When the answer engine decides: Practitioner reflections on generative AI,consumer decision-making,and market transparency in Singapore's tropical platform market

Context and background

I write as both a practitioner and a researcher. I operate an e-commerce venture in consumer electronics in Singapore, teach marketing, and digital economy modules at undergraduate and postgraduate levels, and hold a doctorate researching AI-augmented consumer decision-making and memory. This dual positioning, as seller and buyer, educator and learner, shapes what follows.

Singapore provides a concentrated setting for this analysis: internet penetration stood at 96.0% in early 2024, with cellular mobile connections substantially exceeding the population (Kemp, 2024). Its retail landscape is characterised by pricing that shifts rapidly across official brand stores, telco bundles, and third-party marketplace sellers, creating conditions in which information currency is itself a competitive advantage. Teo and Yeong's (2003) study of the digital consumer decision process in Singapore's online marketplace offers a useful baseline; the current paper extends their analysis to AI-mediated advice as a further layer of information intermediation.

Singapore constitutes a distinctive analytical site because three conditions intersect here in ways that individually exist elsewhere but combine locally. The first is tropical environmental exposure: heat, humidity, and direct equatorial sun create device-specific performance requirements, such as sustained thermal management and peak-nit outdoor brightness, that are rarely foregrounded in the global review corpus generative AI draws upon. The second is regional platform temporality: flash-sale architectures compress the window between ‘best deal’ and ‘expired deal,’ making time-sensitive information currency a structural feature of the retail environment (Hendra and Adiwijaya, 2026; Lamis et al., 2022). The third is cross-border retail fragmentation: Singapore's parallel-import marketplaces offer local-set and export-set variants of the same device at different price points with materially different warranty and service outcomes. These three conditions are not uniquely tropical. Their significance lies in their simultaneous presence within a single, platform-dense, AI-mediated purchase environment. Together, they create a local verification problem that an answer engine ranking primarily on price-to-specification grounds cannot resolve: climate suitability may be omitted because it rarely appears in the global review corpus; promotion-linked prices can change rapidly and therefore require time-stamped verification; and apparently cheaper devices may carry warranty regimes that change the effective cost of ownership. Susanto et al. (2023), studying Indonesian m-commerce users, find that perceived risk, trust, efficiency, and functional benefit shape satisfaction; perceived risk, trust, and functional benefit also directly influence price sensitivity, while efficiency operates through satisfaction and continuance intention. This indicates that m-commerce evaluations extend beyond headline price, a dynamic that answer engines optimised for specification comparison may systematically underweight.

The challenge this creates is cognitive. Consumers want ‘best value now’, but value is multi-dimensional: performance, battery life, camera quality, warranty, and after-sales support. The cognitive load of evaluating these attributes across multiple channels (Paas et al., 2003) accumulates into decision fatigue (Baumeister et al., 1998). Generative AI is increasingly taking over the comparison tasks consumers find taxing (Gerlich, 2025), compressing multi-tab research into a single conversational query. Figure 1 visualises this shift.

OPEN IN VIEWER

Figure 1.

The consumer decision journey compressed. AI mediation delegates search and initial evaluation to the algorithm, shifting consumer agency from information gathering to recommendation interrogation.

Approach and practice description

The vignette. This account is a reflective practitioner narrative, a form of autoethnographic inquiry in which the researcher's own situated experience serves as primary analytical material (Ellis et al., 2010). The intent is illustrative rather than generalisable. On 4 March 2026, I used ChatGPT with web search enabled to find the best smartphone under SGD 500 in Singapore, then followed up with a question about POCO models specifically. The AI retrieved live Singapore marketplace prices and seller ratings, confirming that web retrieval was active. Table 1 traces the decision process through four stages.

OPEN IN VIEWER

Table 1.

Decision trace of the AI-assisted smartphone purchase (ChatGPT, 4 March 2026, web search enabled).

Stage	Prompt issued	AI emphasis	What was absent or conditional	Consumer response
Initial query	‘What is the best mobile phone under $500 in Singapore’	Price tier, chipset, battery, camera, software update longevity. Ten models with star ratings and marketplace prices.	Warranty regime absent. No local-set vs export-set disclosure for any model. No climate-relevant attributes (heat, outdoor brightness, humidity) in any result.	Accepted shortlist. Asked follow-up on POCO.
POCO probe	‘What about the Poco phone models’	Chipset, storage class (UFS 4.0), price-to-spec ratio. X6 Pro named best overall; X7 Pro flagged as a conditional steal.	Heat dissipation, outdoor brightness, local service network absent. Warranty entered only as caution attached to a price anomaly (X7 Pro below official SG retail of S$669).	Warranty identified as decision-critical only at this stage.
Warranty signal	AI-generated caveat on X7 Pro pricing	AI asked: ‘Are you buying local set or ok with export/parallel import?’ Warranty foregrounded as risk flag, not primary attribute.	Whether Galaxy A56 or Pixel 8a were local sets was never stated in either exchange.	Manual verification required on marketplace listings.
Resolution	Decided to verify POCO X7 Pro warranty status independently	Consideration set narrowed to POCO on performance-per-dollar grounds before warranty question arose.	Seller authorisation status unconfirmed from AI interaction alone.	Purchase deferred pending manual verification.

Note: Prompts reproduced from the original conversation log. OpenAI. ( 2026 , March 4). ChatGPT [Large language model with web search]. https://chat.openai.com.

This did not feel like a traditional search. It felt like consulting a product strategist. The AI produced a ranked shortlist with rationale and a narrative hierarchy. It did not merely provide facts; it shaped the decision-making process by foregrounding certain attributes while de-emphasising others, unless I asked. The key challenge shifted from ‘Which phone is best?’ to ‘What is this recommendation made of?’

Insights and lessons learned

AI changes how consumers search and evaluate. Classic consumer decision models such as the Engel–Blackwell–Miniard framework treat purchase decisions as sequential and consumer-directed (Engel et al., 1995). Dual-process theories distinguish slow, effortful System 2 reasoning from fast, heuristic-driven System 1 processing (Kahneman, 2011). Generative AI disrupts this architecture in three ways. First, search became a curated synthesis: the AI offered a ranked shortlist with rationale, pre-empting the external search stage. Second, evaluation became narrative: I was comparing stories (‘performance king’ versus ‘safe mainstream pick’) rather than specification tables. Third, cognitive effort shifted from gathering information to interrogating the recommendation. Under bounded rationality, consumers adopt heuristics to simplify complex decisions (Gigerenzer and Gaissmaier, 2011; Simon, 1955); AI has become one such heuristic, with the consequence that consumer agency shifts from information gathering to verifying a recommendation already structured by another intelligence (Puntoni et al., 2021). The scarce resource is no longer information but the consumer's capacity to evaluate it.

AI assigns salience before the user knows what to ask. In this interaction, the answer engine prioritised performance, battery capacity, camera capability, and price-to-specification value, while not foregrounding warranty regime, heat management, or outdoor visibility. This does not establish a general tropical bias, but it illustrates how locally relevant attributes may remain latent unless consumers explicitly prompt for them. The warranty question arrived not as the opening frame but as a conditional caveat, triggered by a price anomaly. For a consumer who accepted the initial shortlist without probing, the warranty question would not have arisen. Whoever sets the frame shapes the outcome.

The answer layer and generative engine optimisation (GEO). When consumers rely on answer engines rather than multi-tab search, the quality of the decision depends not only on consumer cognition but on the retrievability and structure of upstream information. Aggarwal et al. (2024) formalise this as GEO, demonstrating that structured content with statistics and citations can boost inclusion in AI-generated responses by up to 40 per cent. In this interaction, ChatGPT retrieved and cited specific Singapore marketplace prices, including third-party seller listings. Where web-enabled answer engines draw on marketplace listings, the completeness and structure of listing information can shape what the engine retrieves and foregrounds. GEO can improve answer quality when sellers provide complete, verifiable, and machine-readable product information. It may also create incentives for commercially motivated content to become more visible in answer engine retrieval, although the scale and effects of such practices in Singapore's retail environment require empirical investigation.

The warranty and parallel-import gap demands systemic solutions. The difference between a local set and a parallel import is not merely administrative. It changes warranty enforceability, service-centre access, and return friction. In this interaction, the warranty regime was never a primary attribute. It appeared only as a conditional caveat triggered by a price anomaly. AI recommendations can blur this boundary by presenting prices without signalling warranty regimes. Platforms should adopt a warranty-first design principle: just as price is a mandatory retrieval field, warranty regime should be a mandatory data point that any AI surfaces when recommending a product. This moves the burden from individual vigilance to systemic disclosure.

Open questions. Four practitioner questions feel pressing. First, who becomes invisible in AI-mediated retail advice: small enterprises, informal sellers, and non-English content communities whose discussions appear in local-language forums and regional Telegram channels unlikely to be optimised for AI retrieval? Second, how do we measure whether consumers can resist or reframe AI advice, not just accept it? Third, will answer engines accelerate replacement cycles, or can they be designed to foreground repairability, software-support duration, and warranty quality as components of value? Fourth, when consumers develop a felt attachment to a GenAI interlocutor, what service research describes as an emotional connection with AI that extends beyond functional assistance (Huang and Rust, 2024), does that relational bond amplify compliance with its retail advice beyond what the advice itself warrants, and if so, how should platforms and regulators account for persuasion that operates through relationship rather than content?

Future outlook and recommendations

For consumers: build answer engine literacy. Ask for trade-offs explicitly; request source diversity; verify warranty set type, charging bundle, and return policy manually. Treat the first answer as a draft, not a verdict. In Singapore and similar tropical markets, add climate-relevant prompts: ask how a device handles sustained heat or whether screen brightness suits outdoor use under equatorial sun.

For sellers and brands: optimise ethically. Adopt structured data markup (Schema.org Product and Warranty specifications) to ensure answer engines retrieve accurate local warranty terms. Brands that earn trust in the generative era will be those whose information is consistent, verifiable, and aligned with consumer-relevant outcomes.

For policymakers and platforms: mandate transparency. Mandatory retrieval attributes should include warranty regime (local versus parallel import) and prominent disclosure separating organic recommendations from sponsored or affiliate-linked placements. Google states that Text and Shopping ads are eligible to appear within AI Overviews in English in Singapore, with ad serving informed by both the user's query and the content of the AI Overview (Google, n.d.). Transparency requirements should therefore make clear when a commercially connected option enters an AI-generated answer and on what basis it is presented. Progressive disclosure, where high-level summaries are provided by default and detailed provenance is available on request, represents a workable approach that embeds transparency without sacrificing the efficiency that drives adoption.

Closing reflection

Buying a phone should be mundane. Yet the process revealed how quickly generative AI is becoming everyday decision infrastructure in Singapore's tropical platform market. It compresses complexity into clarity. Sometimes this is helpful. Sometimes it is premature. And as GEO matures, the question ‘What's best?’ may become inseparable from ‘What is most retrievable and citable?’

What surprised me most was not the quality of the AI's recommendations, which were broadly sensible, but how quickly I was willing to trust them. The conversational format created an intimacy that a list of search results never could. That ease of trust is precisely what makes transparency and literacy so urgent.

What the interaction log reveals most clearly is a sequencing problem. The question that mattered most for a Singapore buyer, local set or export set, arrived last. It appeared not as the opening frame but as a conditional caveat, prompted by a price anomaly rather than surfaced as decision-critical from the start. For consumers who do not probe, that question may never arise.

My takeaway is not to reject generative systems but to approach them as powerful intermediaries. Like earlier internet technologies, their effects are shaped by the social and economic structures in which they operate (Curran et al., 2016). The conversation about AI, consumer autonomy, and market fairness needs to happen now, before the answer layer becomes too entrenched to reshape.