Meta Ads

What Actually Changed in Meta Ads This Year

By Sean HolleranAugust 20, 202610 min read
Smartphone screen showing social media platform app icons

Meta changed a lot this year and told almost nobody. Not in the splashy Cannes-announcement way — that part got covered. The changes that will actually affect your reporting, your campaign structure and your year-over-year comparisons shipped quietly, and a fair amount of what is circulating about them is wrong.

Here is what actually changed, what it means, and where the popular version of the story is inaccurate.

1. Your Conversion Numbers Changed Definition in March

This is the most consequential change of 2026 and the least discussed.

In March, Meta narrowed what "click-through attribution" means. It now counts only clicks on links to websites. Everything else — social clicks, and engaged views of 5 seconds or more on video — moved into a new bucket called engage-through attribution, defaulting to a 1-day window. This replaced the old engaged-view setting.

Nothing about your ads changed. The columns changed. Conversions that used to sit in one place now sit in another.

What This Breaks

Any year-over-year comparison that spans March 2026 is not apples-to-apples. If your Q2 looked worse than last Q2, check whether you are comparing the same attribution definition before you rebuild your creative strategy around a reporting artifact.

Practical guidance from Jon Loomer, who documented the change: keep 7-day click for purchase campaigns, and strip 1-day view from non-purchase goals like leads.

2. Placement Exclusions Are Going Away — With Two Big Caveats

In August, Meta began notifying advertisers that it is testing removal of the ability to exclude individual placements, platforms, devices and operating systems at the ad set level.

This got repeated across the industry as "Meta is removing placement controls." That is not quite what is happening, and the details matter:

So if your reflex is to exclude Audience Network, the replacement lever is a bid adjustment rather than an exclusion. Which, honestly, is usually the better instrument anyway — excluding a placement removes cheap impressions along with the bad ones.

3. Advantage+ Shopping Campaigns Stopped Existing as a Separate Thing

Advantage+ Shopping and Advantage+ App campaigns lost their separate API structures and were folded into a single unified Advantage+ framework. New creation was blocked in Marketing API v24.0 last October; full deprecation landed with v25.0 in Q1 2026.

If your team or your agency still says "ASC" as though it names a distinct campaign type, that vocabulary is a year out of date. Advantage+ is now one framework with a state — sales, app, or leads.

For scale context: Meta put Advantage+ at over $75 billion in annual revenue run rate on its Q2 2026 call. This is not an optional beta anymore.

4. New Pixels Now Create Events By Themselves

Meta shipped a rebuilt Automatic Events feature in Events Manager. AI detects activity on your site and creates standard events with no code.

The part to know: it is enabled by default on every pixel created on or after August 3, 2026. Existing pixels have it off by default.

If you launched a new store, a new brand, or a new pixel any time in the last month, go look at Events Manager. There is a management dashboard where you can review and delete auto-created events. The failure mode here is duplicate or inflated event counts quietly polluting your optimization signal — and if the pixel is new, you have no clean baseline to notice it against.

5. The Ranking Engine Got Rebuilt Underneath You

Meta moved its ads ranking onto a foundation-model architecture — GEM, its Generative Ads Recommendation Model. Rather than scoring ads individually, it reasons about ad content and user preference together, at LLM scale.

Meta says the results from its Q2 2026 call were +8.3% ad clicks and +15.7% conversions on Facebook, attributed to advanced user-understanding models. These are Meta's own figures for Meta's own systems. Q1 numbers included a 6% conversion rate lift for landing-page-view campaigns.

You cannot control any of this. What it means practically is that the system reads your creative more thoroughly than it used to, and the structural games advertisers used to play to steer delivery matter less than the content of the ad itself. Which brings us to the biggest practical change of the year.

6. Almost Everything You Learned About Campaign Structure Is Now Wrong

Four pieces of standard 2025 advice are now actively counterproductive:

  1. Splitting ad sets to test creative. Outdated. The current guidance is one campaign, one ad set, optimized for a conversion, with budget consolidated. Valid reasons to split are now narrow: genuinely separate business goals, distinct products needing different messaging, seasonal products.
  2. Building a dedicated retargeting campaign. Remarketing happens by default. Jon Loomer reports Meta spending 20–25% of his own budget on remarketing with no manual input, and Audience Segments reporting will show you your split.
  3. Treating detailed targeting and lookalikes as constraints. They are suggestions now, not restrictions. Meta removed the language saying they are prioritized.
  4. Excluding low-quality placements. Covered above — value rules replaced exclusions.
The structure levers got taken away one by one, and what is left is the creative. That is not a slogan. It is a description of the remaining control surface.

7. AI Creative Tools: What Shipped, and What Nobody Can Prove

Meta released Muse Image in July, its first image model out of Superintelligence Labs, and said advertisers would get access through Advantage+ creative. Advantage+ can also now rewrite the headline text baked into an uploaded image while preserving your design, colors and fonts — generating multiple alternate-angle variants from one asset.

Adoption is real. Meta said 9 million small businesses use at least one of its AI ad creative tools, with image generation adoption more than doubling in Q2.

Here is the honest part, and it matters if you are deciding whether to lean on these tools:

Worth Knowing

Every performance number published about Meta's AI creative tools comes from Meta's own earnings calls or its own case studies. No agency, ad-tech platform or independent study with a disclosed sample size has published a comparison. Meta released no ad-performance claims at all for Muse Image. If you see a Muse ROAS statistic, someone invented it.

Our own read: the text-rewrite feature is genuinely useful for squeezing variants out of a proven static, and it is worth defining your brand kit properly in Ads Manager before you use it, because output quality tracks directly to how well you have specified logo, fonts, colors, tone and restricted words. Generated variants are individually deselectable. Use it, check every output, do not ship unreviewed.

8. AI Labeling Now Happens Whether You Disclose Or Not

Meta updated its generative-AI ads transparency policy this year. The mechanics:

One correction, because this is circulating widely and is not true: there is no confirmed requirement that commercial advertisers disclose AI-generated content. The claim traces back to a page that labels its own sourcing as tertiary and its policy status as "proposed." The only Meta AI-disclosure requirement we could verify applies to political and social-issue advertisers, and has since 2024.

The strategic point is not compliance. It is that if you are running AI-generated spokespeople, the viewer is increasingly being told so in the feed, next to the word "Sponsored" — and that is a creative decision with consequences, not a paperwork question.

The Bottom Line

Three things to actually do this week:

  1. Check whether your year-over-year comparisons cross March 2026. If they do, the attribution definition changed underneath them.
  2. If you created a pixel after August 3, audit your auto-created events before they distort optimization.
  3. Stop splitting ad sets to test creative. Consolidate, and move that effort into producing more distinct creative instead.

The through-line across every one of this year's changes is the same: Meta keeps removing the levers that let advertisers steer delivery, and keeps improving the system that reads the ad itself. The advertisers who do well from here are the ones with more good creative in the auction, not the ones with cleverer account structures.

If producing that volume is the constraint, that is the problem our packages exist to solve.


Sources: Jon Loomer Digital on 2026 attribution (Mar 2026), placement control removal (Aug 2026), Automatic Events (Sept 2026), Advantage+ text rewriting (Jul 2026) and 2026 targeting (Mar 2026); Social Media Today (Aug 2026); PPC Land on legacy campaign API deprecation; Meta Q1 and Q2 2026 earnings calls; Meta Engineering on GEM (Aug 2026); Meta Newsroom, "Introducing Muse Image" (Jul 2026); Meta generative-AI ads transparency page (2026 update).

The Only Lever Left Is Creative

Meta keeps removing the ways to steer delivery. More good creative in the auction is what still works.