Paid Ads — Knowledge Base

v1 — verify/expand. Paid social + paid search best practices for 2026. For ad creative effects, see ../effects_matrix.md → "Ads (A) — top picks". For landing-page conversion, see the L channel picks. For measurement plumbing, see google-analytics.md.

The 2026 reality

Creative is the #1 lever. Platforms automate bidding, targeting, and placement (Advantage+, Performance Max, broad match + AI). Your durable edge is creative volume + quality + the offer.
Feed AI the right signals, not micro-managed settings. Consolidate campaigns, give algorithms conversion data and headroom, use broad/AI targeting + strong creative + clean conversion tracking.
Privacy/signal loss is permanent (ATT, cookie deprecation). Server-side tracking, conversion APIs, and modeled conversions are now table stakes — see Measurement below.
First-party data wins — upload customer lists, use CAPI/server events, lean on lookalikes/seed audiences over manual interest stacking.

Account structure

Keep it simple and consolidated so the algorithm can learn:

Paid social (Meta): few campaigns by objective/funnel stage; broad audiences + Advantage+ where it performs; many creatives per ad set. Don't fragment budget across dozens of tiny ad sets (starves learning phase — needs ~50 conversions/week per ad set to exit learning).
Paid search (Google): by theme/intent. Modern structure leans on fewer, tightly-themed campaigns; broad match + Smart Bidding + good negatives + strong RSAs; Performance Max for full-funnel/shopping (feed it assets + audience signals + good conversion data, and watch it doesn't cannibalize brand search — add brand exclusions).
Naming convention everything consistently (objective_audience_funnel_creative) for reporting sanity.

Creative testing

Volume + iteration beats one "perfect" ad. Ship many concepts, kill losers fast, scale winners.
Test one big variable at a time (hook, format, offer, angle) — not 12 tiny tweaks.
Hook in first 1-2s for video (same rule as organic social). Thumb-stopping frame 1.
Formats: UGC/creator-style and native-looking ads outperform polished brand films in-feed; static + carousel still strong for certain offers. Mix.
Iterate winners: new hooks/openers on proven bodies; refresh before fatigue (watch frequency + declining CTR/rising CPA).
Modular creative: build hooks, bodies, CTAs as swappable parts.

Budget pacing

Give learning phase room — don't edit ad sets daily (resets learning). Make changes in meaningful steps (±20-30%), not constant small nudges.
Scale winners gradually (≈20-30% budget increases) or duplicate to scale; sudden 3x jumps destabilize delivery.
Dayparting only with real evidence; usually let the algorithm allocate.
Set a CPA/ROAS target and let bidding work; intervene on trends, not single-day noise.
Reserve a testing budget (~10-20%) separate from scaling.

Audiences

Paid social: broad + AI-expanded as default; layer first-party (customer lists, site/CAPI events) → lookalikes/seed → retargeting (warm: site visitors, engagers, cart). Retargeting is efficiency, not net-new growth.
Paid search: intent is the targeting — keyword match types + audience signals (in-market, customer match) layered on. Negatives are an ongoing job.
Exclusions: suppress recent converters from prospecting; exclude brand terms from PMax/broad to keep brand search cheap.

Measurement (don't skip — bad tracking = wasted spend)

Server-side conversion tracking: Meta Conversions API (CAPI), Google Enhanced Conversions / server-side GTM. Client-side pixels alone undercount badly post-iOS/cookie-loss.
Define one primary conversion per campaign and optimize to it; track assists separately.
Match the platform metric to reality — platform-reported ROAS is optimistic; triangulate with GA4, backend revenue, and incrementality (geo/holdout tests) for real decisions.
UTM everything consistently (see google-analytics.md UTM hygiene) so GA4/warehouse attribution is clean.
Watch the right metrics by stage: prospecting → CPM/CTR/hook-rate/CPA; retargeting → ROAS/frequency; search → CPC/CTR/conv-rate/CPA/impression share.

Common mistakes

Over-segmenting ad sets/campaigns → starved learning, no exit from learning phase.
Editing live ad sets constantly → perpetual learning resets.
Judging on single-day data; scaling/killing on noise.
Client-side pixel only (no CAPI/server-side) → undercounting, bad optimization.
Letting PMax/broad eat brand search (no brand exclusions).
One creative, run to death; no testing pipeline.
Optimizing to a cheap proxy metric (clicks) instead of the real conversion/revenue.
No negatives in search.

Quick checklist (pre-launch)

Conversion tracking verified (server-side: CAPI / Enhanced Conversions / sGTM)
One clear primary conversion objective per campaign
Consolidated structure (not fragmented), consistent naming
Multiple creatives ready; hook strong in first 1-2s
First-party audiences + lookalikes loaded; exclusions set
Brand-term exclusion on PMax/broad
Budget allows learning phase; testing budget carved out
UTMs tagged consistently
Landing page matches the ad (offer + message), fast + mobile (see L picks)