XCutlery 小型 Shopify 刀具店的 GEO 實作筆記

I look after the site for XCutlery, a small Shopify store selling hand forged kitchen knives. Cleavers, chef knives, butcher knives, fillet knives, that sort of thing. In earlier posts I wrote about indexing fixes and collection page cleanup. This one is about the part everyone keeps asking me about at the moment: how a tiny store gets mentioned when someone asks ChatGPT, Perplexity or Google AI Overviews what knife to buy.

The industry has decided to call this generative engine optimization, or GEO. Some people say AI search optimization. I call it regular SEO for a reader that skims.

Some context so nobody thinks I'm sitting on a secret. The store is young in search terms. Early Search Console showed 41 URLs in the sitemap, 18 of them indexed, and somewhere between 2 and 7 impressions a day. So treat this as a field report from the start line.

How AI answers pick sources, roughly

I don't have access to anyone's ranking internals and neither does the person selling you a GEO course. What you can say in general terms is this.

Most AI search tools pull a set of pages from a search index, read them, and write an answer from what they found. Some show the sources as citations. Google AI Overviews sits on top of Google's own index, so if a page isn't indexed it isn't in the running. Perplexity cites heavily and visibly. ChatGPT with search does something similar, while its answers without search lean on whatever it learned during training.

From that, a few practical things follow:

The page has to be crawlable and indexed first. Boring, but nothing else matters until that's true.
The model favours text that answers the question plainly, near the top, without making it dig.
It seems to trust facts that agree with each other across many places on the web.
It picks up on brands that other sites talk about, not only brands that talk about themselves.
Answer first writing

The biggest change in how I write now is putting the answer at the top of any page that tries to answer something.

The old pages opened with a warm intro about craftsmanship and got to the useful part eventually. A model has to wade through that to find the fact. So does a human, who usually leaves.

New habit: the first one or two sentences answer the question directly, then the detail follows. On the hand forged meat cleavers collection the goal for the short intro is to say what the knives are for and what steel they use before any story about forging. The longer copy sits below the product grid so shoppers see knives first.

Product pages follow the same rule. A page like the Kraken 8 inch hand forged cleaver should open with what it is and what it handles, with the descriptive stuff after. The aim is something that reads like someone answering a question at a counter.

Entity consistency, or stop telling different stories

This is the one I'd fix first if you only had an afternoon.

An AI model builds a picture of a business by stitching together everything it finds. If your About page says one founding year, your Etsy profile another, and a directory listing a third, the model either picks one at random or decides it's not confident enough to mention you. Neither helps.

So I wrote down the core facts once and made every page and profile match them:

Brand name: XCutlery, spelled the same everywhere
Based in Austin, Texas
Founded in 2019
Hand forged, fully handmade knives in high carbon steel, following a Scandinavian forging tradition
Ships worldwide, with a separate Australian domain for that market

The job is making those facts say the same thing on the About page that covers who makes the knives, the footer, the contact page, social bios and any directory profile I set up. If one changes, all of them change. Dull bookkeeping, and probably the most useful kind.

The same goes for policy facts. Out of stock knives are sold as pre orders and forged to order, which can take 2 to 3 weeks before shipping. That line should appear the same way on product pages and in the knife FAQ, so a model doesn't find two versions of the shipping story.

FAQ blocks written as real buyer questions

We have FAQ blocks on the main collections, below the product grid, plus a central FAQ page. The trick was writing them the way people actually type or speak questions, which is also how they ask an AI assistant.

Questions like:

What cleaver cuts through bone?
Is a hand forged chef knife worth it?
How do I care for carbon steel compared with stainless?
Do high carbon steel knives rust?

Each answer starts with a direct reply in the first sentence, then a couple of lines of detail. The bone question links to the cleavers, the chef knife question links to the hand forged chef knives collection, and the care question links to the high carbon steel knives collection.

I was honest in the care answers too. High carbon steel will darken and can rust if you leave it wet in the sink overnight. Pretending otherwise might sell a knife and then earn you a very specific review. The lifetime warranty page covers what's actually covered, and the FAQ should point there instead of inventing a softer version.

Comparison style content

People ask AI tools comparison questions all the time, like whether they need a cleaver or a chef knife, or whether carbon steel is worth the extra care. If your site never compares anything, you're relying on someone else's page to describe your product category for you.

So I'm adding plain comparisons to the guides and collection pages. The kind of thing I mean:

A Serbian chef knife is a wide, heavy blade that sits somewhere between a cleaver and a chef knife. The Serbian chef knives collection explains when that shape makes sense and when a normal chef knife is easier to live with.
A heavy cleaver like the Warthog 6 inch meat cleaver is built for chopping through joints, while a chef knife like the Ironclad 8.7 inch hand forged chef knife is the daily slicing workhorse.
A long fillet blade such as the Barracuda 13 inch fish fillet knife does a job neither of those should be doing.

If a cheaper stainless knife is the better pick for someone who hates maintenance, the page should say so. Balanced pages seem to get picked up more than sales pitches.

Getting mentioned on third party sites

Here's the bit you can't do from inside the Shopify admin.

AI answers lean on what the wider web says. If the only site vouching for a brand is its own domain, the model has little reason to believe it. So part of GEO for us is ordinary, slow, honest presence elsewhere:

Answering knife questions on cooking and knife forums, with a link only where it genuinely helps
Setting up consistent directory listings with the same name, location and founding year
Writing posts like this one, where I'm openly the person who runs the site

On reviews, one thing I'm glad about: the customer reviews page holds genuine reviews imported from the owner's Etsy shop, where he sold for about five years. That history is real, and it means the reviews tie back to an independent platform.

There's one honest to do on my list. Some product images are not real photography yet, and real workshop photos are planned. Buyers trust real photos of the actual forge, and I suspect the machines will too.

Checking AI visibility by hand

There are paid tools that track AI mentions. For a store with 24 products and 11 collections, I just do it by hand and write the results in a spreadsheet like it's 2009.

My routine:

Open ChatGPT with search, Perplexity and a Google search that triggers AI Overviews.
Ask the same set of buyer questions each time, phrased naturally. Things like "best hand forged chef knife from a small maker" or "what meat cleaver can cut through bone."
Note whether XCutlery is mentioned, which page is cited, and what competitors show up.
Look at which sources the tools cite even when we're absent. Those are the pages and forums worth earning a mention on.

Answers shift from one run to the next, even on the same account, so I treat any single result as an anecdote. When a tool gets a fact wrong about the shop, that's a sign some page or profile somewhere is still telling an old story.

What you can't control

A fair amount, unfortunately.

You can't control which sources a model decides to trust on a given day. You can't control training data that was collected before you fixed your About page. You can't force a citation, and anyone promising guaranteed placement in AI answers is selling something I wouldn't buy. Google can also show an AI Overview so complete that nobody clicks through, which is lovely for everyone except the store.

What you can control is the boring list above: get pages indexed, put answers at the top, keep facts matching everywhere, write FAQs in buyer language, compare honestly and earn real mentions elsewhere. All of it also helps normal ecommerce SEO and Shopify SEO, so even if AI search changed its mind tomorrow, none of the work would be wasted.

If you want to see how this looks on a live store, the full hand forged knife catalogue and the guides on the blog are where this work lives. And if a buyer wants something that isn't in stock, the custom knife builder lets them pick the blade style, steel, size and handle wood themselves.

So, a question for anyone doing GEO on a small store: have you actually seen an AI tool start citing you after a specific change, or does it still feel like weather to you too?