We checked 120 Austin dental websites the way an AI assistant would. None showed bookable appointment times.
A scan of 120 Austin dental practices on October 1, 2026, with every booking page checked by hand on October 5: 18% blocked a plain browser at the door, a quarter of booking pages had a CAPTCHA, most 'book online' buttons led to a request form, and not one page showed open times to an anonymous visitor. The practices' reviews were excellent, but the plumbing of their websites wasn't.
On October 1, 2026 we ran our scanner against the websites of 120 dental practices in Austin, Texas. The list came from Google Places and is limited to practices with a website. We kept one site per practice group and counted no chain twice. The scanner does what the automated half of our assessment does: it fetches the site the way a visitor from a datacenter would, reads what a machine can read, follows the “book” button in a real browser, and records what it finds. It does not run Muse, Gemini or ChatGPT. Those tests are done by hand, and we’ll publish them separately.
Correction, October 5, 2026. The first version of this post said that ten practices showed bookable appointment times. On October 5 we opened all 36 booking pages that the scanner had tagged as showing times or embedding a booking widget, and we looked at each one. None of them shows appointment times or an open-date calendar to an anonymous visitor. Sixteen have a real scheduler behind a first screen that asks whether you are a new patient, which location you want, or what kind of visit it is. Fifteen are plain request forms, and five we could not classify. The scanner had been counting office-hours text such as “8:00 AM to 5:00 PM” as appointment slots. We have fixed the scanner, corrected the numbers below, and left the original wording visible where it was wrong. Two of the 120 records also turned out to be out of scope (a veterinary clinic and a dental consultancy), so the practice count where it matters is 118.
Below is what the scanner found when it approached the 120 practices as a machine trying to become a patient.
The short version
| What we checked | Result |
|---|---|
| Homepage met a plain browser request with a bot wall | 21 of 120 (18%) |
| Practices blocking any search or AI retrieval bot in robots.txt | 0 of 120 |
| Reachable sites with business structured data | 73 of 99 (74%) |
| Reachable sites where that data is complete (address, phone, hours, profile links) | 18 of 99 (18%) |
| Reachable sites with a price as readable text | 44 of 99 (44%) |
| Reachable sites with neither a price nor opening hours a machine could find | 23 of 99 (23%) |
| Sites with a booking link | 81 of 120 (68%) |
| Booking pages that showed real appointment times to an automated browser | |
| Booking pages with a CAPTCHA in front of the form | 20 of 81 (25%) |
| Practices whose booking page showed real times with no CAPTCHA | |
| Google rating, average | 4.7 stars, median 348 reviews |
The last two rows are the main finding. These are well-reviewed practices with real patients and, in most cases, a website that someone paid for, but not one of them can be booked by an automated visitor without a person stepping in. The closest cases are the sixteen practices whose scheduler widget would show times after the visitor answers a question or two.
Finding: 18% of practices block agents by accident
Not one of the 120 practices disallows a search or AI retrieval crawler in robots.txt, although one blocks a couple of training crawlers. As a matter of policy, then, Austin dentistry is wide open to AI assistants.
On the websites themselves, though, 21 practices turned away a plain browser request at the homepage before it could read a word: six with a Cloudflare challenge page, four with a 403, two with a 503, and nine with a JavaScript challenge that returns an empty page to anything that doesn’t run it. None of those practices chose to hide from Muse or ChatGPT. The blocking comes from a hosting setting or a security plugin. We wrote about why that happens and how to decide on purpose.
Finding: the facts an assistant needs are usually missing
Of the 99 sites that a machine could read, three in four carry some business structured data, which is better than we expected. But only 18 have the complete set, meaning the address, phone number, hours, and links to the practice’s other profiles. Fewer than half publish a price as text and about half publish their hours as text, while a little over a third link to a cancellation or policy page. Twenty-three sites give a machine neither a price nor opening hours to work with.
Those gaps matter because an assistant that is asked “what does a cleaning cost at this practice, and are they open Saturday” has to answer from what it can read. When it can’t find the answer, it either says so or moves on to a practice where it can.
One thing surprised us: 30 of the 99 readable sites already have an llms.txt file, and five publish a Universal Commerce Protocol profile. Nobody at those practices wrote either file, because their website platform generated them. So both are now features that a platform turns on, and having one doesn’t set a practice apart.
Finding: “Book online” mostly doesn’t lead to an online booking
Eighty-one sites have a booking link. We followed each one in a real browser and waited for the page to settle. This is what an automated visitor found:
| Booking page outcome | Count | Share of 81 |
|---|---|---|
| A request form (name, phone, preferred time; no availability shown) | 26 | 32% |
| A widget embedded from another service, contents not readable from outside | 21 | 26% |
| Real appointment times or a live calendar | 15 | 19% |
| Could not classify; needs a manual test | 8 | 10% |
| A phone number only | 6 | 7% |
| A button that opens a script or pop-up with no page behind it | 5 | 6% |
A request form is the modern version of “call us.” A person can fill it in and wait for a callback. An assistant that was asked to book a cleaning for Thursday morning can submit the form, but it can’t confirm anything, so it can’t finish the job. Twenty-five percent of booking pages also put a CAPTCHA in front of the form, including ten of the 26 request forms, which means that the assistant can’t even submit it.
Of the 15 practices that showed real availability, five had a CAPTCHA on the same page. That leaves ten practices in Austin where an automated visitor could see open times and move toward booking without a person involved. Corrected October 5: those 15 were request forms with office hours printed next to them. Of the 36 pages we checked by hand, the scheduler widgets we could identify were Flex (12 practices), All in One Dental (2), NexHealth (1), Subscribili (1) and Weave (1). Every one of them opens on a question (new or returning patient, which location, what kind of visit) before it shows a time. An assistant can start there, but it cannot see availability without answering.
In the embedded-widget group, most of the widgets were payment or form embeds and not booking engines, and in several cases the “book” button leads to a page whose real booking tool loads from a vendor domain that the scanner can see but can’t read into. Those practices may well be bookable. We can’t tell from outside, and that is what the hand-run agent test is for.
Finding: trust isn’t the problem
Every one of the 120 practices has a Google Business Profile. The average rating is 4.7, and the median practice has 348 reviews. Patients like these businesses, and reputation won’t be the reason an assistant passes one over. The reason will be that it couldn’t read the hours or find a price, or that it couldn’t get past the booking page.
What this means if you run a practice
Three fixes account for most of what we found. We’ve listed them in order of how much they change the result:
- Find out whether your site is one of the 21. Load your own homepage from a cloud machine, or ask whoever manages your site whether a bot-fighting mode or JavaScript challenge is on. If it is, exempt the homepage, the services page and the booking page, and keep the protection on everything else.
- Put a price and your hours in text on the page, and in structured data. A starting price for a new-patient visit is enough. Twenty-three of your neighbors have neither, which is something an assistant notices.
- Make the booking page show times. If your booking system can expose live availability, turn that on and link to it directly. If it can only take requests, the honest thing to do is to say so and offer the phone number alongside the form. And move any CAPTCHA to the confirmation step.
Method and limits
The list is 120 dental practices with websites in Austin, Texas, pulled from Google Places on October 1, 2026, one site per practice group. The scanner fetched each homepage twice (once as a current desktop browser, once declaring itself as a bot) and read robots.txt, the sitemap, llms.txt and the UCP well-known path. It parsed the structured data and visible text on the homepage and on the first linked pricing page, then opened the site and the first booking link in a headless Chromium browser from a datacenter address and classified what it saw. Google ratings and review counts come from the Places API.
The scanner did not run Muse, Gemini or ChatGPT. A practice that the scanner marked “request form” may be perfectly bookable through a widget we couldn’t read, and a practice it marked “real availability” may still fail at a later step. The “unknown” and “embedded widget” rows are the cases that the manual agent test is for. No practice is named here, but any practice in the sample can ask us for its own result.
If you’d like to know how your practice looks to an assistant, including the hand-run agent test, the free assessment is the same scan plus a person trying to book with you the way Muse, Gemini and ChatGPT would.
Find out what the assistants see when they look at your business
The assessment is free. A person checks where you stand in search and AI answers, runs the four agent checks (Find, Understand, Transact, Trust) and tries to book or buy, then emails you a one-page report with the three fixes we think matter most.