We measured 80 Naples businesses. About one in four cannot be read by AI search..
Buyers increasingly ask an AI assistant who to call before they ask Google. That only works if the assistant can actually read your website. So we measured a defined population of independent Naples businesses and counted how many can be read at all. The method, the population, and every denominator are published below so you can reproduce it.
★ 5.0 ON GOOGLE · NAPLES, FL · NO LONG-TERM CONTRACTS
“Within two weeks my business was ranked #1 organically and top 3 in the map pack.”
The Naples AI Visibility Study is original research by SEO Elite Agency measuring whether independent Naples, Florida businesses can be reached and read by AI search crawlers. On July 24, 2026 we measured 80 businesses drawn from a public OpenStreetMap query and reached 69 of them. Roughly one in four of the businesses we reached were either blocked to AI crawlers or served content those crawlers cannot read (SEO Elite Agency, 2026). You can run the same check on your own site with our free AI visibility checker.
What did the study find?
Of 69 Naples businesses successfully measured on July 24, 2026, twelve served content that depends on JavaScript and four blocked AI crawlers outright, so 16 of 69, about one in four, were effectively unreadable or off-limits to AI search engines (SEO Elite Agency, 2026). Not one of the 69 published a markdown twin, and 24.6% of the 65 we could determine published an llms.txt file (SEO Elite Agency, 2026).
The headline number is the rendering one. Twelve of 69 businesses, 17.4%, publish their content in a way that requires JavaScript to assemble (SEO Elite Agency, 2026). AI crawlers fetch raw HTML and do not execute JavaScript (Vercel, 2024), so for those twelve businesses an assistant asked to summarize or recommend them receives a nearly empty page. The site looks perfect to a human and blank to the machine deciding whether to name it.
A smaller group closes the door explicitly. Four of the 66 businesses whose robots.txt we could read, 6.1%, block at least one major AI crawler (SEO Elite Agency, 2026). We make no claim about whether those blocks were deliberate. In our own experience they usually are not: a security product or hosting default writes the rule and nobody reviews it, which is exactly what happened on our own domain in July 2026 and which we documented rather than hid.
The machine-readable layer is close to empty across the board. Zero of 69 businesses published a per-page markdown twin, and 16 of the 65 we could determine, 24.6%, published an llms.txt file (SEO Elite Agency, 2026). Structured data is more common but often incomplete: 65.2% carry schema of some kind while only 37.7% carry a connected graph an engine can resolve to one specific business (SEO Elite Agency, 2026).
How was the population defined?
From OpenStreetMap, not from a ranking. We queried the public Overpass API for every named feature carrying a website tag inside a Naples bounding box, which returned 500 records, then removed social and aggregator hosts, national chains, government and institutional sites, and store-locator subdomains. That left 395 independent local business domains, of which the first 80 were measured.
Population choice is where most marketing research quietly cheats. Sampling "the top 20 results" measures the businesses that already rank, which guarantees a flattering answer and tells you nothing about the market. We wanted a population defined independently of search performance, so we used OpenStreetMap, an open dataset anyone can query without a key or an account.
The exact query is reproducible: the Overpass API, the bounding box 26.10 to 26.30 north and -81.82 to -81.68 west, and every node, way or relation carrying both a name and a website tag. That returned 500 records. We then excluded social and aggregator hosts, recognizable national chains, government and institutional domains, and store-locator subdomains, because the question is whether independent local businesses are visible, and a national chain with a corporate web team is a different animal. That left 395 domains, and we measured the first 80.
We include the exclusions in the open because they are judgment calls that change the answer. Anyone who disagrees with a call can re-run the query and apply their own filter. That is the difference between research and marketing: the reader can check the work rather than trust the conclusion.
How was each site measured?
One pass per domain against public artifacts only: the homepage HTML, robots.txt parsed per named AI crawler, llms.txt and llms-full.txt, and the structured data on the page. No logins, no private data, no paid tools. Eleven of the 80 could not be measured and are reported as unreached rather than counted as failures.
For each domain we fetched the homepage and inspected the raw HTML for how much content was present before any script ran, which is what an AI crawler sees. We fetched and parsed robots.txt and evaluated it against each named AI crawler rather than assuming a single verdict. We fetched llms.txt and llms-full.txt, and we read the page for a declared markdown twin and for structured data.
Eleven of the 80 domains could not be measured: the site was unreachable, timed out, or served a bot-challenge screen instead of its content. Those eleven are reported as unreached and excluded from every denominator. That matters more than it sounds, because the lazy version of this study counts an unreachable site as a failing site and inflates the alarming numbers.
Each metric therefore carries its own denominator. Crawler findings are computed only across the 66 sites whose robots.txt we could actually read, because a verdict about blocking is worthless if the file itself was unreadable. The llms.txt figure is computed across 65, because for three sites the response was ambiguous. Percentages in this study never quietly borrow a bigger denominator to look more dramatic.
What does JavaScript dependence actually cost a business?
It removes you from the answer, not just from the ranking. A JavaScript-dependent page can rank acceptably in classic Google, which renders JavaScript, while being effectively blank to the AI crawlers that do not. The business sees normal search traffic and never learns why assistants recommend competitors instead.
This is the quietest failure in the study because nothing looks broken. The site loads beautifully for customers, Google can usually render it, and analytics look ordinary. The gap only appears in a channel most owners are not measuring, which is why it persists for years.
The fix is usually not a rebuild. Most modern frameworks can render pages on the server or generate them statically, which puts the content in the raw HTML where every crawler can read it. What matters is that the words a buyer would want are present in the initial response, not assembled afterward in the browser.
If you want to know which side of this line your site is on, our free AI visibility checker runs the same rendering and crawler checks used in this study and reports what a machine actually receives. The deeper version, including what to do about it, is our AI crawlability work and the broader GEO service.
What are the limits of this study?
Three worth stating plainly. The sample is 80 businesses from one bounding box, so it describes Naples and not the country. Measurement came from a single vantage point on one date, and some sites serve different content to different networks. And OpenStreetMap coverage skews toward businesses that someone bothered to map.
Sample size first. Eighty measured businesses from a 395-domain population is enough to describe a pattern and not enough to publish a precise national figure. We report it as what it is: a snapshot of one market on one day. We intend to re-run it, which is why the method is written down rather than improvised.
Vantage second, and this one is genuinely interesting. Websites do not always serve the same thing to everyone. Measuring the same domains from a datacenter network rather than a residential one produced different readings, because some sites challenge or degrade responses to server traffic. This study was run from a residential vantage on July 24, 2026, and both facts belong next to the numbers.
Coverage third. OpenStreetMap includes the businesses that contributors have mapped, which is not a perfect census. It skews toward established storefronts and away from home-based and newer businesses. It is still a far better population than a ranking-derived list, because its bias is documented and independent of the thing we are measuring.
How can you check your own business?
Run the same checks yourself in about a minute. Open your robots.txt and look for AI crawler names beside Disallow rules, view your page with JavaScript disabled to see what a crawler receives, and check whether your key content is in the raw HTML. Our free checker automates all of it, and the audit that follows is free too.
The manual version needs no tools. Open your domain followed by /robots.txt and read it. Then load your homepage with JavaScript turned off in your browser settings, or view the page source and search for a sentence you can see on screen. If the sentence is not in the source, an AI crawler cannot read it either.
The automated version is our free AI visibility checker, which runs the fourteen-crawler check, the rendering check and the machine-layer check used in this study and returns the result in seconds with no email required. If you would rather have the whole picture, including rankings, competitors and technical health, the free audit covers it.
We are a Naples firm at 1950 Mayfair Street, Suite 313, we hold a 5.0 rating on Google, and we work month-to-month with no long-term contracts at published pricing. If this study describes your website, the finding is good news: rendering and crawler access are among the cheapest problems in search to fix. Call (843) 955-7727 or write to hello@seoeliteagency.com.
LAST UPDATED 2026-07-24 · WRITTEN BY JAMIE KLONCZ, FOUNDER · SEO ELITE AGENCY, NAPLES FL
