‘Yes,I Am a Slave to AI’: Critical Perspectives of Generative AI in Co-creation and Problem-solving

Over-reliance and Deskilling

As GenAI capabilities evolve, increasing reliance generates psychological, ethical and material risks (Bartikowski, 2026; Grewal et al., 2025). In advertising co-creation, over-reliance emerges when fluent GenAI output is treated as decision-ready, crowding out critique and reflective judgement. Filipa Fernandes (practitioner, Expansion Labs International) describes this tipping point as the moment when GenAI ‘stops provoking questions and only speeds up answers’, aligning with a logic of dependence, in which convenience reduces users’ motivation to engage in evaluative effort, thereby gradually weakening critical thinking (Bartikowski, 2026).

Academic perspectives converge around a valuation drift and its routinisation. Abhishek Mishra (academic, IIM Indore) argues that over-reliance begins when practitioners treat AI-generated content as superior, suggesting that GenAI should be used to ‘finesse the content, not generate it’, in line with Brüns and Meißner (2024). This drift is stabilised by organisational routines, including prompt lock-in, acceptance without critique and speed-over-reflection norms that narrow exploration and displace judgement, as noted by Faizan Ali (academic at the University of Galway). Mohammad Soliman (academic, Sultan Qaboos University) adds an organisational trigger: ‘when humans default to AI for ideation and iteration’, thereby optimising for speed over insight and reducing strategic debate. Together, these views align with concerns that GenAI can inflate perceived quality through surface plausibility, thereby increasing the uncritical uptake of generic or misleading outputs when verification and domain expertise are insufficient (Little et al., 2026). These dynamic risks turn value co-creation into value co-destruction, resulting in homogenised ideas and weaker differentiation (Yang et al., 2025). It may also paradoxically reduce empowerment, which diminishes when automation becomes excessive (Feng & Sun, 2025).

A practical safeguard is therefore governance, not willpower. Implementing ‘human-first checkpoints’ (Filipa Fernandes) preserves performance gains while maintaining human-centric evaluation (Lo Duca, 2025; Yang et al., 2025). This requires that humans frame the brief, develop at least one human-only concept before AI prompting and provide a written critique explaining why the selected idea works and how it was improved beyond the GenAI draft (Faizan Ali, Mohammad Soliman).

Disclosure as a Creative Variable and AI Aversion

Expert opinions on GenAI disclosure expose a strategic fault line between normalisation and differentiation. This tension mirrors prior research showing polarisation among advertising executives regarding disclosure and extends it by framing disclosure not only as a compliance issue but also as a strategic and creative decision (Demsar et al., 2025). One perspective, articulated by António Reis Pereira (practitioner, Automóvel Clube de Portugal), posits that AI’s ubiquity will render disclosure arguable as its use becomes the ‘new normal’. This normalisation paradigm, however, overlooks robust evidence of AI aversion, a phenomenon in which consumers instinctively distrust or reject algorithmic outputs, particularly in creative domains where they perceive them as uniquely human (Feng, 2025). Research confirms that merely disclosing GenAI’s role in content creation may reduce perceived brand authenticity and trigger negative reactions, directly challenging the assumption that audiences will passively accept full automation (Ali et al., 2025; Brüns & Meißner, 2024).

The alternative strategy, advocated by Lino Trinchini (academic, Nottingham Trent University), is to ‘embrace transparency and authenticity’, with a focus on human creativity in the use of GenAI. This approach reframes disclosure from a compliance burden into a creative variable. However, recent studies suggest that transparency can attenuate the positive effects of anthropomorphism, reducing the perceived social presence that fosters trust (Wang et al., 2025). Therefore, a successful disclosure must do more than simply state that AI was used. It must strategically preserve human-like qualities.

A practical approach is the selective disclosure, in which firms use GenAI for tangible ad elements (e.g., backgrounds) but rely on human, trust-critical elements (e.g., service providers) and clearly communicate this division of labour (Grigsby et al., 2025). Implementing such a strategy may be complex in practice, but it can create opportunities for differentiation, positioning and reduced reliance on GenAI as a premium service (Demsar et al., 2025). This human-centric framing, which positions AI as an assistive tool, mitigates aversion and enhances authenticity, transforming a potential risk into a source of competitive advantage by demonstrating that technology serves, rather than supplants, human creativity.

Co-creation Identity and ‘Who Owns the Voice’

GenAI co-creation converges on a shared position among experts, grounded in the principle that human accountability is non-negotiable. At the organisational (Anna Boechat, academic, Universidade Católica Portuguesa) and individual (Patrícia Oliveira, practitioner, Observador; Raquel Carrilho, practitioner, Observador) level, AI is framed as an auxiliary tool, with humans as the ultimate gatekeepers. This view is supported by research showing that the onus for legal and ethical compliance rests with the agency or brand (Demsar et al., 2025), by philosophical accounts defining the human–AI creative relationship as AI-assisted production (Khan et al., 2026) and by consumer psychology. Because AI is perceived as lacking a mind, consumers attribute greater ownership of AI-generated content to the human user, making AI plagiarism seem less immoral than human plagiarism (Wei et al., 2025).

These expert views align with the literature that treats AI as assistive rather than autonomous, but they also expose an already noted tension in prior work. As GenAI becomes more deeply embedded in co-creation, attribution and accountability become harder to separate cleanly (Khan et al., 2026; Little et al., 2026). The core issue is authenticity. Because AI lacks internal emotional states, its attempts to generate emotionally charged brand communications are perceived as inauthentic, thereby triggering moral disgust among consumers (Kirk & Givi, 2025). Furthermore, as AI’s role shifts from assistant to collaborator, it actively shapes outputs in ways that make clean attribution difficult (Little et al., 2026). This attribution becomes more complex when AI is perceived as emotional, which increases its perceived ownership and blurs responsibility (Wei et al., 2025), creating tension between the legal willingness for a single author and the reality of distributed agency.

The challenge is to develop governance frameworks that uphold human accountability while acknowledging AI’s active role. This requires a shift from a narrative of sole authorship to one of accountable stewardship. A practical approach is the CARE framework (collaboration, accountability, responsiveness, empowerment), which provides a structured, human-centric model for mitigating risks and implementing robust accountability systems (Feng et al., 2024).

Ghostwriting Remorse and the Moral Cost of Dependence

The use of GenAI to ghostwrite emotionally meaningful communications imposes a dual moral cost. For senders, it can trigger guilt over perceived dishonesty, as it feels like ‘misrepresenting effort or voice’ (Anish Yousaf, academic, Nottingham Trent University). This is empirically supported by research showing that using GenAI for heartfelt messages elicits guilt (Hass et al., 2026). The sense of responsibility is heightened because consumers attribute more ownership and credit to themselves for AI-assisted work (Khan et al., 2026; Wei et al., 2025). For recipients, AI-authored emotional communications are perceived as inauthentic because AI lacks internal emotional states, which can trigger moral disgust (Kirk & Givi, 2025).

Both experts propose a solution centred on reframing AI’s role from a substitute to an assistant. Teresa Lopes (practitioner, Creative Gate) suggests using AI as a brainstorming engine but not as a replacement for creative thinking, whereas Anish Yousaf argues for framing it as an assistive tool for articulation rather than for idea generation. These expert accounts reinforce existing evidence that the moral cost does not arise from GenAI use per se, but from the perception that authorship, effort or emotional authenticity has been inappropriately displaced (Hass et al., 2026; Khan et al., 2026; Kirk & Givi, 2025). While unacknowledged ghostwriting elicits guilt, acknowledged assistance does not (Hass et al., 2026). Moreover, the negative effects of AI authorship are mitigated when AI edits only the communication (Kirk & Givi, 2025). The moral cost, therefore, stems from deception, not from the technology itself.

Reducing this moral cost requires transparency and a commitment to co-creation. By clarifying that responsibility remains human and framing AI as a supportive tool, the perception of dishonesty is reduced. This involves designing GenAI tools that emphasise co-creation and encourage a culture of human oversight. The moral cost of dependence is not inherent in the technology, but in the failure to maintain transparency and accountability in its use.

Empowerment Paradox in Consumer Co-creation

The integration of GenAI into consumer co-creation presents an empowerment paradox. In fact, it can either enhance or diminish consumer autonomy. Luís Martins (practitioner, Delta Cafés) argues that GenAI empowers when it ‘reinforces the consumer’s choice and autonomy’ but reduces empowerment if it creates ‘dependence or uncritical automatism’. This expert view directly reinforces research showing that GenAI can enhance empowerment under human-led collaboration but can reduce it when automation becomes excessive and displaces user control (Feng & Sun, 2025). While human-led collaboration and communication with GenAI enhance empowerment, excessive reliance on AI for creation can lead to a loss of control and disempowerment. This paradox reflects a broader tension between AI as a tool for human augmentation versus human replacement (Hermann & Puntoni, 2025a).

Empowerment is maximised when GenAI functions as a ‘co-pilot, never as an autopilot’ (Luís Martins), a notion that grounds on perceived controllability. Research shows that perceived controllability is a significant moderator of trust in GenAI. Even beneficial features, such as personalised recommendations, can fail to build trust if users perceive a lack of control (Singu et al., 2026). This requires maintaining ‘agentic empowerment’, a term by Hermann and Puntoni (2025a) that refers to a sense of autonomy, control and meaningfulness. When consumers lead the co-creation process, using GenAI as an assistive tool, their sense of competence and autonomy is enhanced (Feng & Sun, 2025; Lin & Wu, 2026). The suggestion by an anonymous academic at Nottingham Trent University that GenAI should be used as a ‘tutor’ aligns with this augmentation-based approach.

To keep empowerment ‘real’, brands must design co-creation experiences that prioritise human agency and controllability (Singu et al., 2026). This means ensuring the consumer always ‘decides the final result’ (Luís Martins) and framing GenAI as a supportive partner. By focusing on agentic and creative empowerment, that is, leveraging GenAI to expand consumers’ capabilities without sacrificing their control, brands can mitigate the risks of dependence and ensure that co-creation remains a genuinely empowering experience (Hermann & Puntoni, 2025a, 2025b).

Prompt Craft as Hidden Dependence and Power

Experts have conceptualised prompt craft as a focal point of creative and strategic agency rather than merely a technical optimisation process. Maria Camargo (scholar, Universidade Federal de Santa Maria) posited that reliance arises when prompt engineering replaces strategic, creative and critical analyses, resulting in message uniformity, repetitive narrative frameworks and a gradual diminishment of unique brand identity. Ivo Valente (practitioner, Six-Factor) similarly noted that teams may become excessively focused on formulating the ‘ideal phrase’, thereby detracting from audience comprehension, communication goals and contextual strategies. In this regard, prompt craft constrains the decision-making framework and compresses human evaluative capacities.

This interpretation aligns with Bozkurt’s (2024) characterisation of prompt engineering as an art– science interface that mediates human–AI interaction and requires deliberate capability-building and ethical awareness. However, expert accounts extend this view by foregrounding organisational power dynamics. Prompt craft can operate as an informal gatekeeping resource within teams, concentrating influence in those able to elicit outputs that sound right, even when strategic accountability and brand risk remain elsewhere. Dependence thus becomes a governance issue: who controls prompts, which outputs are approved and what criteria define good creative work.

Recent scholarly investigations substantiate this apprehension. Yang et al. (2025) report that advertising practitioners perceive GenAI as enhancing efficiency while also posing risks of deskilling and reduced strategic depth in human–AI co-creation. Similarly, Little et al. (2026) argue that as GenAI shifts from assistant to collaborator, organisations must consciously design capability structures to preserve human oversight and expertise. Lo (2023) further conceptualises prompt engineering as a literacy practice, suggesting that diffusing competence across teams mitigates gatekeeping and cognitive over-reliance. Collectively, these perspectives position prompt craft not as a neutral technique but as a site of redistributed authority and potential cognitive dependence in marketing communication.

Indistinguishability and Creative Legitimacy

In both academic and practitioner circles, creative legitimacy has been characterised as shifting from the identity of the initial author to the methods used to establish and maintain credibility. Luisa Martinez (academic, Iscte-IUL) asserted that legitimacy should be grounded on a truthful representation of authorship and on how transparency is practised: regular, systematic disclosure can safeguard trust without necessarily diminishing uniqueness. Rita Roque (practitioner, Six-Factor) highlighted that brands may need to set themselves apart through signals that are difficult to replicate authentically, such as lived experiences, personal specificity, humour and culturally relevant nuances, as these indicators help audiences perceive authenticity, even when the text appears fluent.

These perspectives align with Zhang and Prebensen’s (2024) experiments that show that tourists struggle to distinguish ChatGPT-generated tourism marketing materials and that GenAI outputs can pass a Turing-style test while matching marketer-written text in semantic fluency and perceived attractiveness. When indistinguishability becomes routine, legitimacy cannot depend on consumers reliably ‘spotting AI’. It shifts to process: when and how GenAI involvement is disclosed, how provenance is documented and how human critical judgement is made visible (e.g., editorial sign-off, source traces and brand-voice governance).

Recent work further complicates this legitimacy calculus. Koning and Voorveld (2025) find that AI disclosures can increase AI-related and persuasion knowledge, yet also decrease trust in both the advertisement and the organisation, implying that transparency is not a frictionless solution but a design choice with trade-offs. Jung et al. (2025) similarly show that trust in AI and perceived humanness form two pathways that shape evaluations of AI-generated luxury advertising. Taken together, our experts’ responses suggest that, as fluency is commoditised, differentiation and legitimacy shift towards verifiable provenance, experiential originality and consistent brand behaviour over time, making legitimacy less a property of the artefact and more a property of the brand’s accountability system.

Human–AI Entanglement as Resource Relations

Respondents viewed GenAI as more than just a mere tool, yet not as an independent creator. Luís Schwab (academic, IPAM) emphasised the risks of dependence throughout the creative process: gathering information, ideation and idea sharing and, most importantly, the final output when teams cease to exercise critical judgement. Practitioners Pedro Rebordão (LISPOLIS) and Guido Santos (Genesis Digital Solutions) characterised GenAI as a ‘super tool’ and a catalyst for value creation, while asserting that the ‘true value’ still originates from humans and that prompts cannot supplant expertise. They also noted a unique aspect compared with traditional tools: GenAI’s output is less predictable, which encourages more iteration while simultaneously heightening the need for oversight (e.g., hallucination checks).

This framing mirrors Kaartemo and Helkkula’s (2025) argument that value co-creation involves entangled human–AI resource relations rather than a simple tool-use model, with multiple types of relations shaping how agency and responsibility are distributed. The expert opinions operationalise this lens for marketing communication: if GenAI is an entangled resource, dependence should be assessed by where the relation becomes non-transparent (e.g., unreviewed outputs, invisible assumptions embedded in prompts or over-trusting ‘AI-sounding’ fluency). Designing entanglement, then, means specifying the relation the team wants GenAI to occupy (background assistant, interpretive partner or near-immersive co-author) and the corresponding accountability.

The comments imply that the core competence is the orchestration of the integration of resources, human judgement, brand knowledge and GenAI capabilities towards value-in-use (He & Yang, 2025). In practice, this suggests instituting checkpoints for verification, hallucination detection and brand-voice consistency, while treating GenAI as a collaborative partner whose contributions must remain inspectable.

Data-driven Storytelling Dependence

Two risks dominated expert comments: factual distortion and manipulative persuasion. Jorge Cunha (practitioner, IT Tech BuZ) argued that GenAI threatens narrative integrity by inventing or distorting facts and ‘manufactures emotion’ through manipulative tactics. His constraint, ‘no source, no claim’, recasts GenAI as a co-pilot that drafts, summarises and proposes hypotheses, but never validates truth; accountable human review becomes mandatory. Tabish Mir (academic, University of Nottingham) similarly warned that GenAI can ‘oversell’ subtly, influencing audiences who are less aware of how the technology works; when the ‘facade’ is perceived, the effect can backfire against the brand.

These concerns align with and tighten Lo Duca’s (2025) framework for co-designing data-driven stories. The framework uses the journalistic 5Ws (who, what, when, where, why) to guide GenAI towards consistent, contextual outputs and explicitly anchors story development in ethical principles as it progresses from data to information, knowledge and wisdom. In our data, the experts’ evidence gate functions as an operational rule in the early AI data and AI information stages: GenAI may help extract patterns, generate candidate storylines and draft narrative options, but any claim included in a public-facing story must remain traceable to underlying data and sources, with the human storyteller retaining responsibility for verification.

Recent evidence underscores the importance of this boundary. GPT-4 can be highly persuasive in conversational settings, and this persuasion is strengthened when arguments are personalised, raising the stakes for ‘ethical persuasion’ in marketing narratives (Salvi et al., 2025). Complementarily, Park and Nan (2025) find that LLMs can generate highly convincing misinformation and that exposure can reduce trust and influence decision-making, illustrating how fluent fabrication can circulate as a data-backed explanation.

In marketing communication, the risk of dependence peaks when GenAI becomes a shortcut for justification rather than a disciplined aid for explanation, evidence tracking and accountable meaning-making.

Synthetic Media Spillovers (Deepfakes) and Trust Decay

Practitioners are now concentrating on maintaining credibility in an era where ‘seeing is believing’ no longer applies. João Matos (practitioner, EY) contended that brands ought to replace superficial transparency with verifiable evidence and consistent behaviour: communicate calmly and clearly during crises, use technical indicators (such as metadata) and establish trust rituals that make authenticity measurable. Ana Telles (practitioner, Studio TUCA) suggested a cultural approach: highlight the qualities that set humans apart from machines, craftsmanship, imperfection and visible ‘errors’, to provide audiences with richer cues of authenticity than mere polished realism.

These proposals align with Momeni’s (2025) findings that audiences struggle to identify deepfakes and that persuasive synthetic media can shape opinions, raise ethical concerns and blur the boundaries of reality. In a marketing context, this creates spillover distrust: even authentic brand content may be treated as suspect, especially when it circulates beyond owned channels. The ‘proof over proclamation’ recommendation responds directly by shifting trust from aesthetic realism to verifiability and consistent conduct, that is, what the brand repeatedly does rather than merely what it declares.

The demand for credible evidence aligns with extensive research on deepfakes, which cautions that synthetic media can accelerate the process of ‘truth decay’ and facilitate reputational undermining and exploitation (Hynek et al., 2025). Collectively, the insights from experts indicate that restoring trust necessitates not only technical verification tools but also communication strategies that render accountability transparent, namely who is accountable, how the content was created and how assertions can be verified.