A customer has a jacket in their basket. They are between a medium and a large. The size chart gives chest measurements, but they want to know whether it runs small, and whether there is room for a jumper underneath. There is nobody to ask, so the basket sits there. Tomorrow they buy a jacket from a shop with a clearer answer.
Fit questions are the quiet killer of online orders. Not price, not delivery. Just uncertainty about whether the thing will work when it arrives, and the hassle of sending it back if it does not.
An AI shopping assistant is good at this narrow job, provided your product data holds the answers. When it does not, the assistant becomes a very efficient way of finding out what your product pages are missing.
Four kinds of fit question
"Will this fit?" means different things depending on what you sell. It helps to separate them, because each needs different data behind it.
- Body fit. Clothing, shoes, helmets, rings. "Does this run large?" "I'm usually a 10 in trainers, which size here?" Needs size charts, fit notes and honest comments about how the cut compares to standard sizing.
- Physical dimensions. Furniture, appliances, planters, bags. "Will this sofa go through a 76cm doorway?" Needs full measurements, including packaged size and whether legs come off.
- Compatibility. Spare parts, accessories, chargers, filters, cases. "Does this fit a 2019 model?" Needs a compatibility list that names models, not "fits most".
- Suitability. "Is this tent OK for two adults and a dog?" "Can this pan go in the oven?" Needs specifications written in terms of use, not just numbers.
Most shops are decent at one of these and weak on the rest. Clothing retailers publish size charts but rarely say whether a style runs narrow. Furniture shops list dimensions but forget the box size.
What the assistant can only know if you tell it
A grounded assistant answers from your content. That is the point: it will not invent a measurement. It also means the quality of its answers is capped by your data. Here is a quick audit worth running before you switch one on.
| Question shoppers ask | Data the assistant needs | Where it usually lives, if anywhere |
|---|---|---|
| Does it run small or large? | Written fit notes per style | Often only in staff heads |
| Will it fit through my door? | Assembled and packaged dimensions | Packaged size is often missing |
| Does it work with my model? | Named compatibility list | Supplier spreadsheet, not the site |
| Can I wear a layer under it? | Garment measurements, not body sizes | Rarely published |
| What if it doesn't fit? | Returns and exchange policy, with timescales | Policy page, usually complete |
The fastest wins tend to be the "staff heads" row. Ask whoever packs returns what the common reasons are. Write those down as fit notes and add them as Q&A pairs or a short document. You can upload a CSV of product specifications, or let the crawler read your product pages, and add the notes on top.
Training an ecommerce chatbot on your product catalogue covers the mechanics of getting catalogue data in.
The honest answer when the data runs out
The dangerous moment is when the shopper asks something your data does not cover. A made-up "yes, that will fit" produces a return, a bad review and a customer who does not trust the chat again. The better answer sounds like this.
Visitor: Will the oak side table fit in a gap that's 42cm wide?
Assistant: The oak side table is 45cm wide at the top, so it would be about 3cm too wide for that gap. The narrower walnut version is 38cm wide and the same height, if that style works for you.
Visitor: Does the walnut one have the same drawer?
Assistant: I don't have details on the walnut table's drawer in the product information I can see, so I don't want to guess. I can pass your question to the team and they'll email you, or if you order and it isn't right, you can return it unused within 30 days for a refund.
Two things happened there. The first answer used real numbers and suggested an alternative, which is what a good shop assistant does. The second admitted a gap, offered a person, and used the returns policy to lower the risk of buying. The 30 days is whatever your policy actually says, which is why the policy needs to be in the assistant's content. When a receptionist should say "I don't know" makes the case for this behaviour.
Using the returns policy without hiding behind it
A clear returns policy is a legitimate part of answering a fit question. "If it doesn't fit, exchanges are free" removes most of the risk a hesitant shopper feels.
What you do not want is an assistant that answers every fit question with "you can always return it". That dodges the question and trains shoppers to over-order and send back. The order of priority should be: answer from the data, suggest an alternative if relevant, then mention the returns policy as a safety net.
Put that order into the assistant's instructions in plain words. It makes a noticeable difference to how answers read.
What to do with the questions it could not answer
Every unanswered fit question is recorded as a knowledge gap. Treat that list as a product page to-do list. If three people this month asked whether a boot is wide-fitting, the boot's page should say so, and so should the assistant's content.
A weekly routine for a small shop can be short:
- Open the unanswered questions list and group them by product.
- For each product with a repeated question, find the answer (supplier, measuring tape, whoever handles returns).
- Add it to the product page and to the assistant's Q&A pairs.
- Check next week whether the question has stopped appearing as a gap.
Over a couple of months, the product pages improve for everyone, including shoppers who never open the chat.
Where to start
Pick the product category that generates the most fit-related returns or pre-sale emails. Gather its measurements, write fit notes for each item, and add your returns policy word for word. Then put the assistant live on those product pages only and read a week of conversations. If you want to see how grounded answers behave before you commit, the demo shows one in action, and the questions people ask right before buying is useful for building your list.