THE SHORT ANSWER
AI is changing commerce by helping systems interpret natural-language needs, rank products, personalize experiences, generate or enrich content, support customers, forecast demand and analyze patterns. Its value depends on data quality, objective design, evaluation, privacy and human control.
Where AI changes the commerce journey
| Application | Potential value | Failure to monitor |
|---|---|---|
| Conversational discovery | Translate needs into product exploration | Invented features or unavailable products |
| Recommendations | Rank relevant products or complements | Popularity bias, repetition or harmful optimization |
| Merchandising | Adapt ranking to context, stock and objectives | Opaque trade-offs and margin-only ranking |
| Product content | Draft, classify, translate or enrich catalog information | Fabricated specifications and inconsistent claims |
| Customer support | Answer routine questions and retrieve order context | Wrong policy, disclosure or unauthorized action |
| Forecasting | Estimate demand or inventory needs | Regime change, sparse data and false precision |
| Analysis | Find patterns and propose questions | Confusing correlation, attribution and causality |
Product data is the foundation
A discovery assistant cannot reliably compare size, compatibility or availability when those fields are missing or contradictory. Establish stable identifiers, variants, attributes, price, stock, delivery and returns before asking a model to infer what the catalog should have said.
Well-structured information also supports feeds, search and machine retrieval. Read Product Pages That Convert and content for AI discovery for the human and retrieval layers.
Evidence & context: Google Merchant Center Help · Google Search Central
Personalization optimizes an objective, not customer welfare
A recommender can rank for clicks, conversion, revenue, margin, diversity or a longer-term signal. These objectives can conflict. Optimizing the easiest proxy can produce repetitive, manipulative or commercially weak experiences.
Define the customer job, eligible inventory and guardrails. Compare the model with a simple baseline, inspect segments and measure downstream paid orders, contribution, returns and satisfaction—not only clicks.
Evidence & context: Google Machine Learning education
Keep the limits visible
- Poor or stale data propagates into confident outputs.
- Generated product claims can be wrong even when they read fluently.
- Personal data requires purpose, minimisation, access control and applicable consent.
- Attribution cannot prove that personalization caused the outcome.
- Automation can scale pricing, content or service errors quickly.
- Over-personalization can narrow discovery or feel intrusive.
Evidence & context: UK Information Commissioner's Office · Google Analytics
Adopt from a bounded decision
Start with one use case, a baseline, an owner, permitted data, human escalation and a rollback path. Evaluate quality and business impact before expanding autonomy. AI in Digital Marketing develops the wider strategy, measurement and governance context.
Sources & further reading
- Recommendation systems: scoring
Google Machine Learning education. Conceptual explanation of scoring and ranking candidates, not a description of every current commercial recommender.
- Product data specification
Google Merchant Center Help. Official product-data requirements covering identity, variants, price, availability, shipping and returns. It is Google-specific, not a universal commerce schema.
- Optimizing your website for generative AI features on Google Search
Google Search Central. Google-specific guidance on AI search and SEO, checked 11 September 2026. It does not establish how all assistants retrieve information.
- How should we assess security and data minimisation in AI?
UK Information Commissioner's Office. UK regulatory guidance, checked 11 September 2026. Jurisdiction-specific context, not individual legal advice or permission for a particular use.
- API dimensions and metrics
Google Analytics. Official GA4 reporting definitions checked 13 September 2026, including session source, medium, referral and landing-page dimensions. Attribution remains limited to observable interactions.
Examples and exercises are illustrative unless attributed to a source. No independent expert review is claimed.
A correction, a counterexample or an experience worth sharing?
Join the conversation ↗