Synthetic Audiences
Are Here.
Trust Is
the New Moat

The Field in Numbers

The synthetic audience research space has moved from fringe experiment to contested commercial territory inside twelve months. A handful of metrics set the context for everything that follows.

These numbers confirm a structural shift. Synthetic respondents are no longer a cost-cutting workaround. They are being positioned as a primary research modality by both startups and global holding companies alike. The question for every player in this space is no longer whether synthetic simulation works, but whose simulation is most trustworthy and most culturally grounded.

For Predikta, the timing is important. The company released the public research backbone of its platform, Sentiment Simulation using Generative AI Agents (Tia et al., arXiv:2505.22125), in May 2025, creating an auditable evidence base before commercial saturation made trust claims harder to differentiate.

Academic Papers Shaping the Science

The academic literature on generative agent simulation has accelerated sharply in 2025. Here are the most relevant developments and how they relate to Predikta's methodology.

Predikta's Own Foundational Paper

Tia et al. (2025), "Sentiment Simulation using Generative AI Agents" (arXiv:2505.22125, Netopia AI / UP Diliman / UP Los Baños). This arXiv preprint is the research foundation the Predikta platform is built on, and it should be evaluated as a research contribution, not only as a product brief.

The core claim is substantial: generative AI agents embedded with psychographically rich profiles, instantiated from a nationally representative survey of 2,485 Filipino respondents, covering personality traits, values, beliefs, and social attitudes, can replicate human sentiment with measurable fidelity. Contextualized profile encoding achieved 92% alignment with original survey responses. In forward-looking simulation tasks, agents reached 81–86% accuracy against ground-truth sentiment. Critically, simulation results were stable across repeated trials (±0.2–0.5% SD) and resilient to scenario framing variation (p = 0.9676, Cohen's d = 0.02), meaning the agents stay anchored to their psychographic grounding rather than being swayed by how a question is worded.

The framework runs in three stages: (1) agent embodiment via categorical or contextualized encodings, (2) exposure to real-world economic and social scenarios, and (3) generation of sentiment ratings with explanatory rationales. The underlying model is a large open-weight LLM, selected for its multi-turn coherence and long-context reasoning. The paper is classified under Multiagent Systems (cs.MA), Artificial Intelligence (cs.AI), and Computers and Society (cs.CY).

What this means competitively: Many commercial products reviewed in this brief make accuracy claims without public, auditable methods. Predikta has disclosed its methodology and validation metrics in a public arXiv preprint. That is a meaningful trust advantage while it lasts.

Closely Adjacent Research

Argyle et al., "Silicon Sampling". An earlier landmark showing that LLMs can emulate response distributions from human demographic subgroups, essentially validating the theoretical basis for what Predikta does. Predikta extends this by grounding agents not in the model's implicit "demographic knowledge" but in actual survey data from real respondents.

LLM Agents Grounded in Self-Reports (arXiv:2411.10109). A study using 1,052 Americans showing that agents built from structured survey data plus semi-structured interviews reached 83–86% accuracy on held-out social attitude items, closely mirroring Predikta's reported numbers. The paper validates the methodology of self-report grounding while also showing its ceiling: demographics-only agents scored only 74%. This is an important benchmark that Predikta's approach consistently clears.

AgentSociety (arXiv:2502.08691). A large-scale simulation framework from Chinese researchers embedding Maslow's hierarchy and emotional need states into generative agents. This work pushes the psychological depth question further, modeling needs, emotions, and cognition simultaneously. It is primarily a social science research tool rather than a commercial product, but it signals where the frontier is moving: toward richer motivational architectures, not just psychographic labels.

RecSysLLMsP, Polarization Simulation (Springer, 2025). A 100-agent simulation grounded in psychometric and demographic data from Serbian social media users, studying how algorithmic content feeds shape attitudinal polarization. Relevant to Predikta because it demonstrates that psychographically grounded agents can maintain ideological coherence across extended, multi-turn interactions, the same stability property Predikta reports.

Open challenge in the literature: Several papers, including a 2025 meta-review of LLM agent simulation, flag that LLMs "struggle significantly in scenarios that more accurately reflect real-world conditions", particularly around emotional nuance, cultural context, and group dynamics. Predikta's localized Philippine dataset is a direct response to the cultural-context critique. The emotional nuance gap remains an open research question for the entire field.

The 200-Campaign Backtest

The public arXiv preprint establishes the research foundation. The more commercial question is whether Predikta's scores can help marketing teams identify stronger ads before live media spend.

In a commercial backtest spanning 200 historical advertising campaigns and nearly 20,000 creative comparisons, Predikta's behavioral scores (sentiment, relevance, behavior, emotion strength, and an overall rating) were compared against actual campaign outcomes including click-through rate (CTR), cost per click (CPC), and cost per mille (CPM). The direction was consistent: higher Predikta scores were associated with stronger CTR and lower CPC, with alignment most pronounced on conversion-oriented campaigns.

Commercial backtest on historical campaign data. A directional predictive signal, not independent validation, peer-reviewed proof, or a guarantee of live campaign performance. Partner attribution and detailed results withheld pending clearance for external use.

Who Is Doing What

The commercial landscape splits into three tiers: well-funded generalist simulators, specialized synthetic research platforms, and incumbent research giants adding synthetic capability. We cover the most relevant players.

Source quality varies across this field. Where claims come from company materials, vendor roundups, or media reports rather than audited academic studies, they should be treated as directional signals rather than independently validated benchmarks.

The Incumbent Move: WPP Media's Open Intelligence

Perhaps the most significant structural development of the past year is not a startup but a global holding company. In June 2025, WPP Media launched Open Intelligence, what it claims is the industry's first Large Marketing Model (LMM). Trained on trillions of signals from 350+ partners across 75 markets, the platform is designed to predict audience behavior and marketing performance. WPP positions it as the marketing equivalent of an LLM: instead of generating language, it generates predictive intelligence about how audiences engage with content, brands, and products.

Open Intelligence is bespoke-model capable. Clients can fine-tune the foundation model with their own first-party data. It aims to reach up to 5 billion adults worldwide. Partners include Google, Microsoft Advertising, Snap, TikTok, and Experian.

What this means: Enterprise holding-company clients now have a proprietary simulation layer embedded in their media buying workflow. This is a different threat vector than Predikta's target market. WPP's tool is an optimization layer for existing campaigns, not a pre-launch simulator. But as WPP extends Open Intelligence to creative testing and concept validation, the overlap will grow.

Feature-by-Feature Comparison

Where Predikta Stands, an Honest Assessment

Genuine Differentiators

• Public research foundation. Among the competitors reviewed here, few have released a public arXiv preprint with auditable methodology and reported accuracy metrics. The paper (Tia et al., 2505.22125) is Predikta's strongest near-term credibility asset. It sets a publicly auditable standard that marketing claims from Aaru, Atypica, and Evidenza do not yet match.
• Cultural specificity as a feature, not a limitation. Many global platforms in this space are implicitly Western-default. Predikta's Philippine model is calibrated from 2,485 nationally representative respondents and projected onto the relevant Filipino adult or consumer universe, depending on the client use case. For brands operating in the Philippine market, this creates a level of local psychographic grounding that generic global panels are unlikely to match.
• Psychographic depth. The combination of Big Five personality traits, personal values, social beliefs, and social attitudes, grounded in actual survey responses from 2,485 real respondents, gives Predikta's agents a richer psychological profile than competitors relying on behavioral signals alone.
• Framing robustness. The paper's finding that agents maintained stable sentiment output regardless of how scenarios were framed (p = 0.9676, Cohen's d = 0.02) is a specific, testable quality claim. It means Predikta's simulation doesn't just tell brands what they want to hear based on how they phrase the question.

Honest Gaps to Watch

• Scale of funding vs. the field. Aaru raised a reported $50M+ Series A at a headline $1B valuation, with Accenture, EY, and IPG among its enterprise relationships and partners. Predikta is operating at a fraction of this resource level. The gap is in capital and distribution, not proven revenue, Aaru's reported ARR remains under $10M, but enterprise sales capacity is asymmetric.
• Geographic coverage. The Philippine-first model is the differentiator for the right clients and a limitation for clients seeking multi-country coverage. The path to Southeast Asia expansion (Indonesia, Vietnam, Thailand) is the natural next chapter, and it requires the same survey-grounding rigor applied to the Philippine dataset.
• Emotional nuance. The broader academic literature flags this as an unsolved problem for all LLM-grounded simulation. Predikta's agents produce sentiment ratings and rationales; they do not yet model affective state dynamics over time or emotional contagion across simulated audiences. Neither do competitors, but it is the next frontier.

What to Monitor Next Month

• Aaru's Lumen expansion. Accenture Song has signaled intent to integrate Lumen into its products across marketing and customer strategy. Watch for case studies or white papers from Accenture that establish Aaru's accuracy claims in a more auditable form. This will directly pressure Predikta's public-evidence advantage.
• Qualtrics Edge Audiences. Announced at Qualtrics X4 in March 2025, Edge Audiences embeds synthetic respondents into the world's most widely deployed survey platform, fine-tuned on 25+ years of Qualtrics' real-human research data. Currently limited to US General Population, with global expansion planned for 2026. If Qualtrics adds Southeast Asia coverage, the enterprise channel threat becomes serious. (qualtrics.com)
• AgentSociety framework adoption. The large-scale agent simulation framework (arXiv:2502.08691) integrating Maslow-level needs modeling is academic today but will be productized. Watch for commercial applications that add motivational architecture depth, the next evolution beyond static psychographic profiles.
• WPP Open Intelligence creative testing expansion. Open Intelligence launched for media optimization; its roadmap likely includes creative concept validation. When it does, Evidenza and Predikta's Campaign Simulation Lab will face direct competition from a platform with 350+ data partners and $B-scale reach.
• Southeast Asia market entry by any Tier 1 player. No major competitor currently appears to have publicly documented, survey-grounded psychographic simulation for the Philippine or Indonesian market. This window will not stay open indefinitely. Monitor Aaru's geographic expansion roadmap and any new Southeast Asia-focused AI research partnerships.