AI Marketing Research: Agents, Influencers & Analytics
If you've built a multi-agent AI pipeline for content, deployed a virtual influencer, or shopped for an AI analytics tool, you've probably felt the same uneasy gap between the demo and the result. The tech works. The output is technically fine. But something is missing — direction, warmth, or a clear sense of who is accountable for the decisions the system is making.
This edition of AI marketing research surfaces three peer-reviewed papers that, taken together, name that gap. A CHI 2026 study shows what happens when AI agents try to coordinate creative work without a human conductor. A survey from Indonesia tests whether humanlike AI influencers actually move purchase intent. And a literature review maps how marketing analytics has evolved into a stage where ethics and explainability are no longer optional.
If you're a brand manager, agency lead, or founder trying to make smart bets on AI without overcommitting to hype, these three studies offer a usable map — plus honest notes on what each one does and doesn't prove.
Quick Takeaway
- Multi-agent AI workflows stall without active human direction at each stage, not just the brief and the review.
- AI influencers with humanlike appearance and empathetic language increase reported purchase intent.
- Marketing analytics has entered a stage where fairness and explainability are becoming business requirements.
- All three findings are early signals — small samples, single-country surveys, or conceptual reviews. Treat as directional.
- The clearest opportunity: build or buy AI tools that structure human judgment into the workflow, not around it.
What This Research Means for Marketers
The common thread across these papers is that AI in marketing performs best when human judgment is structurally embedded — not bolted on. Whether you're orchestrating agents, designing a virtual influencer, or evaluating an analytics vendor, the differentiator is the quality and placement of human decisions inside the system.
That reframes the conversation from "how much can we automate?" to "where do we need human taste, empathy, and oversight, and how do we make those checkpoints explicit?" Teams that answer that question well will outperform teams that simply buy more AI access.
Papers Covered
Paper 1: Understanding Human–Multi-Agent Team Formation for Creative Work
- Source / venue: CHI '26 (ACM CHI Conference on Human Factors in Computing Systems)
- Link: https://doi.org/10.1145/3772318.3791166
- Source type: Peer-reviewed conference paper
- Method: Exploratory qualitative user study using a custom technology probe called CrafTeam. Participants ran three cycles of team formation, ideation, and reflection, followed by post-study interviews.
- Sample: 12 design practitioners working in design teams at IT companies; approximately three-hour sessions each.
- Main finding: Participants initially tried to let AI agents coordinate autonomously, which produced unproductive loops and creative drift. After those failures, they shifted to acting as orchestrators — directly assigning tasks and setting direction at each stage — which improved creative output.
- Evidence strength: Peer-reviewed at a top HCI venue, but small qualitative sample (n=12), single country, single session per participant. Strong directional signal, not quantitative proof.
- Limitation: Findings come from 12 South Korean design practitioners in a single three-hour session with a research prototype. Generalization to marketing teams, other industries, or long-term use is unproven.
- Practical implication: Treat yourself as the conductor of your AI agent team, not a reviewer of its output. Add explicit human decision gates mid-process — not just at the brief and the final review — to keep creative direction from drifting.
Paper 2: Physical Humanlikeness as A Moderator of The Relationship Between AI Influencer Marketing and Purchase Intention
- Source / venue: International Journal of Management Science and Information Technology
- Link: https://doi.org/10.35870/ijmsit.v6i1.7072
- Source type: Peer-reviewed journal article
- Method: Quantitative survey study using Structural Equation Modeling with Partial Least Squares (SEM-PLS) in SmartPLS 4.0. Physical humanlikeness tested as a moderator between AI influencer marketing and purchase intention.
- Sample: 250 respondents in Indonesia with prior experience purchasing through or via AI.
- Main finding: AI influencers perceived as more humanlike — through realistic appearance, empathetic language, and personalized responses — strengthened the relationship between AI marketing activity and self-reported purchase intent.
- Evidence strength: Peer-reviewed but in a non-top-tier journal, abstract-level detail on measures, single-country self-selected sample, and self-reported intent rather than actual purchase behavior.
- Limitation: Self-reported purchase intention does not equal actual purchases. Respondents had pre-existing AI purchase experience, which may bias results. Cultural generalization is unclear.
- Practical implication: If you're deploying AI influencers or chatbots, invest in warmth and personalization cues — empathetic language, expressive design, situation-specific responses — not just product knowledge or technical polish.
Paper 3: Artificial intelligence across social sciences and humanities: The evolution of marketing analytics in the digital era
- Source / venue: Journal of Interdisciplinary Research in Artificial Intelligence and Society
- Link: https://doi.org/10.20897/jirais/18474
- Source type: Peer-reviewed integrative literature review
- Method: Integrative review synthesizing 21 DOI-traceable scholarly sources (2014–2025) into a four-stage framework of marketing analytics evolution.
- Sample: 21 peer-reviewed sources across marketing analytics, AI in marketing, computational social science, and digital humanities.
- Main finding: Marketing analytics has progressed through descriptive, predictive, and generative stages and is entering a fourth stage centered on fairness, trust, and explainability. The author argues that segmentation and personalization decisions are increasingly social judgments, not just technical ones.
- Evidence strength: Conceptual review by a single author in a new, low-citation-history journal. Framework is proposed, not empirically validated.
- Limitation: Small source base (21), single-author synthesis, untested conceptual model, new venue with limited track record, and truncated full text.
- Practical implication: When evaluating AI marketing tools, add explainability and bias review to your vendor checklist alongside performance metrics. Build a habit of asking who an AI segmentation system leaves out, not just how accurate it is.
Plain-English Payoff
AI doesn't run well on autopilot in marketing. Multi-agent creative work stalls without a human conductor. AI influencers convert better when they feel human. And the analytics tools you buy now need to answer ethical questions, not just performance ones. The human layer isn't overhead — it's the product.
Money Move
The clearest market gap across these papers is a structured human-in-the-loop multi-agent workflow for marketing teams — where the marketer explicitly directs specialized agents (copywriter, critic, brand checker) at defined checkpoints rather than watching them auto-coordinate and drift. Adjacent opportunities: a humanlikeness audit service for brands running AI influencers or chatbots, and an AI marketing audit offering that scores vendors on explainability and bias, not just ROI.
Evidence Check
- All three papers were reviewed at full text, but two are early-signal studies, not definitive evidence.
- The CHI 2026 paper is peer-reviewed at a top venue but qualitative with only 12 participants in one country.
- The AI influencer study uses self-reported purchase intent, not observed buying behavior, and a single-country sample of 250.
- The analytics review is a single-author conceptual synthesis of 21 sources in a new journal with limited track record.
- None of these papers establish causality at scale. Treat the findings as directional signals to test, not rules to deploy.
What to Test Next
- Action step. Map your current AI content workflow and mark every point where a human makes a directional judgment call. If those checkpoints exist only at the brief and the final review, add at least one mid-process decision gate and compare output quality across a few campaigns.
- Action step. Audit any AI chatbot, virtual assistant, or AI influencer you currently deploy for humanlikeness cues — empathetic language, personalized references, expressive design — and run a small A/B test on conversion impact.
- Action step. Add an explainability and fairness review to your AI vendor evaluation rubric. For each tool, document how it segments audiences, what it can explain, and who it might systematically exclude.
- Action step. Pilot a structured multi-agent workflow with explicit agent roles (copywriter, critic, brand checker) and human orchestration at each handoff, then compare it to a single-prompt baseline.
How This Connects to AI Marketing Strategy
Across Big Plans Media's AI marketing research coverage, the recurring pattern is that AI tools amplify whatever judgment system surrounds them. When the surrounding system is thin — a vague brief, an auto-coordinated agent stack, a chatbot with no persona, a vendor evaluated only on accuracy — output quality degrades and risk accumulates quietly.
These three papers each hit a different part of that pattern: workflow design, brand expression, and governance. A coherent AI marketing strategy treats them as one system. Where does human direction sit inside your creative pipeline? How human does your AI persona feel to the customer? And how do you justify the segmentation and personalization decisions your stack is making? Teams that can answer those three questions with evidence will be the ones generative AI marketing actually pays off for.
FAQ
What does the research say about multi-agent AI workflows for marketing?
An exploratory CHI 2026 study with 12 design practitioners found that leaving AI agents to coordinate autonomously produced unproductive loops and creative drift. When participants shifted to actively directing each agent at every stage, output improved. The finding is directional, not definitive, but it's consistent with how teams actually report experiencing these pipelines.
Do AI influencers actually drive purchases?
A 250-respondent survey study in Indonesia found that AI influencers perceived as more humanlike — realistic appearance, empathetic language, personalized responses — strengthened the link between AI marketing activity and self-reported purchase intent. The study measures intent, not actual purchases, and is limited to one country, so treat it as a design signal rather than proof.
What is human-in-the-loop AI in a marketing context?
Human-in-the-loop AI means structuring workflows so a human makes directional decisions at multiple points inside the AI process, not only at the start and end. In marketing, that looks like a person explicitly assigning tasks to specialized agents, reviewing creative direction mid-process, and making the brand and taste calls that AI agents struggle to make for each other.
How is generative AI changing marketing analytics?
A 2026 integrative review proposes a four-stage evolution: descriptive analytics, predictive analytics, generative content and conversation, and an emerging stage focused on fairness, trust, and explainability. The author argues marketers should now evaluate AI tools on ethical and governance criteria alongside performance, since segmentation decisions are increasingly social judgments.
Should small businesses use AI influencers or AI agents?
The research suggests value is achievable, but architecture matters more than access. For AI influencers, invest in warmth and persona, not just product knowledge. For agent workflows, keep the owner actively in the loop rather than chaining tools and walking away. Small teams often have an advantage here because the human direction is easier to keep tight.
Are these AI marketing findings proven or still early?
Early. The multi-agent study has 12 qualitative participants. The AI influencer study uses 250 self-reported survey responses in one country. The analytics review is a single-author conceptual synthesis. All are useful directional inputs for strategy, but none should be treated as settled science.
What are the biggest risks of automating marketing with AI?
Three stand out across this episode's research: creative drift when agents coordinate without human direction, weaker conversion when AI personas feel robotic, and governance risk when segmentation and personalization tools aren't explainable or audited for fairness. Each risk is addressable, but only if you design for it intentionally.
Listen to the Episode
Sources and Further Reading
- Understanding Human–Multi-Agent Team Formation for Creative Work
- Physical Humanlikeness as A Moderator of The Relationship Between AI Influencer Marketing and Purchase Intention
- Artificial intelligence across social sciences and humanities: The evolution of marketing analytics in the digital era
Related Big Plans Media:
About Big Plans Media
Big Plans Media helps marketers, educators, entrepreneurs, consultants, and business leaders translate AI marketing research into practical strategy. AI & Marketing Research Radar is produced by Big Plans Media and hosted by Evita, an AI-generated research briefing avatar trained on Dr. Eva Wolf's research framework.
