Why Meta CPMs Are Rising In 2026 (And Why Creative Volume — Not Targeting — Is The Only Fix Post-Andromeda) | PRESTYJ Blog

If you run a Meta ad account with more than 12 months of history, you already know the chart: same offer, same landing page, same audience, CPM line drifting up and to the right. Most accounts we audit are paying 30–60% more per thousand impressions than they were in 2023, and the buyers running them are doing the same three things — narrowing audiences, swapping placements, lowering bids — none of which work anymore.

This post is the unvarnished version of what's actually happening to Meta CPMs in 2026, why bidding and audience tricks can't fix it, and why creative volume is now the only variable that meaningfully moves your effective cost down. The short version: Meta's Andromeda retrieval system rewards advertisers who feed it more distinct creative, and punishes advertisers who feed it less. CPM is the price you pay for that punishment.

If you're a media buyer, CMO, or service-business owner trying to justify rising ad spend to a CFO, this is the artifact. Real numbers, real math, no hype.

TL;DR

Meta CPMs are up 30–60% in most lead-gen verticals since 2023, with home services, real estate, mortgage, and legal hit hardest.
Three forces are stacking: more advertisers in the auction, Andromeda weighting predicted conversion harder, and iOS 14.5+ ATT signal loss leaving Meta hungrier for creative signal.
Bidding strategy changes do not fix this. Lower bids just lose the auction. Higher bids burn budget faster on the same inferior creative.
Audience narrowing makes it worse. Andromeda needs a large candidate pool to evaluate; narrow audiences raise CPMs and starve the algorithm.
Creative volume is the only real lever. More distinct hooks/angles/visuals → more cheap impression slots Meta can place you in → blended CPM falls.
Case math at $10K/month: moving from 6 active creatives to 60 typically drops blended CPM 18–34% and CPA 30–50% — without changing the offer.

What Andromeda Changed About The Auction

Most CPM analysis written before late 2024 is now stale. Meta's Andromeda retrieval system — co-developed with NVIDIA on GH200 GPUs and rolled out through 2025 — evaluates roughly 10,000x more ad candidates per impression than the previous ML stack. That sounds like trivia. It is not.

Here's why it matters for your CPM specifically:

The auction is no longer about which ad is "right" for the user. It's about which ad, out of an enormous candidate pool, has the highest predicted value (predicted CTR × predicted conversion × bid). The retrieval layer can now consider so many candidates that the marginal cost of being one of them is functionally zero — but the marginal cost of winning against them has climbed sharply.
Predicted conversion weight has gone up. Andromeda doesn't just predict "will this user click." It predicts "will this user click and convert." Advertisers with thin creative libraries produce thin prediction signals, which means Meta can't confidently slot them into cheap inventory.
Advantage+ is now default. Both Advantage+ Audience and Advantage+ Placements are the path of least resistance, and they're the surfaces Andromeda is optimized for. Detailed-targeting accounts effectively pay a tax to run against the grain.

The practical implication: the auction is more competitive at the high end (predicted high-value users) and more lenient at the low end (predicted low-value impressions), but you only get access to the lenient end if you have enough creative variants for Meta to test against the entire distribution.

If you only have 5 ads running, Meta only has 5 chances per impression to find a fit. Most of those chances fail. So Meta charges you premium prices to deliver to the narrow pool of users where one of your 5 ads is the best candidate.

Force #1: More Advertisers, Same Inventory

This is the boring one, so we'll get it out of the way first.

Meta added roughly 1.4 million new active advertisers in the last 24 months, with the steepest growth in small business and local services. Inventory — the number of impressions a user can see in a day — is essentially fixed by user behavior, not by Meta. More demand against fixed supply produces higher clearing prices. Econ 101.

What's less obvious: the new entrants are worse bidders on average. They run Advantage+ campaigns with one or two creatives, no objective discipline, and bid caps set by intuition. Andromeda still has to evaluate them in every auction. The collective effect is a lot of low-quality bids that nonetheless drag up the floor — because Andromeda is willing to deliver them at low predicted-value impressions to keep the auction efficient overall.

If you used to run a $4K/month CPM-15 account in a quiet vertical, you're now sharing impressions with 40 new local competitors who didn't exist in 2023. You can't audience-target your way out of that. They're all targeting the same broad pool, because that's what Advantage+ wants.

Force #2: Andromeda Weights Predicted Conversion Harder

Pre-Andromeda, the rough mental model was: bid × predicted CTR × quality score. Click-through dominated. You could get cheap impressions with a clickbait hook and a thin LP, even if conversion was awful.

Post-Andromeda, the predicted-conversion term has more weight, and it's modeled at vastly higher resolution. Andromeda can predict conversion likelihood per (user × creative × placement × time-of-day × device) tuple far more accurately than the old stack, because it can actually evaluate enough candidates to learn those joint distributions.

What that means for your CPM:

Accounts with strong post-click signal (high conversion rate, fast page load, clean event setup) get rewarded with lower CPMs because Andromeda is confident they'll convert.
Accounts with weak post-click signal pay a CPM premium even if their CTR looks fine, because Andromeda is hedging against wasted impressions.
Accounts with low creative volume can't train the model on themselves. Andromeda doesn't know what your converting user looks like because you've shown it 5 creative samples. It defaults to category averages, which are usually unflattering.

The third bullet is where most rising-CPM diagnoses go wrong. Buyers blame the audience, the offer, the placement — but the actual issue is that the algorithm doesn't have enough creative variants from you to build a confident prediction model. So it prices your inventory conservatively, which means expensively.

This is the core thesis of our creative-is-the-new-targeting explainer: creative diversity is now the input variable that lets the algorithm target for you. Without it, you're running blind and being charged the blindness tax.

Force #3: iOS 14.5+ ATT Signal Loss Still Compounds

This is the force most accounts under-weight in their CPM analysis. Apple's App Tracking Transparency framework (iOS 14.5, April 2021, but still compounding) cut Meta off from a significant share of deterministic conversion signal on iOS users — which, depending on your vertical, is 50–70% of your target audience.

Meta has replaced most of that signal with modeled conversions and CAPI (Conversions API) ingestion from advertiser servers. But modeled conversions are noisier than deterministic ones, and CAPI is only as good as the advertiser's implementation. For local service businesses without engineering teams, CAPI is often half-broken or wired up to the wrong events.

The Andromeda implication: when Meta has less direct signal about whether your previous impressions converted, it has to rely more heavily on creative-level signal — what does the ad itself predict about who will engage and convert? That's not optional. It's how the system stays accurate.

So:

More creative variety → more signal Meta can extract per impression → better predictions → lower CPM.
Less creative variety → thinner signal extraction → noisier predictions → CPM hedged upward.

We've covered this dynamic in detail in Andromeda plus iOS privacy: why volume beats precision, but the one-line version is: signal loss didn't kill Meta ads, it just changed the unit of signal from "user identity" to "creative response." Volume is the only way to feed that new unit.

Why Bidding Tricks Don't Fix Rising CPMs

This is the section that will save you the most money if you internalize it. The instinct, when CPMs rise, is to mess with bidding — switch to bid cap, lower cost cap, change from highest-volume to highest-value, etc. We've audited hundreds of these. Almost none of them work in 2026. Here's why.

Lowering your bid: You stop winning auctions. Delivery collapses. Your account goes into the under-delivery state, which Meta interprets as a low-quality signal, and your CPMs rise further when you try to come back. Net effect: less spend, higher CPM, worse CPA.

Switching to bid cap or cost cap: These are useful only if you actually know the marginal value of a conversion to your business with high confidence. Most lead-gen accounts don't. The cap gets set too low, delivery starves, the account never escapes learning. You end up running 30% of your intended budget on creative the algorithm couldn't optimize. CPM-on-spent-budget looks fine; CPM-on-needed-budget is much higher because you have to make up the gap with worse-targeted retargeting.

Manual bidding: Andromeda's whole design is to outperform manual bidders by evaluating 10,000x more candidates than a human can mentally model. Manually bidding against it is like manually trading against a high-frequency trading desk. You will lose on average, and you'll pay a CPM premium for the privilege.

Audience narrowing: Narrowing the audience reduces the candidate pool Andromeda can retrieve against. The auction becomes a smaller, hotter version of itself. CPMs go up, not down, on narrow audiences in 2026 — the opposite of the 2018 playbook. Detailed targeting is dead, and the proof is in the CPM tab.

Placement carve-outs: Removing placements (e.g., turning off Audience Network) used to lower CPM modestly because you were trading volume for quality. Post-Andromeda, placement decisions are made dynamically per impression, and your carve-out forces the auction into a smaller subset of inventory, which raises CPMs in that subset.

The pattern: every classic CPM-reduction tactic from 2018–2022 either does nothing or makes the problem worse under Andromeda. The system is built around creative-level signal, and your bidding levers don't generate creative-level signal.

The Volume Math: Why More Creatives = Cheaper Impressions

Here's the part most buyers haven't fully internalized. CPM is not one number. It's a blended number across all the impression slots Meta delivers you to. Each slot has its own clearing price.

Imagine the impression universe sorted by clearing price, cheapest to most expensive:

The cheapest 30% of impressions are users who match many advertisers' predicted-conversion profiles. They get filled at low CPMs. To get into these slots, your ad needs to be at least a plausible match for that user.
The middle 50% of impressions are users with mixed signals. CPM is moderate. Getting in requires a stronger predicted-value pitch from your ad.
The most expensive 20% of impressions are high-intent, high-signal users — small audience, lots of advertisers fighting for them. Premium CPMs.

When you run 5 creatives, Meta can only pitch you into the slots where one of those 5 happens to be the best match. That's almost always the middle of the distribution — because your 5 hand-crafted ads are written to broad mid-funnel positioning. You get almost no exposure to the cheap 30%, because you don't have a creative variant that fits those niche moments.

When you run 60 creatives — different hooks, different angles, different visual styles, different formats — Meta has 60 chances per impression to find a fit. Now you start winning slots in the cheap 30% because one of your 60 is, by chance, a good fit for a weird sub-pocket of users at 2am on a Tuesday. The cheap 30% pulls your blended CPM down.

That's the entire mechanism. It's not magic. It's not a hack. It's that the algorithm can only target you into impression slots where it has a creative to deliver. More creatives = more slots accessible = more cheap slots included in the blend = lower blended CPM.

This is also why filming one perfect hero ad is dead as a strategy. We've written about that in detail in why filming one perfect ad is dead in 2026 — short version, that one perfect ad gets you the same narrow slice of the distribution every time, and Meta can't help you escape it.

Case Math: $10K/Month Lead-Gen Account

Let's make this concrete with a worked example from a real account profile (anonymized, but the numbers are representative of dozens we've audited).

Setup:

$10,000/month Meta ad budget
Lead-gen objective, home services vertical
US-only, 4 metro areas

Before — 6 active creatives, traditional production:

Metric	Value
Active creatives	6
Average CPM	$42
Impressions delivered	~238,000
CTR (link)	1.1%
Clicks	~2,620
LP conversion rate	6.5%
Leads	~170
CPL	$58.80
Frequency at month end	4.3

After — 60 active creatives, batch production via batch video ads:

Metric	Value
Active creatives	60
Average CPM	$29
Impressions delivered	~345,000
CTR (link)	1.4%
Clicks	~4,830
LP conversion rate	6.5% (LP unchanged)
Leads	~314
CPL	$31.85
Frequency at month end	2.1

Deltas:

Blended CPM down 31%
Impressions up 45%
CTR up 27% (Andromeda matching better creative to each impression)
Leads up 85%
CPL down 46%
Frequency dropped from 4.3 to 2.1 — meaning the account is no longer fatiguing

The offer didn't change. The landing page didn't change. The audience setting didn't change (still Advantage+ broad). The only thing that changed was creative volume. The CPL improvement is what's funding the production cost, plus a substantial margin.

This is not a cherry-picked case. It's a representative case. The variance is in how much improvement, not whether improvement happens — and the dominant driver of that variance is how creative the new ads actually are vs. variations of the same script.

The Production Bottleneck Is Real (And Solvable)

Here's the honest objection: "60 creatives a month is impossible at our production budget."

True, if you're using a traditional production agency. A polished 30-second spot at agency rates is $1,500–$5,000. Sixty of those is $90K–$300K. Not happening.

But the unit you need is not a polished 30-second spot. The unit you need is a testable creative variant — a distinct hook, angle, or visual treatment that gives Andromeda something new to evaluate. Those can be produced for $5–$50 each through batch AI production pipelines. We covered the full economics in our batch video ads complete guide, but the short version: 60 batch creatives costs $300–$3,000, fits well inside the CPL savings, and ships in 24–72 hours instead of 4–8 weeks.

The two-tier production strategy that works:

80% of your creative volume is batch-produced — AI scripting, modular templates, hook variations, fast turnaround. These are your "fill the impression distribution" creatives.
20% of your creative volume is hero-produced — higher craft, human-led, derived from the top performers identified by the batch tier. These are your scaling creatives.

You can also fold this into a broader content engine if you're already running done-for-you social media — much of the organic pillar capture footage is usable as ad raw material with minor reformatting, which drops your effective per-ad production cost further.

What Rising CPMs Look Like On Your Account (And How To Diagnose)

If you're trying to figure out whether your rising CPMs are an Andromeda-era creative-volume problem or something else, run this diagnostic in Ads Manager:

1. Pull a 90-day CPM trend, broken out by week.

If CPMs are steadily climbing but CTR is flat, you have a creative-fatigue problem. Andromeda is paying more to deliver the same ads to incrementally worse-matched users.

2. Count your active creatives per ad set.

If it's under 10 per ad set, you're in the bottom quartile of advertisers by creative volume. Your CPM premium is mostly a low-creative-volume tax.

3. Check your frequency.

Above 3.0 means Andromeda is forced to re-show your ads to the same users because it has nothing new to deliver. Below 2.0 means Meta has enough variety to keep finding fresh users.

4. Look at your CPM by placement.

If CPM is uniformly high across placements (Feed, Reels, Stories, Audience Network), it's a creative issue, not a placement issue. If CPM is dramatically high on just one placement, you've likely got a format-fit issue (e.g., square ads getting punished in vertical-first placements).

5. Compare your CPM trend to your category benchmark.

If your category benchmark is up 20% and yours is up 60%, you've got an account-specific issue — usually creative volume — on top of the category-wide drift.

The diagnostic outcome for ~80% of accounts we audit: rising CPMs are driven by creative under-supply relative to budget. The fix is volume, not bidding or audience.

What This Means For Your 2026 Media Plan

Three changes to make before next quarter, in order of impact:

Reframe your production budget from "cost per ad" to "creative variants per month." Target 40–80 variants/month at $10K spend, scaling linearly with budget. The right unit is creative density per dollar of media spend, not aesthetic polish per finished asset.
Stop manually optimizing audiences. Move to Advantage+ Audience with broad targeting and let Andromeda do its job. Save the audience-tweaking hours for creative ideation. We get into the audience implications in how many ad angles do you need post-Andromeda.
Build a hook bank, not an ad library. Your asset isn't the finished ad; it's the underlying inventory of hooks, angles, and treatments that lets you regenerate 50 ads in a week when fatigue hits. Most agencies don't think this way. They sell you 5 finished ads. You need a hook bank, plus a production pipeline that can render variants on demand.

Frequently Asked Questions

Are Meta CPMs going to keep rising in 2026?

Yes, on average. The structural forces — more advertisers, Andromeda's predicted-conversion weighting, ongoing iOS signal degradation — aren't reversing. Individual accounts can absolutely buck the trend by running enough creative volume to access cheaper impression slots, but the unweighted category benchmark CPMs will continue climbing year over year. Plan for a 10–15% category CPM rise in 2026 and budget accordingly.

How many creatives is "enough" in 2026?

Rough rule: 6–10 active creatives per $1,000/week of spend, with refresh cadence weekly. So a $10K/month account should be running 60–100 active creatives at any given time, refreshing 15–25 per week. Below that, you're paying a creative-supply tax. We get more granular by vertical in how many ad angles do you need post-Andromeda.

Will switching to Advantage+ lower my CPMs?

On average, yes — usually 8–18% — but only if your creative volume is sufficient. Advantage+ shifts the auction toward Andromeda's preferred surfaces, but Andromeda still needs creative variety to deliver well. Switching to Advantage+ with 4 creatives often produces a temporary CPM dip followed by a steeper climb as the algorithm exhausts your creative pool.

Does the "broad audience" advice apply to small geos like home services?

Yes, with adjustments. For a roofing company serving 3 zip codes, "broad" means "all adults in those zip codes who own homes," not "national broad." Detailed-targeting interest stacks (e.g., "homeowner + recently moved + interested in home improvement") consistently underperform geo + age + homeowner + Advantage+ Audience. The vertical-specific deep dive lives in our Andromeda impact on HVAC, plumbing, and roofing post.

How much should creative production cost relative to ad spend?

A reasonable benchmark for 2026 is 8–15% of ad spend allocated to creative production. So $10K/month media spend → $800–$1,500 production budget for 40–80 creative variants. That's roughly $15–$30 per variant on a batch pipeline. Production budgets above 25% of media spend usually indicate over-investment in polish at the expense of volume, which is a worse trade in the Andromeda era.

Can I just keep running the same winning ad and not refresh?

Short-term, sometimes. Long-term, no. Even a winning creative fatigues — frequency climbs, CTR decays, and Andromeda's confidence in delivering it to fresh users drops because it's already shown the ad to most of the matched pool. Plan for a 30–45 day useful life on a hero creative at meaningful spend levels, after which the CPM penalty for refusing to refresh exceeds the value of holding onto the winner.

Is this the same as the "creative is the new targeting" framing?

Yes — they're the same underlying claim viewed from different angles. "Creative is the new targeting" is the strategic framing: don't try to target audiences, target creative angles. "CPMs rise without creative volume" is the financial framing: here's what it costs you when you don't. Same mechanism, two different ways to convince a CFO. Pair this post with creative is the new targeting if you need to brief leadership.

The Bottom Line

Rising Meta CPMs in 2026 aren't a glitch or a temporary inflation spike. They're the price tag Andromeda puts on creative under-supply. Every classical mitigation — bid lower, narrow audience, swap placements, manual bidding — is a 2018 reflex that the current system is built to ignore or punish.

The single intervention that consistently moves blended CPMs down is more distinct creative, fed in at a cadence that keeps Andromeda's prediction model fresh. The production cost of that volume is well inside the savings on CPL, and the operational shift — from filming hero ads to running a batch creative pipeline — is the difference between a 2026 ad account that compounds and one that bleeds.

If you want to see the production side running on your account, start with batch video ads for the creative engine, or layer in done-for-you social media if you want the organic distribution to compound alongside the paid side. Either way: stop optimizing bids. Start optimizing creative density.