already aligned, the problem is neither the channel nor the bid — it is the advertorial, and it needs rewriting before another dollar goes in.