Generative AI is fundamentally reshaping the planning, execution, and optimisation of marketing, with significant effects on strategy, research, creative development, media, and customer experience (Grewal et al., 2025; Farseev et al., 2024; Harkness et al., 2023).

1. From predictive to generative marketing

Analytical AI has traditionally supported marketing predictions, such as identifying target audiences, offers, and pricing, using structured data such as transactional data and demographics. In contrast, generative AI leverages unstructured data (text, images, audio, video) to produce new content and ideas, including ad copy, product descriptions, images, chat responses, scripts, and synthetic customers (Harkness et al., 2023; Grewal et al., 2025).

These approaches are converging with predictive models that identify who to target and when, while generative models determine messaging and presentation, often in near-real-time (Grewal et al., 2025; Farseev et al., 2024).

A retailer might use analytical AI to identify high–value segments, then deploy generative models to create and test thousands of creative variants and messages for those segments, updating them daily (Grewal et al., 2025; Harkness et al., 2023).

2. How GenAI is changing marketing work

Generative AI is transforming daily marketing tasks across three key areas: speed, scope, and sophistication (Grewal et al., 2025; Farseev et al., 2024; Harkness et al., 2023).

Content volume and speed

• Organisations now use GenAI to draft ad copy, social posts, emails, landing pages, and sales scripts within minutes rather than days (Grewal et al., 2025; Farseev et al., 2024; Harkness et al., 2023).
• Vanguard increased LinkedIn conversions by 15% using GenAI-generated ad copy, and Emirates NBD achieved a 177% increase in credit card leads through AI-personalised offers (Grewal et al., 2025).

Service, sales, and operations

• Brands including Unilever and a major Asian bank use GenAI assistants to summarise customer interactions, suggest replies, and retrieve information, reducing handling time by 20–90% and allowing staff to focus on higher-value conversations (Grewal et al., 2025).
• Walmart and other companies use negotiation and vendor-facing chatbots, resulting in measurable cost savings and strong partner acceptance (Grewal et al., 2025).

Research, insight, and strategy

• GenAI can summarise qualitative interviews, synthesise product reviews, simulate synthetic respondents, and analyse competitor reports, transforming lengthy research cycles into continuous decision-making inputs (Korst et al., 2025; Grewal et al., 2025).
• Emerging systems such as MindFuse analyse ads and performance data to identify content pillars, personas, and themes, then automatically draft campaign narratives and briefs for agency refinement (Farseev et al. 2024).

This shift moves teams from creating content from scratch to focusing on editing, interpretation, and orchestration, which increases the need for strategic and creative judgment (Grewal et al., 2025; Farseev et al., 2024).

3. A practical framework: inputs and human control

Grewal et al. (2025) introduce a four-quadrant framework for selecting GenAI tools, based on input data and the degree of human involvement.

Input: General vs. Custom

General Models (Large Language Models (LLMs), e.g. Chat GPT, Gemini, Copilot)

• Trained on broad public data; excellent for generic tasks such as brainstorming, first‑draft copy, social posts, and generic educational content (Grewal et al., 2025; Harkness et al., 2023).
• These models are most effective when broad knowledge is required, and the risk of error is low, such as during early ideation or for internal summaries (Grewal et al., 2025).

Customised Models and Retrieval-Augmented Generation (RAG)

• These models are fine-tuned or retrieval-augmented using proprietary brand, product, and customer data, making them essential for precise, brand-safe, or regulated tasks such as product advice, regulated content, or store-level information (Farseev et al. 2024; Grewal et al., 2025).
• A custom GenAI assistant can answer questions about SKU location, stock status, and alternatives using current store data, leading to efficiency gains and increased sales in challenging categories (Grewal et al., 2025).

Output: low vs. high human augmentation

Low augmentation (nearly autonomous)

• Suitable for low‑risk tasks: automated review summaries, internal Q&A, basic store info, or ranking creative variants (Farseev et al. 2024; Grewal et al., 2025).
• This approach maximises speed and scale but also raises the risk of hallucinations, bias, or tone errors (Grewal et al., 2025).

High augmentation (human‑in‑the‑loop)

• Needed when brand risk, legal exposure, or customer impact is high: public social content, pricing or financial disclosures, sensitive service communications (Grewal et al., 2025).
• Many firms use GenAI as a co-pilot, generating drafts that humans review, adapt, and approve (Farseev et al. 2024; Grewal et al., 2025).

In practice, most solutions combine these approaches, such as using a general LLM with RAG on internal content, rule-based guardrails, and mandatory human review before external publication (Farseev et al. 2024; Grewal et al., 2025).

4. Beyond content: GenAI as explainable co‑strategist

Next-generation systems integrate GenAI throughout the entire marketing lifecycle, extending beyond copywriting (Farseev et al. 2024).

Strategy co‑creation and “narrative intelligence”

• MindFuse shows that LLMs can process thousands of competitor and brand ads, extract content pillars such as needs, value propositions, and tone, cluster audiences and challenges, and generate persona-challenge narratives, such as strategist-written briefs (Farseev et al., 2024).
• Agencies using these systems report completing tasks such as audience research, brand analysis, and content planning 2.5 to 12 times faster, while maintaining explainability through attention maps and cluster labels (Farseev et al. 2024).

Explainable creative optimisation

• By combining click-through rate (CTR) models with visual attention maps, marketers can identify which creative elements drive engagement and simulate performance impact before launch (Farseev et al. 2024).
• Experiments show that removing attention-critical elements such as key visuals, call-to-action (CTA) buttons, or backgrounds can reduce CTR by 22–64%, providing designers with clear guidance on essential elements (Farseev et al. 2024).

Live campaign steering

• Multimodal LLMs can analyse performance data such as reach, cost-per-result (CPR), CTR, cost-per-mille (CPM), and conversion rates, along with sample creatives, to generate human-like diagnoses and recommended actions, such as identifying creative fatigue or suggesting adjustments to bidding and audience targeting (Farseev et al. 2024).
• While not flawless, these systems already function as junior performance strategists and are especially valuable for lean teams (Farseev et al. 2024).

This trend suggests marketing environments where humans set goals and constraints, while GenAI continuously analyses, recommends, and drafts content, with humans curating and making final decisions (Grewal et al., 2025; Farseev et al., 2024; Harkness et al., 2023).

5. Risks, limits, and governance

The features that make GenAI powerful also introduce significant structural risks (Farseev et al. 2024; Grewal et al., 2025).

Hallucinations and inaccuracy

• GenAI is probabilistic, not deterministic; it can fabricate facts, misinterpret nuance, or over‑confidently state wrong answers (Grewal et al., 2025).
• This may be acceptable during early ideation but poses significant risks in regulated domains, financial communications, or sensitive service interactions (Grewal et al., 2025).

Bias, ethics, and IP

• Models inherit bias from their training data. Studies reveal ideological and demographic skews in major LLMs, which can affect targeting, messaging, and creative outputs (Grewal et al., 2025).
• There is active legal debate around copyright for training data and generated outputs, particularly for image and video models (Grewal et al., 2025; Farseev et al., 2024).

Privacy and data control

• Executives are concerned about proprietary data being incorporated into model weights or exposed through outputs, particularly when using public Application Programming Interfaces (APIs) (Grewal et al., 2025).
• Many organisations are moving toward private deployments, strict data contracts, and careful selection of data inputs for models (Grewal et al., 2025; Farseev et al., 2024).

Societal and regulatory pressure

• Concerns about deepfakes, misinformation, and election interference are already driving regulation, such as the European Union AI Act, which includes explicit provisions on generative transparency and watermarking (Grewal et al., 2025).
• In summary, GenAI should be viewed as a high-impact yet high-variance capability that requires clear governance rather than simple experimentation (Grewal et al., 2025; Farseev et al., 2024)

Recommendations: How to prepare your marketing strategy

For businesses and entrepreneurs, GenAI readiness requires designing the right architecture for data, people, processes, and safeguards, rather than focusing on a specific tool (Grewal et al., 2025; Farseev et al., 2024; Harkness et al., 2023).

1) Align use cases with risk and value

Align your marketing activities with the input and augmentation framework:

• Low‑risk, generic tasks (idea generation, internal summaries, generic social drafts): use general models with light oversight (Grewal et al., 2025).
• High‑risk or high‑stakes tasks (regulated claims, pricing, financial comms, large‑scale public campaigns): use customised/RAG models plus mandatory human review and legal checks (Grewal et al., 2025).
• Prioritise use cases where GenAI provides clear speed or scale benefits, such as content pipelines, testing and variation, qualitative synthesis, and frontline enablement (Grewal et al., 2025; Farseev et al., 2024; Harkness et al., 2023).

2) Invest in your proprietary data and knowledge layer

• Consider internal content and data, such as brand guidelines, historical campaigns, CRM data, FAQs, product specifications, and playbooks, as competitive assets for GenAI (Grewal et al., 2025; Farseev et al., 2024).
• Develop or partner to implement a retrieval layer, including vector search, content tagging, and access policies, to ensure models base outputs on verified internal knowledge rather than public sources (Grewal et al., 2025; Farseev et al., 2024).
• Begin with small-scale initiatives, such as knowledge-based copilots or campaign Q&A and expand to customer-facing experiences as quality and governance improve (Grewal et al., 2025; Farseev et al., 2024).

3) Design “human‑in‑the‑loop” workflows by default

• Clearly define which content requires human review, editing, or approval, including brand voice, sensitive topics, offers, targeting rules, and any material that could pose reputational or regulatory risks (Grewal et al., 2025).
• Transition creative and strategy teams to editor-curator roles, focusing on reviewing AI-generated options, selecting and refining content, and incorporating performance data into prompts and knowledge bases (Farseev et al. 2024).
• Establish and document standard prompt libraries, review checklists, and escalate procedures to maintain consistent quality as GenAI usage increases (Grewal et al., 2025; Farseev et al., 2024).

4) Build explainability and measurement into your stack

• Select systems that provide explainability, such as attention maps, content-pillar contributions, or source citations, rather than relying solely on black-box scoring (Farseev et al. 2024).
• Connect creative features and narratives to funnel metrics such as CPM, CTR, conversion, and Return-On-Ad-Spend (ROAS), enabling teams to identify which stories, tones, and visuals drive revenue rather than just engagement (Farseev et al. 2024; Harkness et al. 2023).
• Conduct controlled tests comparing human-only and human-plus-GenAI workflows in terms of speed, cost, and outcome quality, and use these results to refine your operating model (Grewal et al., 2025; Farseev et al., 2024).

5) Upskill and reorganise your marketing talent

• Develop hybrid skills, including prompt design, critical evaluation of AI outputs, data literacy, cross-channel experimentation, and traditional positioning and storytelling (Farseev et al. 2024).
• Introduce new roles as needed, such as AI product owners for marketing, creative technologists, and GenAI ethics and governance leads, particularly in larger organisations (Farseev et al. 2024).
• Agencies and in-house teams should focus on orchestration and brand stewardship in an AI-driven environment, rather than manual production (Holte & Bublies 2025; Farseev et al. 2024).

6) Establish clear guardrails and policies

• Define which data may be used for training or prompting and clarify ownership of AI-assisted content with employees, partners, and clients (Grewal et al., 2025; Farseev et al., 2024).
• Adopt guidelines for disclosure, bias mitigation, and acceptable use, including restrictions on sensitive targeting or the use of synthetic personas, in alignment with emerging regulations in your markets (Grewal et al., 2025).
• Regularly audit key GenAI systems for performance drift, hallucination patterns, and bias, particularly in customer-facing journeys and automated decision points (Grewal et al., 2025; Farseev et al., 2024).

By treating GenAI as an integral part of your marketing operating system, grounded in proprietary data, guided by human judgment, and continuously measured, you will be well-positioned to realise its benefits while safeguarding customers, brand, and society.

REFERENCES

Farseev, A., Ongpin, M., Yang, Q., Gossoudarev, I., Chu-Farseeva, Y. Y., & Nikolenko, S. (2025). MindFuse: Towards GenAI Explainability in Marketing Strategy Co-Creation. ArXiv. https://doi.org/10.1145/3746027.3758167

Grewal, D., Satornino, C.B., Davenport, T., Guha, A. (2025). How generative AI is shaping the future of marketing. J. of the Acad. Mark. Sci. 53, 702–722. https://doi.org/10.1007/s11747-024-01064-3

Harkness L, Robinson K, Stein E, Wu W (2023), How generative AI can boost consumer marketing, McKinsey & Company,  https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/how-generative-ai-can-boost-consumer-marketing

Holte S, Bublies T (2025), From Automation to Innovation, From Data To Impact, https://fromdatatoimpact.com/index.php/2025/07/03/from-automation-to-innovation/

Korst J, Puntoni S, Toubia O (2025). How Gen AI is Transforming Market Research, Scribd, https://www.scribd.com/document/884716689/How-Gen-AI-is-Transforming-Market-Research