Meta's September quarter shows AI-driven targeting and ranking lifting both how many ads it serves and what advertisers pay for each one. That closes the gap with Google Search. The same quarter also showed a steep rise in costs and a business that depends almost entirely on advertising. The revenue side is well measured. The return on the spending behind it is not. The question is whether AI-led pricing power can keep outrunning the cost of producing it. The stock's swings in the period say the market has not decided.
What the quarter established
Start with what many sources agree on. Ad impressions rose 10% 11,15,16,18,22,25,29,56. Average price per ad rose 14%, the most repeated data point in the set, at five sources 15,17,18,22,56. Advertising revenue growth is quoted at 27% across five sources 3,5,9,11,12. Other items give 26% 23,45 or 28% 37. This looks like rounding or a different measure, such as "core" advertising 53, not a real conflict. North America accelerated to 33% 23,56. Europe grew 18% 30.
Two details temper the headline.
- Impressions slowed. Q2 impressions growth was 14% 4,10,13,25, so Q3 is a deceleration, even though one poster describes three consecutive double-digit quarters 50.
- Pricing strengthened. Q1 price growth was 12% 1,4,11,13,16,25.
One poster calls simultaneous double-digit growth in impressions and price a first in Meta's history. The poster adds that the two metrics usually offset each other 50. That is a single-source assertion, so treat it as interpretation, not established fact.
The cause of the higher pricing is also contested. Management credited better-performing ads and a better economy than a year earlier 19. JPMorgan puts more weight on AI-driven targeting 24. That matters. A price increase can come from better ads, a stronger economy or both. Only the first is a durable platform advantage.
Why AI gets the credit, and how much of it is proof
The supporting evidence is mostly management-reported performance gains:
- Ranking improvements drove 8.3% more clicks and 15.7% more conversions on Facebook 31.
- Newer ranking models lifted conversions across Facebook and Instagram 42, with a figure of up to 9.6% cited 43,45.
- A large-language-model ad-selection system lifted conversions by nearly 2% in targeted parts of the Facebook feed 19.
- Andromeda improvements were +1.4% on Facebook 20.
- Advertisers following Meta AI Business Assistant recommendations saw roughly 5% more conversions at unchanged cost per conversion 9,27,34.
- Almost 10 million advertisers used at least one Meta AI tool in Q3 34.
The 5% result at unchanged cost per conversion is the most useful of these. It speaks directly to cost-per-acquisition integrity. An Advantage+ ROAS gain of 22% is self-reported 14. It is weaker evidence, and none of these figures is an independent incrementality test.
JPMorgan links the gains to results. It attributes them to AI-driven improvements in ad retrieval and targeting 56. It says AI-driven targeting is lifting conversions 47. It estimates that Meta's AI-boosted ads brought in about $13 billion of new revenue last quarter, roughly twice the increase in Google Search revenue 24,46,57. That is an attribution estimate. It is not a measured incremental result. It is plausible, but the method behind it is not in the material.
The Google gap
Google Search and other properties grew 12% to $63.4 billion 29. Google's overall advertising revenue grew 10.6% 42. Against Meta's 26% growth, the roughly 14-point gap is described as the widest in Meta's favor since 2021 34.
The scale of the catch-up is the notable point. Three years earlier Meta's advertising revenue was roughly 70% of Google Search's 27. In the September quarter the gap was under US$300 million 27. The two are now described as practically equal in size 34. Net of traffic acquisition costs, Google's lead was roughly $2.6 billion, down from about $9 billion a year earlier 42. JPMorgan expects Meta's Family of Apps advertising revenue to overtake Google Search from Q4, a view repeated across four sources 47,48,56,57.
Read this carefully. Google "remained slightly ahead" in the September quarter 40. The overtake is a forecast, not a reported result. The comparison also mixes definitions: worldwide ad revenue 42 against Search only. Meta is closing on a narrower measure of its rival. That is still a real result, but the headline is looser than it sounds.
The cost side and the concentration problem
The cost evidence is the counterweight. Operating margin fell from 43% to 31%, a figure repeated across nine sources and reported since July 2,5,6,7,8,15. One post gives a lower reading of 40% to 20% 25. Another cites 2025's margin at 41% 36. Total costs rose 68% 15,30. The earlier comparison was 55% 4,25, so cost growth is accelerating. COGS is up 45% 41,49 and G&A up 323% 41,49. AI buildout costs are climbing 32. A LinkedIn post citing the Financial Times says further increases are expected in 2027 37. Leverage has risen from 0.1x in 2025 28. Meta's notes due 2056 priced almost 0.9 percentage points wider than a comparable April note 21.
Revenue has little diversification behind it. Advertising is about 97.6%–98% of revenue 15,25,38. There is also a waste fraction to consider. Reuters-based documents estimate that about 10% of 2024 sales came from fraudulent or suspicious ads 35. Meta said reports of fraud ads fell 58% in 18 months and that it removed over 134 million scam ads in 2025 35. Both claims describe past practice. Their effect on current results is unclear. Advertiser pushback is anecdotal: commenters say rising costs have made returns unworthwhile 39. That is only anecdote, but it matches the reported price increases. A 14% rise in price per ad is also a 14% rise in what advertisers must earn back.
What the market made of it
The share price was volatile. Meta lost 14% in October 34 and fell 10.75% over a month 17. It fell alongside Alphabet despite the strong quarter 27. It then rose about 5% on November 9 51 and 2.9% on November 13 54, closing at $641.69 33. That sits against a late-September peak near $780 25. One article reports a 30% gain for the month 38. That conflicts with the October and one-month declines and cannot be reconciled from the material, so I do not rely on it.
Sell-side targets remain well above the price. The average 12-month target is about $858–$859 25,38. The median across 33 analysts is $860 26. The implied upside depends on the reference price: about 32% 55, 36% 25 or 42% from $605.08 38.
The sensitivity analysis is the most revealing item. An extra 1 percentage point of annual ad-price growth over 2027–2035 adds roughly $82 per share 20. The valuation therefore depends heavily on sustained ad pricing, which is the least independently verified part of the story.
Interpretation
The evidence supports one conclusion and leaves a second unproven.
Supported. Meta's AI gains show up in revenue. Impression growth, pricing growth and the narrowing gap with Google all point the same way, and they come from several sources.
Not yet shown. The size of the AI contribution, as opposed to the economy or other factors, rests mostly on management-reported and single-source figures, and some figures carry no stated period 44,52. The market is not simply rewarding growth. A 68% rise in total costs, margin compression and a revenue base that is almost entirely advertising help explain why the stock fell even as the Google comparison improved. The case for the shares depends on pricing and conversion gains continuing to outpace spending.
Takeaways
- Price and volume both rose. Q3 impressions rose 10% and price per ad rose 14%, which supports 26–27% ad revenue growth. Impressions slowed from Q2.
- The Google gap is nearly closed. JPMorgan expects Meta to overtake Google Search from Q4. That is a forecast, and the Q3 gap was still under $300 million.
- Costs are the main tension. A 68% rise in total costs and a margin falling from 43% to 31% weigh against the revenue story.
- Valuation hangs on ad pricing. Targets near $860 imply large upside. The sensitivity to ad-price growth and the roughly 98% advertising concentration leave little room for a slowdown.
The question is not whether it works, but how you know it works. If advertisers are paying 14% more per ad, how much of that is incremental sales, and how much is attribution credit for sales that would have happened anyway?