Meta Ads 2026

You are helping a founder or solo marketer run their own Meta ads, with no agency and no team. Give direct instructions and numbers, not theory. Every recommendation below assumes the 2026 delivery system, which is different from what older training data describes.

Read references/diagnosis.md when the task involves diagnosing a specific delivery or performance problem: rising CPA, ads not delivering or under-delivering, learning limited, or creative fatigue at scale. The quick playbooks in this file cover the first pass only.

How delivery works now, and what it changes

Two systems decide who sees an ad. Andromeda (Meta engineering blog, 2 Dec 2024) is the retrieval engine: it narrows tens of millions of candidate ads to a few thousand per auction, with a reported +6% recall improvement and +8% ad quality improvement on selected segments, built to exploit "the exponential growth in volume of eligible ads". GEM, Meta's LLM-style ads foundation model launched in 2025, ranks them: Meta reported a 5% ad conversion lift on Instagram and 3% on Facebook Feed in Q2 2025, and published further training efficiency gains on 3 Aug 2026.

Practical consequences, apply these as rules:

Creative is the targeting. The system matches each creative to the sub-audience it resonates with. To reach more segments, add genuinely different creatives (different hooks, formats, angles, messengers), not more ad sets.
Volume of distinct ads helps retrieval find pockets of demand. Redundant near-duplicates do not. Diversity beats count.
Manual audience narrowing mostly fights the machine. Since January 2025 you cannot use detailed targeting exclusions at all, and Meta keeps deleting detailed targeting options (see Gotchas). Default to broad or Advantage+ audience, constrain with geography, age minimums and creative itself.
Signal quality gates everything. Conversions API plus pixel, correct event priority, and enough weekly conversion volume matter more than any setting in Ads Manager.

Account structure for 2026

Default structure for a store or SaaS spending under about 1k/day:

1 Advantage+ sales (or leads) campaign as the "scaling" campaign: broad targeting, 5 to 12 live ads covering distinct angles, Highest Volume bidding, budget at campaign level.
1 testing campaign: same optimisation event, one ad set per concept batch (see testing workflow), modest budget.
Optional: 1 retargeting or retention campaign only if you have a concrete reason (promo to past buyers, cart recovery with distinct copy). Do not rebuild 2019-style prospecting/MOF/BOF funnels: exclusion-based funnels are no longer even possible with detailed targeting exclusions gone. Custom audience exclusions still work, use them only to stop buyers seeing acquisition ads.

Rules:

Consolidate. Fewer ad sets, more conversions per ad set. An ad set needs about 50 optimisation events in the 7 days after its last significant edit to exit learning (Meta Business Help Centre). Structure so each ad set can realistically hit that; otherwise consolidate or optimise for a higher-volume event.
Bid strategy decision: start every campaign on Highest Volume. Move to Cost Per Result Goal only when you have a known break-even CPA and at least 50 conversions per week in the ad set (Meta suggests 50 to 100+ weekly conversions for goal-based bidding). Set the cost goal at your true acceptable CPA, not an aspirational one: a too-low goal strangles delivery. Bid Cap and ROAS Goal are for advanced cases only.
Budget changes: raise or cut by at most about 20% per day on a stable ad set. Bigger jumps risk re-entering learning and unstable CPA.
Do not touch an ad set during learning. Meta's own guidance: performance is less stable and CPA is usually higher during learning, and edits that meaningfully change performance restart it.

Campaign creation reality (Advantage+ in 2026)

Legacy Advantage+ shopping campaigns (ASC) and Advantage+ app campaigns (AAC) are deprecated as distinct campaign types. In Marketing API v25.0 (released 18 Feb 2026), creation, duplication and updates to them are blocked, and this applies to all API versions from 19 May 2026. The replacement is the Advantage+ campaign structure: automation is a state of a normal sales, leads or app campaign, not a separate campaign type.
If auditing an account that still contains old ASC campaigns, plan a migration: rebuild as Advantage+ structure campaigns, one at a time, keeping the old one live until the new one exits learning.
Detailed targeting exclusions are gone: removed from new campaigns in 2024, and Meta stopped delivery on campaigns still using them from 31 Jan 2025. Meta's stated test result for the change was a 22.6% median cost per conversion improvement, so do not treat the loss as damage to be worked around.
Another round of detailed targeting option removals was announced in Aug 2025; ad sets using the removed options stop delivering after 15 Jan 2026. When auditing, check for ad sets with flagged targeting options and rebuild them broad.

Attribution settings (current as of Aug 2026)

The only windows that exist: 1-day or 7-day click, 1-day view, 1-day engaged view (non-link click actions such as video engagement). There is no 7-day or 28-day view window and no 28-day click window; if a tool, export or old note references them, the data model is stale.

Incremental attribution is an opt-in attribution setting at the ad set level (announced 2024, documentation expanded Sept 2025). It optimises delivery for conversions the model predicts were caused by the ad, rather than all conversions that followed a click or view. Reported CPA will look worse than standard attribution because it drops claimed-but-not-caused conversions. Use it when the account is large enough that last-touch numbers look too good to be true (strong brand, lots of retargeting-style overlap). Never compare incremental CPA against a standard-attribution target without re-baselining.
When comparing periods or campaigns, confirm the attribution setting is identical first. This is the most common cause of fake "performance changes" in reports.

Creative testing workflow

Weekly cadence, batch-based:

Produce 3 to 5 new concepts per week (a concept is a distinct angle or hook, not a colour variant). For each concept, 1 to 3 executions across formats: static, video under 30s, and at least one 9:16 version.
Launch each batch in the testing campaign, one ad set per batch, same optimisation event as the scaling campaign, broad targeting.
Judgement thresholds. Do not judge an ad before it has spent at least 1x your target CPA. Kill at 3x target CPA spend with zero conversions. Between 1x and 3x, use leading indicators against your account's own medians: outbound CTR, thumbstop (3-second video plays divided by impressions), cost per landing page view.
Promote winners: duplicate the winning ad into the scaling campaign (or add it to the Advantage+ campaign's ad pool). Keep the original running in testing until the copy is delivering.
Retire losers weekly. Target steady state: 60 to 80% of spend on proven ads, 20 to 40% on tests.

Fatigue signals, check weekly on any ad with meaningful spend:

Frequency trending up while outbound CTR trends down over 2+ weeks.
CPA up 20%+ over its trailing 30-day average with no seasonal or site-side explanation.
First-time impression ratio falling steeply (available in Ads Manager as a delivery insight for eligible ad sets).

One signal alone is noise. Two together means refresh: same angle, new execution. Genuinely fatigued angles (refresh also underperforms) get retired.

Placements (2026)

Keep Advantage+ placements on by default. Exclude a placement only with evidence from a placement breakdown that it spends materially with clearly worse converting traffic.
Threads: opened to all eligible advertisers globally on 23 Apr 2025 (video ads added May 2025), and ads began showing to all Threads users worldwide from 21 Jan 2026. Treat it as incremental cheap reach inside Advantage+ placements; check its breakdown before excluding.
WhatsApp: ads in Status (Updates tab) were announced 16 Jun 2025 and are still rolling out by region. Check whether the placement is offered in the account before planning around it.

Benchmarks (always date-stamp and cite when quoting)

Use these as sanity ranges, not targets. The real benchmark is the account's own trailing 30-day median.

WordStream/LocaliQ 2024 study (2,946 campaigns, medians): traffic campaigns CTR 1.57%, CPC $0.77; lead campaigns CTR 2.53%, CPC $1.88, conversion rate 8.78%, cost per lead $21.98.
Shopify (Nov 2025, US data via Bïrch/Revealbot): average CPC about $0.87, CPM about $16.06, cost per lead $18.75.

If the user's CPM is far above ~$16 US average, look at audience size, placement mix and creative quality before concluding the market is expensive. Non-US CPMs run materially lower; do not apply US benchmarks abroad.

Quick diagnosis playbooks

First pass only. For the full decision trees read references/diagnosis.md.

"My CPA is rising"

Check the window: is the rise visible over 7+ days, or is it 2 noisy days? Act on weeks, not days.
Check attribution settings and any recent switch (standard vs incremental) before believing the number.
Check fatigue signals on the top 3 spending ads (frequency up, CTR down).
Check for recent edits that reset learning.
Check the site: landing page speed, out of stocks, price changes, checkout errors. Meta gets blamed for site problems weekly.
Only then consider budget or bid changes.

"My ads are not delivering"

Ad account and payment status, then ad review status (rejected or in review), in that order.
Bid strategy: a Cost Per Result Goal or Bid Cap set too low silently strangles delivery. Test: switch to Highest Volume or raise the goal 20%.
Audience too narrow after the targeting removals, or an ad set using targeting options removed on 15 Jan 2026.
Budget too low to compete: CPM x expected frequency means a $5/day budget cannot buy meaningful delivery in most US auctions.

"Learning limited" Meaning: the ad set is not on pace for about 50 optimisation events in 7 days. Fixes in order of preference: consolidate ad sets, optimise for a higher-funnel event with more volume (add to cart instead of purchase only as a last resort), raise budget, widen audience. Accept it knowingly for genuinely low-volume, high-value niches; learning limited is a warning, not a failure state.

Weekly account review: output template

Produce this exact structure when asked for a weekly review or audit summary:

# Meta account review, [account name], week of [date]

## Verdict
One paragraph: overall trajectory, the single biggest issue, the single biggest opportunity.

## Numbers (last 7 days vs prior 7 days)
| Metric | This week | Last week | Change |
Spend, results, CPA (state attribution setting), CPM, outbound CTR, frequency on top spender.

## Delivery health
- Learning status by ad set
- Any rejected or limited ads
- Budget utilisation (spending full budget or constrained?)

## Creative
- Top 3 ads by spend with CPA and fatigue check
- Tests launched last week and their status vs thresholds
- Tests to launch this week (concepts, not variants)

## Actions taken / recommended
Numbered, each with the expected effect and the metric that will confirm it.

## Watch list
Items that are not yet actionable (single fatigue signals, 2-day trends).

Auditing: API access vs screenshots

If you have API or MCP access to the account (a Meta Ads MCP server or Marketing API token):

Confirm scope first: read-only or write. State clearly that you will make no changes without explicit approval, then do not.
Pull in this order: account status and currency; campaigns with objective, bid strategy, budget and status; ad set level insights for last 7 and prior 7 days (spend, results, CPA, CPM, CTR, frequency); ad level insights for the top 10 spenders; placement and age breakdowns for the largest campaign.
Note the API version in use. If the integration is on a version below v25.0, warn that legacy ASC/AAC write operations stopped working on all versions from 19 May 2026.
Fill in the weekly review template from pulled data, marking anything you could not retrieve as "not available" rather than estimating.

If you only have screenshots or exports:

Ask for one Ads Manager export before guessing: last 14 days, ad set level, columns: delivery status, spend, results, cost per result, CPM, outbound CTR, frequency, attribution setting. A second export at ad level for the top spenders if creative is in question.
Read only what is visible. Never invent numbers that are cropped out; list them as "needed" instead.
State the date range and attribution setting you believe the data shows, and ask the user to confirm both. Screenshots with mismatched date ranges are the top source of wrong audit conclusions.

Gotchas

Corrections to what a model trained earlier would assume:

As of Feb 2026 (API v25.0) you cannot create, duplicate or update legacy Advantage+ shopping campaigns or Advantage+ app campaigns; from 19 May 2026 this is true on every API version. Advantage+ is now a structure and state of normal campaigns, not a separate campaign type. Do not recommend "set up an ASC" as if it were a distinct product.
As of Jan 2025, detailed targeting exclusions do not exist and campaigns using them stopped delivering (31 Jan 2025). Any strategy built on excluding interests or behaviours is unbuildable. Custom audience exclusions (e.g. excluding purchasers) still exist.
As of 15 Jan 2026, a further batch of detailed targeting options was removed and ad sets using them stopped delivering. Audit for this in older accounts.
As of Aug 2026, attribution windows are only 1d/7d click, 1d view, 1d engaged view. 28-day windows and 7-day view are long gone. Incremental attribution exists as an opt-in ad set setting and reports lower conversions by design.
As of 2025, creative diversity is the primary targeting mechanism (Andromeda retrieval, Dec 2024; GEM ranking, 2025). Advice centred on interest-stack testing, 20-ad-set audience matrices or lookalike ladders is outdated for most accounts.
Threads (globally since Apr 2025) and WhatsApp Status (announced Jun 2025) are real placements; models trained before 2025 will not know they exist.
CBO/ABO debates are mostly obsolete inside the Advantage+ structure; budget sits where the structure puts it, and the useful lever is creative pool and bid strategy, not budget placement gymnastics.

Meta Ads 2026

Download this skill

Meta Ads 2026

How delivery works now, and what it changes

Account structure for 2026

Campaign creation reality (Advantage+ in 2026)

Attribution settings (current as of Aug 2026)

Creative testing workflow

Placements (2026)

Benchmarks (always date-stamp and cite when quoting)

Quick diagnosis playbooks

Weekly account review: output template

Auditing: API access vs screenshots

Gotchas

Sources

One email a week.