Run a Useful Product Data Audit

Most online shops do not have a product problem. They have a product data problem, and they have not noticed yet.

The catalogue grew over years. Someone added a line, copied an old description, left a field blank because it did not seem to matter, used "Various" for a colour, put the size in the title on some items and in an attribute on others. None of it was wrong, exactly. It all just drifted. And for a long time it did not cost anything, because the only thing reading the data was a human looking at a page.

That has changed. Four different things now read your product data, and three of them cannot ask you a question when something is unclear.

The same fields, read four ways

Here is the thing worth understanding before you spend a penny on anything clever. Your product data does not do one job any more. It does four, and they all read the same fields.

The first reader is the shopper. They want to know what it is, whether it is in stock, what it costs, and whether it is the right size, colour, or material. Vague or missing details here cost you a sale directly, while the person is deciding.

The second is ordinary search. Google reads your product pages and your feed to decide what you sell and when to show it. Its feed expects the basics done properly: a clear title, a real description, a price, availability, a good image, and a consistent identifier. Miss those and products get disapproved or quietly ranked below shops that filled them in.

The third is AI search. When someone asks ChatGPT or Google's AI Mode for "a waterproof jacket under £80 for a tall person", the tool is not browsing your shop. It is reading structured data and picking the products it can describe confidently. Thin data does not get chosen, because the tool has nothing specific to say about it.

The fourth, and newest, is the shopping agent: software acting for the customer, comparing options and sometimes buying. An agent cannot squint at a photo to work out the size. If the attribute is not in the data, the product effectively does not exist to it. The industry view forming through 2026 is blunt about this: if a product cannot answer a machine's question cleanly, it gets excluded from the shortlist.

Notice that only two of those four are AI. This is not an AI project. It is the groundwork that happens to make AI work too.

A product data audit checks five areas

You do not need a tool for the first pass. You need a spreadsheet of your products and a couple of hours. Go field by field and ask, for each one, "could a stranger who cannot see the photo act on this?"

• Titles. Consistent shape, the things people search for, no dumping every keyword in. "Merino base layer, men's, long sleeve" beats "BEST Thermal Top Warm Winter Mens Ladies".
• Prices. Right, current, and matching what shows at checkout. Agents and feeds both punish a price that disagrees with the page.
• Options. Size, colour, and material as proper structured variants, not buried in a sentence. This is where most catalogues are messiest.
• Stock and availability. Accurate and reasonably live. "In stock" that is not is worse than "out of stock", because it burns trust at the worst moment.
• Descriptions. Written for a person, but specific. What it is, who it suits, the dimensions, the care, the catch. Specific beats poetic for every one of the four readers.
• Identifiers. Brand, and a GTIN or part number where the product has one. Dull, and the single thing that lets every system match your product to the same product elsewhere.
• Images and alt text. A real image on every item, and alt text that describes it. The alt text helps screen-reader users, ordinary search, and AI tools at the same time.

You will find the same five sins on most catalogues: blank fields, sizes trapped in titles, "Various" doing the work of real options, descriptions copied across products that are not the same, and prices that have quietly fallen out of sync. Fixing those is unglamorous and it moves all four numbers at once.

An audit will not fix missing facts

Clean data is the floor, not the ceiling. It will not sell a product nobody wants, rescue a price that is wrong for the market, or fix a checkout that loses people on the last step. If sales are flat because the range is wrong, a tidy feed just describes the wrong range more accurately.

It is also possible to over-do it. A small shop with forty products does not need forty attributes per item. Fill in the fields your category actually uses to compare and decide, and stop. The goal is data a stranger can act on, not a museum of metadata nobody reads.

Clean product data before adding AI features

We learned this most clearly on Modern Printworks, our own online art shop. The interesting part is usually described as the AI: a vision model looks at each incoming artwork image and drafts the title, description, alt text, and suggested tags, and an on-site assistant answers from the real catalogue and can add a configured item, with its size and materials, straight to the basket.

None of that works on thin data. The assistant can only add a framed print to the basket because size and material exist as proper options, not as a line in a paragraph. The AI answers are only trustworthy because the underlying product data is clean enough to answer from. The same investment in good structured data made each product better at all four jobs at once. Skip it, and the clever features would have had nothing solid to stand on, and the failures would have been the kind a customer sees.

That is the buying heuristic, and it is worth stating plainly: do the product data audit before you buy any AI feature, assistant, or feed tool, not after. An AI layer on top of messy data does not clean the data. It amplifies the gaps and states them more fluently. If a salesperson is pitching you an AI shopping assistant and has not asked to see your product data first, that is the question they skipped.

We tend to look at a client's product data before we look at anything else they want to add, because it is the cheapest work that improves the most things, and because it is usually the part holding the rest back. It rarely makes it onto a roadmap, which is exactly why it is worth putting there first.

If your catalogue has drifted over the years and you want a clear read on what to fix before you add anything to it, have a 20-minute chat. Bring the messy spreadsheet. That is the useful one.