Case study: getting The Electric Fire Company ready for AI
How Wyshbone audited The Electric Fire Company's Shopify catalogue of about 85 electric fires, worked through the fixes with the owner, and put an AI assistant on their website that answers from the cleaned-up data.
- Before
- Named in 1 of 24 AI answers. ChatGPT: 0 of 8
- Biggest find
- 75 of 85 products had no product type
- Our side
- All 85 products read and labelled by our AI
- Owner's side
- A plain-English fix list. Minutes per change, no special skills
- Fixed
- Product data, Google titles, web addresses and pages, each checked live
- AI assistant
- Live since 13 October, 97.5% of facts right in testing
"Just a quick thank you for all your hard work on the assistant... I've no doubt there's a lot of your time and expertise in there too."
Michael Paine, The Electric Fire Company
The shop
The Electric Fire Company sells electric fires and stoves from the leading brands, online through their Shopify store and from a showroom in central London. In September 2026 their catalogue had about 85 products, which puts them on our small-shop plan.
Visit The Electric Fire Company
Before: AI barely knew they existed
On 13 October 2026 we asked ChatGPT, Gemini and Claude eight questions a customer might ask, such as where to buy a particular electric fire in London. The Electric Fire Company was named in 1 of the 24 answers. ChatGPT didn't mention them once. The assistants sent shoppers to other fire shops instead.
What the audit found
We read every product page against the store's own collections and product fields, the way Google and AI assistants read them. The website looked fine to a shopper. Underneath, the data told a different story:
- 75 of 85 products had no product type. The shop's own "Electric" filter showed 10 fires, although every product they sell is electric.
- Products whose data contradicted itself: a suite described as freestanding that also sat in the Wall Mounted collection, a ceiling-suspended fire listed as a stove, and an insert in the Stove collection although its description says it does not include a stove.
- Four product pages created by copying another product, carrying the other product's settings and, in one case, a misspelt name.
- 32 products that describe themselves as inset fires but were missing from the Inset collection, so the shop's own filter under-showed them.
- Google titles and descriptions that didn't name the product properly, two web addresses with typos, and typos on information pages.
What we did on our side
Our AI read every one of their 85 products and labelled what each one actually is: its fuel, its form (stove, insert, suite, wall fire) and how it fits, each with the line from the shop's own page that proves it. A fire that looks like a cast-iron wood burner but runs on electricity is labelled as electric. Where the shop's own data disagreed with itself, it was flagged for a person to check rather than guessed.
We also read the heat output from 82 of the 85 product pages, collected the questions customers ask that the website didn't answer, and added the owner's answers so the assistant can give them.
Easy for the owner: no special skills needed
The shop's owner fixes their side in Shopify, and we do ours. Each item on the fix list is written in plain English: which page, what to change it to, and why. The owner makes each change in the normal Shopify admin, the same way they would edit a product, and marks it done. Most changes take a few minutes. No coding or technical knowledge is needed, and when an item is marked done we check the live page to make sure the fix is really there.
Fixes have gone live across their product pages, Google titles and descriptions, web addresses (with the old addresses redirected) and information pages, and every one has been checked on the live site. The owner tracks it all on a check-in page we built for them, and a weekly check reads their listings for anything new.
The AI assistant
The Wyshbone AI Assistant has been live on The Electric Fire Company's website since 13 October 2026. Before it went live we tested it with hundreds of realistic customer conversations; in one full run of 324 answers, 97.5% of the facts it gave were right, and the mistakes found were fixed and retested. It answers customers from their real products, prices and stock, plus their showroom hours, delivery and returns. When a customer asks something the website doesn't answer, it gives the shop's phone number and email rather than making something up. They were pleased enough with it to email the link to their own suppliers so they could try it too.
Keeping the assistant up to date is just as easy. We read what customers ask it, and questions the website couldn't answer, such as which fires are on display in the showroom or what a fire costs to run, went back to the owner as a simple list. The owner just types an answer in plain English, and it becomes part of what the assistant can say. It takes minutes, and no special skills are needed.
Why it matters
AI assistants recommend the shops they can read and understand. A filter that hides 75 products from itself, or a fire that is listed as two things at once, confuses an AI tool as much as a shopper. Fixing the data once makes the shop easier for Google, ChatGPT and Gemini to understand, and gives the shop's own assistant something accurate to answer from.