Why AI isn't recommending you

Your store is one of the most AI-readable stores we have tested. Every crawler we sent at it got the full page. The problem is not access. It is that the answers you have already written are sitting where AI cannot use them, your main product is missing from every route your store publishes, and the two assistants we tested disagree completely about you.

What we found

Good news first, and it is real. Your store passes the test that most stores fail. We sent six different crawlers at your homepage, including OpenAI's, Perplexity's and Anthropic's, and every single one received an identical 377,096 bytes of complete HTML. No blank shell, no JavaScript wall, no bot getting a worse version than a human. Your security headers are all in place. Shopify is also auto-publishing the new AI agent files, so you are on the right side of a platform shift that happened quietly in the last few months.

Now the part that matters. You have already written genuinely good answers to the exact questions guitarists ask an AI assistant before they know a product like yours exists. There is a real, technical answer to why strings rust on your site right now. There is a named corrosion scientist quoted by name. There is a signed, dated lab report. This is the material AI systems reward most heavily, and you have it natively, which most brands do not.

It is all invisible. The answers sit as unmarked prose with none of the structure that tells an AI system "this is an answer to that question." Your flagship case cannot be found through any route your own store publishes. And the one file on your site whose entire job is to introduce your brand to AI contains not one word about guitars.

The three things that matter

• The hard part is already solved. Being readable by AI is where stores usually fail, and you pass it cleanly on every crawler we tested. You do not need to fix your platform, your theme, or your rendering.
• Your standalone case cannot be found by anyone who does not already have the link. It is absent from every discovery route your store publishes. That follows automatically from its Unlisted status, which is the correct setting for a bundle component. So the question is not the setting, it is whether you intend to sell that case on its own.
• You already own citable material that nobody can cite. A named scientist, a dated lab report, and eighteen real buyer questions with real answers. Marking these up is a few hours of work and it is the difference between having evidence and having evidence AI can use.
• Your visibility is not one number, it is different on every assistant. On one assistant you place sixth of seven on your core query. On Perplexity you are the lead recommendation on that same query, by name. Any single figure for "AI visibility" would have hidden that.

What we tested

The question we set out to answer is a practical one. When an AI assistant reads your store, what does it actually see, and is that the version of your brand you would want it to repeat? So we bypassed JavaScript entirely, which is what these crawlers do, and read the raw HTML each one is served. Then we checked whether the things worth citing are marked up in a way a machine can lift.

This is the third pull on these surfaces. We re-fetched everything and compared it to the 1 and 2 September pulls rather than trusting notes, because the store has been actively changing. In two days the only change was a line Shopify added to a file automatically. Nothing else moved.

Every AI crawler gets the full page

This is the most common failure point in the audits we run, and you pass it without qualification.

Identical byte counts across all six matter more than they look. It means no crawler is being fingerprinted and served something different, which is a real and common problem. Your product titles, descriptions, prices, reviews and structured data are all visible the moment a model crawls you.

Your headings are also in good shape on the pages that count. The homepage leads with a single clear heading and ten descriptive section headings underneath. That is the anchor AI uses to work out what a page is about, and yours is correct. One exception: your About page has no top-level heading at all. That is the page an AI system reads to work out who you are, and it currently opens with no anchor.

The agent layer is live, and completely empty

Shopify has started publishing a set of files written specifically for AI assistants and AI shopping agents. Your store exposes the full set, which puts you ahead of most stores by default.

You did not pay for any of this and you did not configure it. It arrived with the platform. The problem is what is inside.

We searched all three of your AI-facing text files for the words guitar, rust, patent and corrosion. The result across every file was zero occurrences of any of them. These files describe how to browse and buy from a Shopify store in general. They contain no brand description, no product claim, no mention of your lab test, no answer to any buyer question. An AI agent that reads them learns how to transact with any Shopify store. It learns nothing about why it should recommend yours.

The agent-specific sitemap is the sharpest version of this. It is listed first in your sitemap index, ahead of your products. It contains exactly one URL, and that URL is the generic instructions file above. So the one surface on your store built to tell an AI agent what is here resolves, in full, to a document about how to shop at any Shopify store.

One detail worth noting because it shows how little of this is yours. Over the two days we watched these files, the only change to any of them was a developer link Shopify added automatically. The "full detail" version of your brand file is five bytes longer than the short version.

Your main product is invisible to AI agents

This is the biggest finding in the audit and the cheapest to fix.

Think of your store as a shop with a catalogue at the door. An AI shopping assistant walks up, reads the catalogue, and can only consider what is listed. Your $59.99 protection case is not in the catalogue. Not in any version of it.

We did not assume this from one check. We tested every route your store publishes for discovering products, individually:

The cause is the product's Unlisted status. Shopify responds to that by serving the page a noindex, nofollow instruction and excluding it from your sitemap. Both happen silently.

To be clear, that status is not a mistake. Unlisted is the correct, documented setting for a product that exists as a component of a bundle, which is what this case is. We are not telling you to flip it. Everything above describes a consequence of a correct setting, not a misconfiguration.

The product is still reachable if an agent already knows its exact address. Nothing your store publishes ever supplies it, and no page on your site links to it: we checked your homepage, bundle page and About page, and found zero links to it between them. We tested the trademark symbol in the address too, in case that was the culprit. It is not.

One important qualification, because we tested it rather than assuming. Perplexity cites this page anyway. It appears as a source on a live query we ran, despite the hidden instruction, the sitemap exclusion and the absence of any link to it. So the page is not unreachable by every AI system. It is missing from every route you publish, which is a different and more fixable problem. What you cannot do today is rely on discovery: an assistant that has not already picked it up has no way to find it.

You already wrote the answers. Nothing can read them

This is the finding we did not expect, and it changes the shape of the work.

AI assistants reward content that answers a specific question in a self-contained way, and they reward it most when it carries named sources and real numbers. You have this. On your bundle product page there are nine real buyer questions with real answers, and nine more on your homepage. These are not filler. One of them reads:

That is a genuinely good answer. It is specific, it is technical, and it explains something a guitarist actually wants to know. And when we asked a live AI assistant that exact question, your brand did not appear anywhere in the answer or in any source it cited. The assistant built its response from guitar forums and blogs instead. Two of those sites answer the question with their own dedicated articles.

You are not missing the content. You are missing the markup that tells a machine the content is an answer.

The same is true of your strongest asset. Elsewhere on that page you quote a named corrosion scientist by name and title, answering whether the product still helps on coated strings. A named expert quote is one of the two highest-value citation signals there is. It is sitting in unmarked prose.

Your product and pricing data is marked up properly, which is why an agent can read your prices. Nothing else is. There is no question-and-answer markup and no step-by-step markup anywhere on the store.

Worth stating plainly, because it corrects our own earlier read. Our first two passes recorded that this kind of content did not exist on the site at all. It does, and there is more of it than most brands your size have. The gap is narrower and much cheaper to close than we first wrote down.

Your patent claim contradicts itself

Your site says two different things about your patent, and it does so on four separate pages.

Those two phrases mean different things. One says you have applied. The other says you have been granted. No patent number, filing date, or registry reference appears anywhere on the site, so there is nothing an AI system could check against.

The sharpest version is on your flagship product page, where both claims appear on the same page: one "Patent Pending" against four uses of "patented". An AI system does not need to compare two pages to catch that one. It is visible inside a single document, on the page most likely to be an agent's target.

This matters more for AI than for search. AI systems weigh whether a source is internally consistent before deciding to repeat what it says. A page that contradicts itself about its own central claim is a page a model is less willing to vouch for. There is also a plain accuracy question here that is worth settling with whoever handles your filing, separately from anything to do with AI.

No independent source corroborates you

AI assistants do not just cite the most relevant page. They cite the most corroborated one. An assistant recommending a product is effectively vouching for it, and it prefers to vouch when someone other than the brand has said something.

We ran two assistants. They disagree sharply about you, which is worth understanding on its own. But underneath the disagreement they share one thing, and it is the more important half.

Assistant A

Perplexity, tested 3 September

Two things follow from the disagreement. First, on your core commercial query you are already winning on one major assistant and losing badly on another, so there is no single number for your AI visibility and you should not accept one. Second, the problem-aware absence is the finding that held on both assistants. That is the durable gap, and it is the one your existing unmarked answers would fix.

We checked the sources rather than counting them

Perplexity's answer to the branded question listed 38 sources, which looks like healthy coverage. It is not. We opened them.

Three were Reddit threads. We read the first one: it is a discussion about Fast Fret, a string lubricant sold in a can, and it does not mention your product once. One cited Amazon listing is an electronic mouse trap. Another source is a badminton string. Others were about a Toyota dashboard and an iPhone case.

So we asked Reddit directly instead. Searching Reddit's own index for your brand returns no results at all, on both spellings, "string protech" and "stringprotech". Reddit does not know you exist. It offered racket-stringing communities instead.

This is the finding that has not moved across three days and two assistants. Everything else in this audit is a fix you can ship. This one is a slower build, and it is the reason a good answer on your own site is not yet enough: when an assistant wants a second opinion about you, there isn't one.

Your entity signals are close to empty, which makes this worse. Your store publishes a machine-readable list of your other verified profiles so an AI system can confirm you are a real brand. It has nine slots and one is filled, with Instagram. Your brand description, contact email and phone are all absent from that same block. There is no review platform an assistant can point to for a rating, and no organic forum discussion for it to quote.

For context, your own analytics already shows this. Two sessions arrived from an AI assistant in the trailing ninety days, against roughly 2,400 sessions total. That is not zero, so the door is open and a couple of people have walked through it. There is almost nothing waiting for them.

What we could and could not test

We want to be exact about this, because it is the one real limit on the citation findings above.

Two assistants were queried directly with live web search on real dated runs. The first was tested across three separate days. Perplexity was added on 3 September, and adding it changed two conclusions, which is exactly why it was worth doing.

We have not tested ChatGPT, Gemini or Microsoft Copilot. On the evidence that two assistants disagreed this sharply, we would expect those three to differ again rather than to confirm either result. Do not read this document as a census of all AI assistants.

We did not take any assistant's source list on trust. We opened the cited pages and searched Reddit's own index directly, which is how we established that several of the sources offered as evidence about your brand are not about your brand at all. Where we state that something exists or does not exist, we checked it rather than counting a citation.

The technical findings in Sections 03 through 07 carry none of these limits. They were tested directly against your live store and are reproducible on demand.

Where to aim

AI visibility splits into tiers by how fast you can realistically win each one. The useful thing in your case is that you have already written material for the first two.

Problem-aware is the unusual opportunity. Normally this tier is expensive because you have to create the content first. You do not. It exists, it is good, and it is one markup pass away from being usable. Two competitors currently own those answers with ordinary blog articles, and you have a named corrosion scientist.

What to do, in order

Ordered by impact against effort. The first three are the ones we would do this week.

For context

The usual shape of this audit is that a store fails at the front door and the work is a rebuild. Yours is not that. You pass every readability test we have, on every crawler, with identical payloads. That part is genuinely done.

What is left is unusual in a good way. You are not short of citable material. You have a signed lab report, a named scientist, and fifteen answered buyer questions, which is more raw substance than most brands your size can put on the table. It is sitting in a form nothing can use, behind a product setting that hides the thing you sell.

Next step

This audit is a snapshot of one moment. AI search moves quickly: crawlers appear, schema handling shifts, and citations rotate every few weeks. A document can tell you what to fix. It cannot tell you whether the fix worked.

The first three items in Section 11 are worth doing this week regardless of what we agree next, and the first one is worth doing today. If you want, the practical next step is a citation baseline across your target queries before anything changes, so that when the flagship becomes visible and the answers get marked up, the difference is measurable rather than assumed.

The one gap in this audit that is worth closing directly is the assistant coverage in Section 09. We tested one assistant honestly rather than four badly. Widening that is a short piece of work and it would firm up the only findings here that carry a caveat.