Product catalogue optimisation: choose which products to improve first

Catalogue optimisation becomes manageable when you weigh commercial priorities, customer questions, information gaps and the evidence available to close them. Without that order, teams fall into two expensive habits: polishing the easiest records while important problems remain, and starting work on products nobody can substantiate.

This guide is a prioritisation method for a team with limited time. It decides which products go into the next batch. The product data optimisation guide covers what happens to each product once it is in the batch: the fact map, the claim-by-claim review and publication.

Start with the decision you want to help a customer make: checking compatibility, choosing a size, comparing capacity or establishing what a package contains.

Then name how the current record fails that decision. “Needs better content” is too vague to assign or verify. “Missing external dimensions, although they exist in the approved specification” is a task someone can complete and someone else can check.

Keep the objective and the metric apart. More complete attributes are an operational outcome. More qualified visits or purchases are commercial outcomes that need their own measurement, and a higher readiness score on its own cannot show that a change produced additional sales.

Urgent corrections

Misleading claims, incorrect offers and material contradictions need a correction route of their own. A known incorrect specification should never sink to the bottom of a list because a revenue-based score ranks the product low. For destination requirements, check the relevant specification: Google's Merchant Center product data specification is the reference for Merchant Center, not a universal checklist for every shopping system.

Evidence-ready improvements

These are useful changes backed by accessible, approved sources. They are the candidates for a focused enrichment batch, because the team can complete them and verify the result.

Evidence required

These are worthwhile questions whose answers are not yet established, such as a compatibility detail the supplier has not confirmed. Give each one an owner and a source-gathering task. Keeping them in their own queue stops the drafting queue turning into a holding pen for unanswered supplier questions.

The following is a suggested planning tool and no validated prediction of revenue or of how likely an AI assistant is to recommend a product. Score each factor from zero to two and keep a short reason beside each score.

• Commercial relevance: zero for low current importance, one for a useful supporting range, two for a documented priority range. Use your own demand, margin and strategic context, and do not read a sales forecast into the score.
• Decision importance: zero for a cosmetic improvement, one for helpful context, two for a gap that blocks an important purchase decision. Customer-service questions and on-site search terms are good evidence here.
• Evidence readiness: zero when no approved answer exists, one when the evidence needs clarification, two when the source is accessible and applies to the exact item.
• Completion feasibility: zero when unresolved dependencies prevent delivery, one when some coordination is needed, two when an owner can approve, publish and verify the change within the planned batch.

Use the total to organise the discussion, with urgent corrections handled separately. When evidence readiness is zero, the task moves to the evidence queue, however high its commercial score.

Imagine three records. This is an illustrative demonstration: none of the scores is a Sighted client result.

Product A is an available, commercially important backpack missing dimensions that an approved specification already confirms. It scores two on all four factors, eight out of eight. The team can answer a meaningful purchase question and verify that the answer was published.

Product B is a strategic accessory missing a compatibility specification the supplier has not confirmed. It scores two for commercial relevance, two for decision importance, zero for evidence readiness and one for feasibility: five out of eight. Because evidence readiness is zero, it goes to evidence gathering rather than straight to AI drafting.

Product C is a low-priority seasonal item with complete specifications and generic lifestyle wording. It scores zero for commercial relevance, zero for decision importance and two each for evidence readiness and feasibility: four out of eight. It is easy to do, which does not make it the best use of this week's review time.

If Product C also showed an incorrect price, that would go straight to urgent corrections. The scoring makes decisions clearer; it is no reason to leave a known error in place.

Estimate review and verification effort, not only drafting time. A description can be drafted in minutes while its approval waits on an unavailable expert or an integration change.

Keep the batch small enough to follow all the way through publication, and include a few representative edge cases, so you do not find out later that the process only works for straightforward products.

Name an owner for every record you hold back. At the next review, ask what evidence or dependency would move it forward, rather than rescoring the same unresolved task without changing its next action.

A batch you can name is also one you can measure. In Sighted's pilot with The Game Collection, 200 products were enriched and 17 collections created; AI citation coverage went from 0% to 32% of the priority products tested, and the store's organic conversion rate rose 64%.

Once the batch is chosen, the product data optimisation guide shows how to build the fact map for each product, and the AI product readiness checklist helps you assess the catalogue before you build the queues.

If the goal is visibility in AI shopping, favour information that answers real shopping requests. A complete but irrelevant field is not necessarily the next best investment.

Google says the best practices for SEO remain relevant to its AI features, such as AI Overviews and AI Mode, and that there are no additional requirements or special optimisations needed to appear in them (Google Search Central, “AI features and your website”, as of October 2026). That is a useful check against turning a planning score into an invented ranking formula for a platform; what OpenAI documents for ChatGPT is in the ranking factors guide.

The GEO vs SEO guide connects the batch to your wider search priorities. To see what changes afterwards, follow a defined measurement approach, such as the one in the AI shopping measurement guide, rather than declaring cause and effect from one favourable answer.

Record what was approved, held, published and verified, and compare the effort you estimated with the review time it actually took. If a whole group keeps failing approval because the supplier records are poor, improve the source process before generating more drafts for that group.

Look for changes that repeat. One resolved category mapping can improve many records, and a shared specification can support a group of verified updates. Check that the value really applies to each product before copying it across a product family: a size chart for one fit is wrong on another.

Should bestsellers always come first?

Bestsellers deserve consideration, but urgent corrections, blocked purchase questions and whether evidence is ready matter too. A bestseller with complete information may need less work than an important range with one specific gap.

Should you optimise every product to the same score?

No. No single score shows that every product has equal commercial value. Use scores to support a decision, with the underlying gaps and evidence visible beside them.

How often should the priority list change?

Review it when availability, campaigns, source evidence or business priorities change. Keep a change log so the team understands why a product moved up or down.

1. Name the customer decision each improvement should help, and the specific way the current record fails it.
2. Sort the work into three queues, urgent corrections, evidence-ready improvements and questions that still need evidence, and give each evidence question an owner.
3. Score each remaining product from zero to two on commercial relevance, decision importance, evidence readiness and completion feasibility, keeping a short reason beside each score.
4. Move any product with zero evidence readiness to the evidence queue, however high its commercial score, and send a known error straight to urgent corrections.
5. Choose a batch small enough to follow through to publication, with a few representative edge cases, and estimate review and verification effort as well as drafting time.
6. Build each product's fact map with the product data optimisation guide, and measure what changes afterwards with the AI shopping measurement guide.
7. After each batch, record what was approved, held, published and verified, and improve the source process for any group that keeps failing approval.

Sighted scores every product out of 100 on the data AI assistants read, lists the gaps that cost the most points first, and orders a bulk enrichment run so the products with the largest gaps come first, within your plan's limit. It drafts missing fields from your own descriptions, options and photos and, unless you switch on automatic publishing, holds every draft for your approval.

Sighted does not know your margins, campaigns or which supplier answers are still outstanding, so commercial relevance and evidence readiness stay with your team. The free AI visibility audit gives the gap list for up to 50 of your product pages before you commit to a batch; the enrichment that follows, written back on Shopify, is the technology behind SO-ME.