Outlook
Part of planning a shopping feeds budget without ignoring regulatory shifts
Eight shopping feeds AI applications worth planning for now
Most teams bolt AI onto a broken feed. Here are eight shopping feeds AI applications, the risks they carry and the triggers that tell you when to act.
What to take away
- The most common mistake is adding AI to a feed that is already inaccurate, which scales the errors rather than the sales.
- Eight applications are realistic todaytitle and description generation, attribute completion, image clean-up, taxonomy mapping, anomaly detection, bid and budget signals, duplicate detection, and enquiry triage.
- Google Merchant Centre changes its policies and features continuously, so any AI workflow needs a review trigger tied to that changelog.
- Compliance risk sits with you, not the tool, so the Information Commissioner's Office enforcement record is worth reading before you automate personal data.
Where teams go wrong first
Most retailers buy an AI tool before they fix the underlying catalogue. If 15% of your product identifiers are wrong, a model will happily generate polished copy for the wrong product. The output looks better while the feed performs worse.
Fix data before buying AI
15% of IDs wrong
Model writes copy for the wrong product
Output looks better, feed performs worse
Start with a data quality baseline. Count missing attributes, duplicate entries and mismatched prices. Only then decide which application earns its place. The wider shopping feeds trends and outlook for England in 2027 sets out the shifts that make this sequencing matter.
Eight applications to plan for
- Title and description generation from structured attributes. Named tools: OpenAI ChatGPT, Anthropic Claude, Google Gemini, Shopify Magic.
- Attribute completion where suppliers send partial data. Named tools and services: Akeneo, Salsify, Productsup, Feedonomics.
- Image clean-up and background removal for non-compliant photography. Named tools: remove.bg, Cloudinary, Adobe Photoshop.
- Taxonomy and category mapping between your systems and each channel. Named standards and tools: Google Product Category, Shopify Taxonomy, Channable.
- Anomaly detection on price, stock and delivery time changes. Named tools and rules: Google Merchant Centre diagnostics, DataFeedWatch, Channable.
- Bid and budget signals built from feed-level performance data. Named platforms: Google Ads, Microsoft Advertising, Amazon Ads.
- Duplicate and near-duplicate detection across variants. Named tools and rules: Google Merchant Centre duplicate item warnings, OpenRefine.
- Enquiry and returns triage, routing customer messages by cause. Named services: Zendesk, Gorgias, Intercom, Freshdesk.
The named tools, providers, rules and services above are real examples you can evaluate.
Each application has a different risk profile. Copy generation is low risk if a human reviews it. Price signals are high risk because a wrong change hits margin immediately.
Platform changes drive timing
Google Merchant Centre publishes policy and feature changes in a public changelog, so the Merchant Centre announcements change log is the single best trigger for reviewing your automation. When a specification changes, your mapping rules and generated attributes may need rewriting.
The public Google Merchant Centre product data specification sets required attributes such as GTIN, brand, MPN and condition.
Set a monthly review. Someone owns the changelog, checks it against your feed rules, and logs any required work. Without that owner, AI workflows drift out of compliance quietly.
Compliance and data risk
Automation that touches customer data brings data protection duties. The ICO's enforcement action record shows penalties imposed on organisations that mishandled personal data, which is a useful reality check before you feed order data into a model.
UK GDPR sets the lawful basis and retention rules for that customer data.
Keep personal data out of feed generation where you can. Where you cannot, document the lawful basis and retention period. Our shopping feeds risk scenarios in England covers the failure modes that follow weak controls.
New sellers and first feeds
If you are launching a shop, the GOV.UK guidance on starting your business covers registration, tax and structure before you worry about feed automation. Get the legal basics right, then build the catalogue.
For a first feed, apply AI to attribute completion only. It removes the most tedious manual work and the errors are easy to spot.
Glossary
- Feed
- the structured product file sent to a sales channel.
- Attribute
- a single product field, such as colour or size.
- Taxonomy
- the category tree a channel expects.
- Anomaly detection
- flagging unusual changes in feed data.
- Enrichment
- adding or improving product data before submission.
Common questions
Do I need AI to run a shopping feed?
No. A clean, well-mapped feed outperforms an automated messy one. Add AI only where a specific manual task is slow and measurable.
Which application should I adopt first?
Attribute completion is usually the safest start. It fills gaps from existing structured data and a reviewer can check the output quickly.
How often should I review AI feed rules?
Monthly is a sensible baseline, plus a review whenever a channel publishes a specification change. Assign one named owner for that check.
What is the biggest compliance risk?
Feeding personal or order-level data into a model without a documented lawful basis. Keep customer data out of generation workflows unless you have a clear reason and record.
