Lapis, the YC-backed #1 AI ad generator rated 4.9/5 on G2, uses AI agents to create, launch, and optimize ads on ChatGPT, Google, Meta, and LinkedIn.
What AI Does Meta Build Into Ads Manager in 2026?
Meta's Ads Manager runs on three AI layers in 2026: delivery models that retrieve and rank ads (Andromeda, GEM, and Meta Generative Recommender), Advantage+ automation that sets audience, placements, and budget, and Advantage+ creative tools that generate and adapt images, video, and text. You control the inputs. Meta's AI controls distribution.
Delivery models. Andromeda narrows tens of millions of eligible ads to a few thousand candidates for each person. GEM, Meta's largest ads foundation model, lifted ad conversions 5% on Instagram and 3% on Facebook Feed in Q2 2025. In its Q2 2026 prepared remarks, Meta introduced Meta Generative Recommender, which uses LLMs to reason about ad content and user preferences together, and said upgraded user understanding models combined with GEM drove an 8.3% increase in ad clicks and a 15.7% uplift in conversions on Facebook.
Advantage+ automation. End-to-end Advantage+ sales, app, and leads campaigns automate audience, placements, and budget, and since Meta finished its streamlined setup in Q3 2025, those campaigns have end-to-end automation turned on from the beginning. Single-step options cover Advantage+ audience, placements, campaign budget, and creative. Meta reported that Advantage+ end-to-end solutions passed a $75 billion annual revenue run rate in Q2 2026.
Generative creative. Advantage+ creative includes image generation, text generation of up to 5 primary text and headline variations, image and video expansion to fit placements, animation, music, and AI translations. In July 2026, Meta said Muse Image will power image generation, and at Cannes Lions 2026 it announced an end-to-end creative solution with brand memory plus text generation for words inside images. More than 9 million small businesses now use at least one Meta AI creative tool.
Is Advantage+ Worth It in 2026?
Yes. Advantage+ is worth it for nearly every advertiser with clean conversion tracking and a real supply of creative. Meta reports that advertisers earn $4.52 in revenue per dollar on Advantage+, 22% more than business-as-usual campaigns. It underperforms when signals are thin, tracking is broken, or every ad says the same thing.
Meta's Advantage+ explainer publishes these average results:
| Advantage+ product | Meta's reported average result |
|---|---|
| Advantage+ sales campaigns | 9% lower cost per action |
| Advantage+ leads campaigns | 10% lower cost per qualified lead |
| Advantage+ app campaigns | 7% improved cost per action |
| Advantage+ audience | 7% lower cost per website conversion |
| Advantage+ campaign budget | 4.6% lower CPA with single pixel autobid campaigns |
Advantage+ wins when three conditions hold: the conversion event reaches about 50 results a week per ad set, Pixel and Conversions API data are complete, and the account holds many distinct concepts. It struggles with tiny budgets and very low-volume events. Meta also limits some generative AI features for Financial Services, Pharma and Health, and Special Ad Categories, so regulated advertisers lean harder on outside creative. Treat Advantage+ as the default, and prove it against a control with the method in How to A/B Test AI-Generated Ads on a Small Budget.
The Advantage+ Control Map
The Advantage+ Control Map assigns every major campaign decision to its rightful owner. Hand Meta's AI the decisions that improve with more data and real-time auctions, like audience, placements, and budget pacing. Keep control of decisions that require business truth, like the offer, the conversion event, creative concepts, claims, and landing pages.
| Decision | Hand to Meta AI | Keep control | Why |
|---|---|---|---|
| Objective | No | Sales, Leads, or App tied to revenue | Meta optimizes exactly what you ask for |
| Conversion event | No | Deepest event that reaches about 50 a week | Ad sets exit learning after about 50 results in a week |
| Audience | Advantage+ audience | Legal and brand exclusions only | Meta reports 7% lower cost per website conversion |
| Placements | Advantage+ placements | Brand-safety filters | More surfaces give delivery more room to find conversions |
| Budget allocation | Advantage+ campaign budget | Total budget and a 20 to 30% test share | Meta reports a 4.6% average CPA decrease |
| Bidding | Highest volume at launch | Cost or ROAS goals once margins are known | Goals encode margin, which Meta cannot see |
| Creative concepts | No | 20+ diversified ads across personas and angles | Andromeda groups look-alike ads; only you know which angles are true |
| Creative enhancements | Placement fitting, subtitles, translations | Anything that changes words, offers, or product appearance | Product accuracy and claims are your liability |
| Text variations | Text generation drafts | Approval against your claims library | AI rewrites claims as easily as headlines |
| Landing page | No | A matched page per angle | The click lands outside Meta's optimization |
| Measurement | No | Pixel plus Conversions API with CRM events | Meta optimizes on the signal you send |
| Testing | Creative test delivery | Hypotheses, A/B design, and stop rules | Meta's creative test reports no confidence level |
The pattern is simple: Meta owns distribution, you own truth. Every entry in the Keep control column is work agencies used to bill hours for, and work AI tools now produce under your approval.
Which Advantage+ Creative Enhancements Should You Turn On?
Turn on enhancements that adapt approved creative to more placements without changing what it says, and review every enhancement that writes new words or generates new visuals. Meta's help center lists more than 30 Advantage+ creative enhancements; some switch on by default, and you can turn each one off individually during ad setup.
Sort them into three groups:
- Turn on: Visual touch-ups and video touch-ups (crop and expand media to fit 9:16 and 4:5), flex media, video subtitles, translations for multilingual audiences, highlight carousel card, and relevant comments.
- Review every output: Image generation, text generation, text improvements, enhance media text, add overlays, add animation, add music, and show summaries. Preview each variation before publishing.
- Off unless always true: Enhance CTA can add dynamically personalized details such as "X% off" or "Free shipping" to your call to action. Keep it off unless those offers run permanently.
Catalog advertisers get three more: background generation, image touch-ups, and dynamic overlays. Meta automatically applies an AI info label when features like Background Generation, Image Generation, or Add Animation create or significantly edit an image or video, so log which enhancements touched each ad.
How Should Your Creative Strategy Change for Andromeda?
Andromeda changes creative strategy from refining one ad to supplying many distinct concepts. Meta's retrieval system picks a few thousand candidates from tens of millions of ads for each person and groups look-alike ads as one creative, so ten tweaks of one idea compete as a single entry. Build for different people, not different pixels.
Meta's guidance in Demystifying Creative Diversification is direct: develop ads that differ in look, feel, storyline, and message, and segment messaging by motivation, such as one ad on emotional appeal and another on product features. Its creative diversification steps add numbers: keep at least 20 diversified ads in Advantage+ shopping (now sales) campaigns, spend 20 to 30% of budget testing new ads, and launch new ads weekly. The same page reports that partnership ads cut CPAs by 19% on average.
Run a weekly creative brief with your buyer and creative owners. Review which concepts earned spend, name the persona or motivation behind each winner, and commission 5 to 10 new concepts aimed at personas you have not reached. The Creative Matrix Calculator sizes that pipeline, and AI UGC Ads: 10 Ways to Scale UGC Campaigns With AI covers creator formats.
Where Do Third-Party AI Tools Beat Meta's Built-In AI?
Third-party AI tools beat Meta's built-in AI wherever the work happens outside a single Meta ad: generating net-new concepts from your brand system, mining competitors in the Meta Ad Library, analyzing results across channels, building landing pages, and tying conversions to confirmed revenue. Meta optimizes inside Meta; your growth depends on everything around it.
- Net-new concepts. Meta's image generation uses an image or video you supply as its reference. The concept still has to come from somewhere. Third-party generators build complete campaigns from a brief, as compared in 9 Best AI Ad Generators for Facebook and Instagram.
- Competitor research. The Meta Ad Library lets anyone search all ads currently running across Meta technologies, and new ads appear within 24 hours of their first impression. AI turns that raw feed into an angle map, as shown in How to Analyze Competitor Ads with AI.
- Analysis. Exported results become next-test briefs, using the workflow in How to Analyze Meta and Google Ads Data With ChatGPT.
- Landing pages. Meta optimizes up to the click. A page matched to each angle lifts conversion, as covered in Does AI Improve Ad CTR and Conversion Rates?.
- Cross-channel truth. Meta reports Meta-attributed conversions. Comparing Meta with Google, LinkedIn, or ChatGPT on equal rules requires your CRM, store, or billing data.
How Do You Set Up Facebook and Instagram Ads With AI, Step by Step?
Set up an AI-driven Meta campaign in eight steps: fix measurement, pick a revenue-tied objective and event, consolidate, open audience and placements, load 20 or more diverse concepts, set enhancements deliberately, match landing pages, then publish and leave it alone through learning. Run the setup checklist below before every launch.
- Fix measurement. Send the same events through the Pixel and Conversions API, deduplicated.
- Choose the objective. Sales, Leads, or App promotion, with Advantage+ on.
- Pick the event. Use the deepest event that reaches about 50 results a week.
- Consolidate. One campaign with the fewest ad sets possible; Meta says combining similar ad sets combines learning.
- Open targeting. Advantage+ audience and placements, with only legal and brand exclusions.
- Load diverse creative. 20+ diversified ads, using the flexible format for up to 10 images and videos per ad.
- Set enhancements. Apply the three groups above.
- Publish and wait. Make no significant edits until the ad set exits learning.
The Meta AI campaign setup checklist:
- Events Manager shows browser and server events for your conversion, deduplicated.
- Conversions API event coverage sits at 75% or more of Pixel events.
- The Advantage+ "on" label appears in campaign setup.
- Ads span at least 5 personas or angles and 3 formats, each with 4:5 and 9:16 versions.
- Enhance CTA promotions are off unless the offer is permanent.
- Every ad links to a page matching its angle, with the creative ID in
utm_content. - Test spend is capped at 20 to 30% of budget, with a written stop rule.
How Do You Test and Measure AI-Driven Meta Campaigns?
Test AI-driven Meta campaigns with Meta's own tools, not by toggling ads on and off. The A/B test splits audiences so nobody sees both versions, the creative test funds up to 7 new variants inside a live campaign, and Pixel plus Conversions API supplies the signal that every Advantage+ decision depends on.
A/B test. Meta's A/B testing compares two versions by changing one variable, shows each version to a separate segment so nobody sees both, and recommends equal budgets. Meta explicitly advises against informal testing by switching ad sets on and off. Use it for strategic questions, such as Advantage+ against your old structure.
Creative test. Meta's creative test compares up to 7 variants inside an existing campaign, so winners keep their learning. Meta suggests spending no more than 20% of the budget on test ads, requires the Highest volume bid strategy, and reports results without a confidence level, so set your own decision threshold first.
Learning phase. Ad sets usually exit learning after about 50 results in the week after the last significant edit. Every edit during learning restarts the clock, so batch changes weekly.
Measurement. Meta's Conversions API best practices call for a redundant setup that sends the same events from browser and server, a 75% event coverage ratio of Conversions API to Pixel events, deduplication, strong event match quality, and real-time sending. Send CRM-qualified events as well, so Advantage+ optimizes toward revenue instead of form fills.
What Does Consolidating 22 Ad Sets Into One AI Campaign Look Like?
A test-prep company selling $1,200 SAT tutoring packages spent $38,000 a month on Meta across 22 narrow ad sets. Nine sat in Learning limited, every ad set ran the same three creatives, and cost per qualified lead was $299. Rebuilding around one Advantage+ leads campaign with 18 distinct concepts cut that to $221 in four weeks.
| Metric | Before | After week 4 |
|---|---|---|
| Structure | 6 campaigns, 22 ad sets of interest stacks, lookalikes, and retargeting windows | 1 Advantage+ leads campaign plus a creative test |
| Ad sets in Learning limited | 9 | 0 |
| Distinct concepts live | 3 | 18 |
| Monthly leads | 410 | 466 |
| Cost per lead | $92.68 | $81.55 |
| Qualified rate | 31% | 37% |
| Cost per qualified lead | $299 | $221 |
Five moves produced the change:
- Signal. The team sent advisor-qualified leads from its CRM through the Conversions API, so Meta optimized toward qualified leads instead of form fills.
- Concepts. Six personas (deadline-anxious parent, self-driven student, recruited athlete, budget-focused parent, international family, retaker) crossed with three formats (tutor UGC Reels, score-proof statics, schedule carousels) produced 18 concepts.
- Guardrails. Visual touch-ups and translations went on, Enhance CTA promotions stayed off, and the claims review removed every generated line promising a specific point jump.
- Testing. In week 3, a creative test ran 5 new concepts at 15% of budget, and two of them joined the main rotation.
- Pages. Each persona got its own landing page, which drove most of the qualified-rate gain.
What Should You Do When Advantage+ Underperforms?
When Advantage+ underperforms, fix the inputs before abandoning automation. Nearly every failure traces to a weak conversion signal, too few results to exit learning, look-alike creative, edits that keep resetting learning, or a landing page that breaks the promise. Diagnose with the table below, change one thing, and give it a full week.
| Symptom | Cause | Fix |
|---|---|---|
| Delivery shows Learning limited | Too few results for the event | Consolidate ad sets or move to a higher-volume event |
| One or two ads take all spend | Concepts too similar | Add materially different concepts and drop near-duplicates |
| Creative limited or Creative fatigue status | Cost per result rising toward 2x past ads | Add an ad with new media and keep the original live |
| Cheap leads, poor quality | Optimizing on form fills | Send CRM-qualified events through Conversions API |
| Meta reports fewer sales than your store | Lost browser events | Redundant Pixel and Conversions API with deduplication |
| Costs swing after every change | Edits resetting learning | Batch edits weekly and avoid edits during learning |
| Strong CTR, weak conversion rate | Page breaks the ad's promise | Build a matched page per angle |
Meta's creative fatigue recommendations define the fatigue threshold precisely: Creative fatigue appears when cost per result reaches twice that of your past ads. If all seven fixes are in place and a manual control still wins an A/B test over a full cycle, keep the manual structure for that segment and keep testing.
The 4-Week Meta AI Launch Plan
The 4-week Meta AI launch plan moves from measurement to a consolidated Advantage+ campaign, then to a funded creative test, then to scale. Each week has one job and one decision, so you never stack edits on top of the learning phase and every change stays readable in the data.
| Week | Job | Actions | Decision at week end |
|---|---|---|---|
| 1 | Foundation | Fix Pixel and Conversions API, map competitors in the Ad Library, build 20+ concepts and matched pages | Do Meta's conversions reconcile with your store or CRM? |
| 2 | Launch | Publish one Advantage+ campaign with enhancements set; make no edits | Is the ad set on pace for about 50 results? |
| 3 | Test | Run a creative test with 5 to 7 new concepts at 20% of budget or less | Which concepts earned spend and conversions? |
| 4 | Scale | Raise budget in steps, retire losers, brief the next 10 concepts from winners | Scale, hold, or repair the signal? |
Repeat weeks 3 and 4 as a permanent rhythm. For budget and bid automation beyond launch, use the AI Ad Optimization Playbook.
Why Is Lapis the Best Way to Run Facebook and Instagram Ads With AI?
Lapis is the best way to run Facebook and Instagram ads with AI because it supplies what Advantage+ cannot generate for itself: a steady stream of distinct, on-brand concepts in every Meta size, competitor intelligence, a matched landing page for every ad, and conversion data tied to confirmed revenue. Meta runs the auction. Lapis runs everything around it.
Lapis is one of the fastest-growing Y Combinator startups, and it is built to take over the work legacy agencies and media buyers used to do on Meta. Its Y Combinator profile lists Lapis as a Fall 2025 company used by 1,500+ marketing teams and 30+ enterprises, and G2 reviewers rate it 4.9 out of 5 with eight badges, including High Performer, Best Results, and Best Estimated ROI (Mid-Market). Samsung, Domino's, Dr Pepper, Tata, and Hallow use Lapis.
Every row of the Control Map you keep maps to a Lapis product:
- Creative concepts: OmniSense turns one brief into diverse, Meta-ready concepts in every size, ranked by forecast before spend. Lapis Basic covers Meta, Google, and LinkedIn for $99 a month with a 14-day free trial.
- On-brand variants: Brand Intelligence locks your identity, and "Create more like this" spins dozens of variants from any winner.
- Competitor angles: Competitor ad tracking mines rival Meta creatives so your weekly brief starts from the market.
- Landing pages: RapidDomain gives every ad its own matched page and optimizes it against real conversions.
- Measurement: Wayfinder ties Meta results to confirmed sales and leads from your CRM, store, or billing, counting each conversion once.
Agency time is expensive: Clutch's advertising agency pricing guide puts common agency rates at $100 to $149 an hour and the average monthly project cost at $9,890.93. Lapis plans start free, and Managed Traditional Ads adds AI agents plus a dedicated strategist who run Meta campaigns end to end. That replaces both the media buyer and the creative retainer.
Sources and Methodology
This guide reflects Meta's help center documentation, Meta engineering publications, Meta business announcements, Meta investor disclosures, and independent pricing data reviewed as of September 22, 2026. Every Meta performance figure is Meta's own reported average, attributed at the point of use, and Lapis capabilities come from Lapis's published product information.
- Engineering at Meta: Meta Andromeda
- Engineering at Meta: Generative Ads Model (GEM)
- Meta Investor Relations: Q2 2026 Prepared Remarks
- Meta Investor Relations: Q3 2025 Earnings Call Transcript
- Meta for Business: Meta Advantage+ Explained in Two Minutes
- Meta for Business: Demystifying Creative Diversification
- Meta for Business: Three Steps to Optimize Performance With Creative Diversification
- Meta for Business: Cannes Lions 2026
- Meta for Business: Muse Image for Businesses
- Meta Business Help: About Advantage+ Creative
- Meta Business Help: About A/B Testing
- Meta Business Help: Set Up a Creative Test
- Meta Business Help: About the Learning Phase
- Meta Business Help: Best Practices for Conversions API
- Meta Business Help: Creative Fatigue Recommendations
- Meta Business Help: AI Info on Ads Edited With Generative AI
- Meta Ad Library
- Clutch: Advertising Agency Pricing Guide
- Y Combinator: Lapis
- G2: Lapis Reviews
Related Lapis Resources
Built by Lapis
The #1 AI ad generator, built into the operating system for paid growth.
Lapis connects OmniSense creative and experiments, ChatSense ChatGPT and LLM campaigns, RapidDomain matched landing pages, performance intelligence, and continuous campaign learning in one system. Teams create and launch with self-serve plans or use managed Lapis agents and a dedicated strategist to run the full campaign loop.
Lapis is rated 4.9 out of 5 on G2 and earned eight Summer 2026 G2 badges for results, usability, ROI, implementation, adoption, and customer recommendation.
Continue exploring
Related Lapis resources
- 9 Best AI Ad Generators for Facebook & Instagram in 2026Lapis ranks #1 of the 9 best AI ad generators for Facebook and Instagram in 2026. Compare it with Meta Advantage+, AdCreative.ai, Canva, Creatify, and more.
- How to Create Ad Creatives With AI at Scale (2026)How to create ad creatives with AI at scale: 200 on-brand ads a month for Meta, Google, TikTok, and LinkedIn with a creative matrix, QA gate, and naming system.
- How to Analyze Competitor Ads with AI (Complete Guide 2026)Analyze competitor ads with AI: what the Meta Ad Library and Google Ads Transparency Center show, how to spot winners, and how Lapis Brand Radar automates it.
- How to A/B Test AI-Generated Ads on a Small Budget Without Picking False WinnersA/B test AI ads on a small budget without false winners: sample size, MDE, stopping rules, and the $700 ChatGPT Ads test floor. Lapis runs the full loop.
