Buyers now ask ChatGPT, Perplexity, and Google AI which jeweler to trust, which ring to buy, and where to take a repair. Half of consumers are now asking AI for business recommendations. So we asked a different question: can the AI engines even see the jewelers?
In June 2026 we pulled 706 independent US jewelry store websites from the industry’s own public directories, the American Gem Society locator, Preferred Jewelers International, Jewelers of America, IJO, plus metro-level searches across 25 cities, and measured every site on the technical signals AI engines depend on: crawler access, structured data, FAQ markup, and llms.txt.
The short version: the average independent jeweler is not losing the AI shelf to competitors. They are losing it to defaults nobody ever looked at.
This is part seven of our 2026 fine-jewelry operating series, and it is the one built on our own first-party data. Part three made the case that AEO is a separate channel from SEO. Part five covered the agentic storefronts that sell inside the AI surfaces. This study is the proof, measured across an entire industry.
Of 706 directory-listed jewelers, 542 sites were reachable and 521 served a parseable robots.txt. 17.7% block at least one major AI crawler, and it is the same bundled list every time (GPTBot in 100% of blocklists, Perplexity in almost none), which means it is a copy-pasted platform default, not a decision. 22.1% have no structured data at all. Only 1.5% carry FAQ schema. The blocking is concentrated almost entirely in jewelry-specific website platforms, which ship the block for every store at once. The jeweler never set it and usually does not know robots.txt exists.
IThe headline numbers.
Of 706 directory-listed jewelers, 542 sites were reachable for a full assessment and 521 served a parseable robots.txt. Across those:
- 17.7 percent block at least one major AI crawler, 92 of the 521 sites with a readable robots.txt. And it is the same 92 every time: every jeweler that blocks any AI crawler blocks GPTBot, OpenAI’s crawler. GPTBot sits in 100 percent of the blocklists.
- The blocklists are bundled, and one omission gives the game away. Among those 92 sites, ClaudeBot is blocked by 91 and Google-Extended by 87, but PerplexityBot by only 2. Nobody hand-writes a list that bans ChatGPT, Claude, and Google’s AI crawler yet deliberately welcomes Perplexity. These are copy-pasted platform defaults, not decisions a jeweler made.
- 22.1 percent of reachable sites have no structured data at all on their homepage. Not a misconfigured schema. None.
- Only 1.5 percent have FAQ schema, eight stores out of 542. The other 98.5 percent have none. Question-and-answer markup is the most direct feed into AI answers, and it is effectively unused by an entire industry.
- Product schema appears on 2 percent of homepages. LocalBusiness schema does better, at 51.1 percent, meaning roughly half the industry is at least minimally machine-readable as a business.
- 26.2 percent publish an llms.txt file. That number looks like sophistication. It is not, and the explanation matters (section IV).
- 16.1 percent of reachable sites are fully absent from the AI layer. These are sites that do not block anything but also offer nothing: no structured data, no llms.txt, nothing for an engine to parse beyond raw HTML. Passively invisible, as distinct from the active blockers above.
IIThe blocking is not a choice. It is a platform setting.
The 17.7 percent overall is an average, and the average hides the real story. Split the same sites by the platform they run on and the blocking is almost entirely concentrated in one place:
- Shopify jewelers: about 4 percent block AI crawlers
- Wix: 0 percent
- WordPress: 5 to 6 percent
- Squarespace: 5 to 6 percent
- Jewelry-industry website platforms (the single largest group, about half of all sites): roughly 31 percent block AI crawlers
These rates do not add up to anything. They blend. Roughly half the independent trade runs on website vendors built specifically for jewelers, and about a third of those vendors ship a robots.txt that bans GPTBot, ClaudeBot, and Google-Extended for every store on the platform. Average that 31 percent against the near-zero rates of Shopify, Wix, and WordPress and you land back on the 17.7 percent for the industry as a whole. The blocking is not spread thinly across everyone. It is one category of vendor making the decision for hundreds of stores at once. The jeweler never sees it, and usually does not know robots.txt exists.
The result: a buyer asks ChatGPT for a trustworthy jeweler in that store’s own town, and the engine cannot read the store’s website to recommend it. The competitor on Shopify down the street, about 96 percent likely to be readable, gets the citation instead.
We also found five stores that block AI crawlers in robots.txt while simultaneously publishing an llms.txt file inviting AI agents to read the site. The front door is locked and the welcome mat is out. Nothing says “configured by accident” more clearly.
IIIThe structured-data gap is the bigger opportunity.
Blocking is the dramatic finding. The quiet one is more useful if you own a store.
Half the industry (51.1 percent) carries LocalBusiness schema, which tells an engine the basics: name, address, hours, that you are a real business. That is table stakes, and half the industry does not have it.
Past the basics, the cupboard is bare. Product schema sits on 2 percent of homepages. FAQ schema on 1.5 percent, eight stores in the entire sample. FAQ schema is the most direct path into an AI answer, because it hands the engine clean question-and-answer pairs it can lift verbatim. An entire industry has left it on the table.
This is the inverse of the blocking problem, and better news. Blocking is a vendor decision the jeweler has to escalate to fix. Structured data is something a jeweler (or whoever runs their site) can add directly. The store that ships real LocalBusiness, Product, and FAQ schema is not fighting the whole market for the AI shelf. As the data shows, on FAQ markup specifically, they are competing against 1.5 percent of the field.
IVThe llms.txt surprise.
We expected llms.txt adoption near zero. It came back at 26.2 percent, which would make independent jewelers one of the most AI-forward retail segments in the country.
They are not. Reading the files shows a large share follow identical boilerplate templates (“Agent Instructions” headers with the store name swapped in), generated automatically by website vendors. It is adoption by template, not by intent. Which is still better than nothing, and genuinely useful to the engines, but it means the jeweler with a deliberate, accurate llms.txt still stands out.
It also means the same vendors making one good AI decision (llms.txt) and one bad one (blocking the crawlers) are deciding the AI fate of hundreds of stores at once, in both directions, without the stores knowing.
VWhat this means if you own a jewelry store.
Three checks, ten minutes, no vendor required:
- Open yourdomain.com/robots.txt and look for GPTBot, ClaudeBot, Claude-Web, anthropic-ai, or Google-Extended next to “Disallow: /”. If they are there, your site is invisible to the engines your buyers are asking. Ask your website vendor to remove the blocks, in writing.
- Paste your homepage into Google’s Rich Results Test. If it reports no structured data, you are in the 22.1 percent. LocalBusiness schema is table stakes. FAQ schema, which only eight of 542 stores carry, is the single cheapest visibility advantage left in this industry.
- Ask ChatGPT and Perplexity who the best jeweler in your town is. If you are not in the answer, the first two checks usually explain why.
The window matters. Half the industry is minimally readable, a sixth is invisible, and almost nobody answers questions in markup. A jeweler who fixes crawler access, ships real schema, and publishes citable answers is not competing against the whole market for the AI shelf. They are competing against the 1.5 percent.
This is the same argument we made in the AEO playbook and the agentic storefronts playbook, now measured across 706 real stores instead of asserted. The gap is real, it is industry-wide, and it is mostly invisible to the jewelers sitting inside it.
Straight answers
How many jewelry websites block AI crawlers?
In our June 2026 study of 706 independent US jewelry stores, 17.7 percent of the sites with a readable robots.txt (92 of 521) block at least one major AI crawler. GPTBot, OpenAI’s crawler, appears in 100 percent of those blocklists. The blocking is overwhelmingly concentrated in jewelry-specific website platforms that ship the block as a default for every store on the platform.
Why would a jewelry store block ChatGPT from its website?
Almost none of them chose to. The blocklists are identical across sites (GPTBot, ClaudeBot, and Google-Extended blocked together, Perplexity left out), which is the signature of a copied platform default, not a decision. Most jewelers do not know robots.txt exists, let alone that their vendor ships one that bans the AI engines their buyers are asking for recommendations.
How do I know if my jewelry website is visible to AI search?
Three checks. Open yourdomain.com/robots.txt and look for GPTBot, ClaudeBot, or Google-Extended next to Disallow. Paste your homepage into Google’s Rich Results Test to see whether you have any structured data. And ask ChatGPT and Perplexity who the best jeweler in your town is. If you are not in the answer, the first two checks usually explain why.
What structured data should a jewelry website have?
At minimum, LocalBusiness schema (name, address, hours), which about half the industry has. Then Product schema on product pages and FAQ schema on pages that answer buyer questions. FAQ schema is the highest-leverage gap: only 1.5 percent of the jewelers we studied have it, and it is the most direct feed into AI answers.
Does llms.txt help a jewelry store show up in AI search?
It can, but most of the llms.txt files we found are vendor-generated boilerplate rather than deliberate, accurate descriptions of the store. A real llms.txt that genuinely describes your business and catalog is more useful, and because most adoption is template-driven, a deliberate one still stands out.
How was this study conducted?
We assessed 706 unique independent US jewelry store domains collected in June 2026 from the American Gem Society locator, Preferred Jewelers International, Jewelers of America, IJO, and “best jeweler” searches across 25 metros, excluding chains, marketplaces, and wholesalers. Each site received three requests: robots.txt, the homepage, and /llms.txt. 542 were reachable; sites behind aggressive bot protection are recorded as unreachable rather than guessed at. Full methodology is in the section below.
VIMethodology.
Sample: 706 unique independent US jewelry store domains collected June 2026 from the American Gem Society public locator, Preferred Jewelers International member list, Jewelers of America directory, IJO member searches, and “best jeweler” searches across 25 US metros. National chains, marketplaces, wholesalers, and appraisal-only businesses were excluded. 542 sites were reachable for full assessment; sites behind aggressive bot protection are recorded as unreachable rather than guessed at.
Each site received three requests: robots.txt, the homepage, and /llms.txt. “Blocking” means the crawler’s user-agent group (or the wildcard group covering it) carries Disallow: / without an overriding Allow. Schema detection covers JSON-LD on the homepage only, so product schema living solely on product pages is not counted, a deliberately conservative choice. “Fully absent” means a reachable site that blocks nothing yet carries no structured data and no llms.txt: passively invisible. Platform detection uses markup fingerprints (Shopify CDN, wp-content, and so on); the jewelry-industry-platform bucket is everything not matching a mass-market platform, validated by manual sampling. Per-platform blocking rates are reported as ranges where rounding is close. Raw per-domain results: available on request.
Study by Endico Data Strategic, a New York City marketing studio. The founder is on every account. We do not work with two jewelers in the same trade area.
If you want to know exactly where your store stands on every signal in this study, that is what our AEO audit measures. Request an AEO audit or email info@endicodatastrategic.com directly. The founder reads every inquiry within one business day.
