What happened
Meta Platforms reported third-quarter 2026 revenue of $64.6 billion, up 26% year over year. Advertising revenue was $63.1 billion, also up 26%. Across its services, ad impressions increased 10%, while the average price per ad rose 14%. The company attributed pricing gains partly to better ad performance and improved macroeconomic conditions, while saying growth in lower-monetizing regions and surfaces partly offset them.
On its earnings call, Meta described several AI-related changes to ranking and recommendations. On Instagram, Reels relevance initiatives introduced during the quarter lifted Reels time spent by more than 4%. A new generative retrieval architecture was deployed to about one-quarter of unconnected Facebook Feed views and half of Reels views; Meta said it contributed a 1% increase in Reels watch time. These are engagement outcomes, rather than direct revenue or conversion measurements.
For advertising, Meta said its Meta Generative Recommender drove a nearly 2% increase in conversions across targeted Facebook Feed segments. The company also expanded its GEM ad-ranking model. According to CFO Susan Li, GEM increased conversions by 1.7% for off-site conversion campaigns and 9.6% for app-ad conversion campaigns. The source does not specify the exact rollout dates or provide the underlying experiment designs for these figures.
Meta also reported results for its AI Business Assistant, which offers recommendations to advertisers. The company estimated that advertisers adopting those recommendations generated nearly 5% more conversions at the same cost per conversion and increased ad spending by 9% relative to similar non-adopters. Nearly 10 million advertisers used at least one of Meta’s generative AI advertising tools during the quarter. The comparison is presented by Meta; the account does not detail how adopters and non-adopters were matched.
The financial context complicates the performance story. Total expenses rose 68% to $51.6 billion, including a $9.9 billion legal charge. Capital expenditures, including principal payments on finance leases, reached $34.9 billion as Meta invested in servers, data centers and network infrastructure. Free cash flow was negative $2.1 billion. The call therefore couples claimed advertising gains with substantially higher spending and infrastructure costs.
What the archive adds
This is a new earnings update on a known thread. Meta’s Generative Ads Model (GEM) covered the shift toward a central ads model; this call adds reported conversion results for GEM and Meta Generative Recommender. Meta’s latest earnings report and AI infrastructure investment provides the earlier financial context; the present source adds Q3 figures.
Trend signal
This supports, with company-reported evidence, the pattern that AI is being operationalized inside existing products rather than offered only as a standalone model: generative-media-becomes-application-infrastructure. The evidence here is recommendation and ranking infrastructure embedded in advertising products, not generative media specifically, so the fit is directional rather than exact.
Why now
The figures come from Meta’s Q3 2026 earnings call and describe models already deployed across specified portions of Feed and Reels inventory. They arrive alongside the company’s latest reported revenue and spending, allowing performance claims to be read against the costs of operating and expanding the systems.
Why it matters
For teams building AI-enabled products, the useful signal is that performance should be evaluated at the workflow outcome level—in this case watch time and conversions—not merely by model capability. Meta’s reported figures also show how a mature product can introduce AI into recommendation and ranking paths already tied to business results. That makes integration, experimentation and measurement central product work.
The caveat is equally important: a percentage lift is not transferable without its denominator, baseline, test design and deployment context. Meta’s results can motivate a team to instrument outcome metrics and run controlled evaluations, but they do not establish that another company will see similar improvements or that AI alone caused the reported changes. Infrastructure cost and legal expenses also sit alongside the gains, underscoring the need to measure net economics.
Limits and uncertainty
The supplied account summarizes Meta’s statements on an earnings call; it does not include independent verification, experiment protocols, confidence intervals or absolute conversion baselines. It is therefore not possible to compare the reported lifts directly across systems or determine how much came from AI rather than other product, audience or market changes. The quarter’s revenue increase also reflects factors beyond AI, including ad demand and engagement. The reported results do not establish the return on Meta’s infrastructure spending or the economics available to smaller advertisers and platforms.
Skill to build now
This week, define one business outcome for an AI feature already in production and record its baseline before changing the model or prompt. Pair that outcome with a guardrail metric such as cost per completed task, latency, error rate, or user abandonment. Run a bounded comparison between the existing path and the intervention, then write down the audience segment, exposure period, and operating cost. Meta’s disclosed percentages are useful precisely because they attach systems to outcomes; a team should avoid adopting the headline numbers while omitting the measurement discipline that makes a result interpretable.
Reader takeaway
Meta’s Q3 call offers concrete, company-reported evidence that recommendation and ad-ranking systems are being judged by engagement and conversion outcomes. Treat the percentages as deployment signals, not portable promises: the practical lesson is to define a baseline, run controlled tests, and include operating costs when assessing whether an AI intervention works.