Incentive-Aware Multi-Fidelity Optimization for Generative Advertising in Large Language Models

Jiayuan Liu; Barry Wang; Jiarui Gan; Tonghan Wang; Leon Xie; Mingyu Guo; Vincent Conitzer

cs.GTApr 7, 2026

Incentive-Aware Multi-Fidelity Optimization for Generative Advertising in Large Language Models

Authors: Jiayuan Liu, Barry Wang, Jiarui Gan, Tonghan Wang, Leon Xie, Mingyu Guo, Vincent Conitzer

Organizations: Carnegie Mellon University · University of Oxford · Harvard University · Adelaide University

Abstract

Generative advertising in large language model (LLM) responses requires optimizing sponsorship configurations under two strict constraints: the strategic behavior of advertisers and the high cost of stochastic generations. To address this, we propose the Incentive-Aware Multi-Fidelity Mechanism (IAMFM), a unified framework coupling Vickrey-Clarke-Groves (VCG) incentives with Multi-Fidelity Optimization to maximize expected social welfare. We compare two algorithmic instantiations (elimination-based and model-based), revealing their budget-dependent performance trade-offs. Crucially, to make VCG computationally feasible, we introduce Active Counterfactual Optimization, a "warm-start" approach that reuses optimization data for efficient payment calculation. We provide formal guarantees for approximate strategy-proofness and individual rationality, establishing a general approach for incentive-aligned, budget-constrained generative processes. Experiments demonstrate that IAMFM outperforms single-fidelity baselines across diverse budgets.

Figures & tables

Panel of Figure 1 — Figure 1: Visual Over-Saturation in Generative Advertising (advertisers: Alpha sports jacket and Beta boba). (a) Baseline produces a natural image lacking commercial utility (unbranded cup and jacket). Balanced configurations naturally prioritize a single brand: (b) sharp focus on Alpha logo with a blurred Beta branded cup; (c) prominent Beta logo, pushing Alpha to the periphery. Conversely, extreme parameters (d) cause catastrophic structural collapse, hallucinating an amalgamated “AB” logo onto the boba cup.

Figure 2

Method	Mean Welfare $\pm$ Std. Dev.
Method	Low-Budget ( $\leq 16\text{k}$ tokens)	High-Budget ( $\geq 32\text{k}$ tokens)
IAMFM-MFBO	186.26 $\pm$ 5.02	190.73 $\pm$ 5.53
IAMFM-ASH	187.26 $\pm$ 6.00	187.43 $\pm$ 7.93
UCB Bandit	182.78 $\pm$ 4.15	187.04 $\pm$ 1.58
Uniform Sampling	184.26 $\pm$ 4.62	185.87 $\pm$ 2.41
IAMFM-SH	184.42 $\pm$ 10.10	185.21 $\pm$ 11.68

Table 1: Aggregated performance (mean welfare

\pm

std. dev.) stratified by budget regime. Methods are sorted by High-Budget performance. Bolded are proposed methods.

Budget	IAMFM-MFBO	IAMFM-ASH	IAMFM-SH	UCB Bandit	Uniform Sampling
500	231.25 $\pm$ 5.29	233.12 $\pm$ 1.97	231.24 $\pm$ 3.64	225.41 $\pm$ 2.14	228.95 $\pm$ 5.45
1,000	232.07 $\pm$ 5.86	230.56 $\pm$ 5.79	231.96 $\pm$ 7.39	229.12 $\pm$ 2.87	230.80 $\pm$ 5.08
2,000	233.29 $\pm$ 7.93	231.99 $\pm$ 5.99	233.19 $\pm$ 5.40	229.67 $\pm$ 4.56	230.56 $\pm$ 5.65
4,000	238.24 $\pm$ 6.85	231.43 $\pm$ 5.20	233.41 $\pm$ 5.04	233.14 $\pm$ 5.23	232.25 $\pm$ 4.85
8,000	240.26 $\pm$ 5.92	232.88 $\pm$ 5.79	231.48 $\pm$ 6.79	235.58 $\pm$ 2.16	229.98 $\pm$ 5.05

Table 2: Algorithm performance (Mean Welfare

\pm

Std. Dev.) on the Image Generation task across 20 independent trials. IAMFM-MFBO demonstrates a widening performance gap over baselines as the budget scales.

Appendix figures & tables9 assets

Supplementary material from the paper’s appendix.

Appendix

Strength	Label	Description for Prompt
0	Very Weak	minimal presence (indirect references only)
1	Weak	weak presence (subtle, background references)
2	Moderate	moderate presence (balanced, natural integration)
3	Strong	strong presence (prominent, featured examples)
4	Very Strong	very strong presence (dominant, detailed focus)

Appendix

Table 3: Mapping of numerical strength levels to prompt descriptions.

Regime	Comparison	Mean Diff.	p-value	95% CI 3 3 3 We report Welch 95% confidence intervals throughout. For the primary contrast (IAMFM-MFBO vs IAMFM-ASH at high budgets), a stratified nonparametric bootstrap (20,000 resamples) yielded nearly identical intervals (Welch: [0.21, 6.38]; bootstrap: [0.42, 6.33]), supporting the robustness of our conclusions.
Low-Budget ( $\leq 16\text{k}$ )	IAMFM-MFBO vs. UCB Bandit	+3.48	0.0003	[+1.63, +5.33]
	IAMFM-MFBO vs. Uniform	+2.00	0.043	[+0.06, +3.93]
	IAMFM-ASH vs. UCB Bandit	+4.47	$<0.0001$	[+2.40, +6.55]
	IAMFM-ASH vs. Uniform	+2.99	0.0069	[+0.84, +5.14]
	IAMFM-MFBO vs. IAMFM-ASH	+0.99	0.38 ns	[-3.22, +1.24]
High-Budget ( $\geq 32\text{k}$ )	IAMFM-MFBO vs. UCB Bandit	+3.69	0.0002	[+1.83, +5.54]

Appendix

Table 4: Statistical significance (Welch’s t-test) for key comparisons across budget regimes. We report p-values and 95% CIs for the mean difference. Bolded methods are our novel proposals.

Panel of Figure 3 — Figure 3: Qualitative comparison. The naive greedy strategy (Config A) results in disjointed text where one advertiser overpowers the other, leading to hallucinations. IAMFM (Config B) identifies the optimal influence configuration, improving total welfare by $+36\%$ through enhanced semantic coherence.

User Weight ( $w_{U}$ )	Optimal $\mathbf{s}^{*}=(s_{A},s_{B})$	$E[V_{A}]$	$E[V_{B}]$	$E[V_{\text{user}}]$
$w_{U}=0.2$	(4, 4)	52.41	73.56	60.02
$w_{U}=1$	(2, 2)	51.59	72.46	63.77
$w_{U}=5.0$	(2, 4)	44.18	77.17	65.62

Appendix

Table 5: Optimal configuration

\mathbf{s}^{*}

for different user utility weights (

w_{U}

), calculated using F4 mean utilities aggregated across all 5 users. Restaurant A is “QuickBite” and B is “Thai Spice Garden”.

Scenario	Optimal $\mathbf{s}^{*}=(s_{A},s_{B})$	$E[V_{A}]$	$E[V_{B}]$
A-Only ( $\arg\max E[V_{A}]$ )	(3, 0)	52.44	57.07
B-Only ( $\arg\max E[V_{B}]$ )	(0, 4)	45.18	77.72

Appendix

Table 6: Optimal configuration

\mathbf{s}^{*}

in advertiser-only extreme scenarios, using F4 aggregated means. The optimum is found by maximizing only

E[V_{A}]

E[V_{B}]

Panel of Figure 4 — Figure 4: Absolute error (compared with the offline ground truth) in finding the best counterfactual welfare as a function of additional budget ( $B_{cf}$ ). We compare our warm-start methods against their cold-start counterparts. The GP-based methods are shown in blue: ACO-MFBO (solid blue line) and Cold-Start MFBO (dashed blue line). The elimination-based methods are in green: ACO-ASH (solid green line) and Cold-Start ASH (dashed green line).

Figure 5: SAE Steering for promoting ice creams and lemons. Prompt: “Help me plan the perfect movie night tonight with 8 guests coming. What do I need to buy?” Columns steer for ice cream, and rows steer for lemon-related things.

Online Advertising

Online advertising studies how digital ads are targeted, ranked, priced, measured, and delivered while balancing budgets, business goals, and user interests.

See all 87 papers

LLM-Auction: Generative Auction towards LLM-Native Advertising
Dec 11, 2025Chujie Zhao, Qun Hu, Shiping Song +4Online Advertising Mechanism Design
Taobao & Tmall Group of Alibaba Beijing, China
Token-Level Advertising
Aug 27, 2026Hanbing Liu, Bowei Zhang, Changyuan Yu +2Online Advertising Mechanism Design
Renmin University of China Beijing, China · Beijing, China · Stanford University Stanford, CA, USA
Generative Optimization for Incentivized Advertising with Global Level Constraints
Aug 5, 2026Gege Chen, Ning Luo, Hao Jiang +7Constrained Generative Modeling Online Advertising
University of Electronic Science and Technology of China · Kuaishou Technology

Dec 11, 2025J/K move · Enter open · S save

LLM-Auction: Generative Auction towards LLM-Native Advertising

Chujie Zhao, Qun Hu, Shiping Song, Dagui Chen+3

Taobao & Tmall Group of Alibaba Beijing, China

Online Advertising Mechanism Design Large Language Model-Guided Optimization

Open paper AskPDF arXiv

The commercialization of LLM applications is the next frontier in online advertising, with LLM-native advertising emerging as a promising paradigm by integrating ads into LLM-generated content. However, classic mechanisms are no longer applicable in this setting where the auction object is shifted from discrete ad slots to distributions over LLM outputs, and existing methods are impractical in industrial scenarios due to ignored externalities or high inference costs. To address these issues, we propose LLM-Auction, the first learning-based generative auction mechanism that integrates auction and generation. By formulating the allocation as preference alignment between LLM outputs and a mechanism objective that balances advertisers' value and user experience, we optimize the LLMs to inherently model allocation externalities without extra inference cost. Theoretically, we identify the allocation monotonicity and continuity of the optimal allocation rule of LLM-Auction, and prove that a simple first-price payment rule exhibits favorable incentive properties under this allocation rule. Furthermore, we build an LLM-as-a-judge simulation environment for quantitative evaluation, and experiments demonstrate that LLM-Auction achieves the state-of-the-art allocation efficiency while approximately satisfying key mechanism properties.

Abstract

Online Advertising

Explore similar work

LLM-Auction: Generative Auction towards LLM-Native Advertising

Token-Level Advertising

Generative Optimization for Incentivized Advertising with Global Level Constraints