What we check, and what it costs you

An owner should be able to read any line of one of our reports and know what it means, why it matters and what to do about it, and this page is the key to all of them. It also says how sure we are, because some of what follows is documented by the platforms, some we've observed in our own tests, and some is inference.

The ladder: how far your customer's assistant can get

The headline of every report is a rung, which is a stop on the line. It answers the question an owner actually has, which is how far a customer's assistant gets with their business when someone asks it for a business like theirs.

Rung 0: Not found
The assistant never mentions you.
24 of 117 Austin practices

Usually a bot wall on the homepage, a noindex tag, search crawlers blocked in robots.txt, or listings that disagree, so the assistant can't confirm that you're one business.
Rung 1: Found, little to recommend on
It knows you exist but picks a business it can say more about.
0 of 117 Austin practices

There's no readable price, hours, policies or specifics (new patients, Saturday, insurance), so it can't match you to what the person asked for.
Rung 2: Recommended, then on your own
It names you and hands the person a phone number or a link.
25 of 117 Austin practices

There's no online booking, only a phone number, or a booking button that opens a script with no page behind it.
Rung 3: Booking starts, a person must finish
It opens your booking page and gets stuck.
68 of 117 Austin practices

It meets a request form instead of open times, a CAPTCHA, a login wall, or a widget it can't operate.
Rung 4: Bookable end to end
The appointment is confirmed and the person is told.
0 of 117 Austin practices

The page shows real availability, nothing blocks the form, and there's a confirmation page at the end.

Search and AI answers: what we track every month

The ladder above is about agents. We measure search and AI answers in the ordinary way, on a schedule and against the businesses you compete with.

Technical health

What we check: Crawl errors, indexing, broken links, redirects, page speed on a phone, structured data, and bot settings that turn machines away.
What we do about it: Fix it on your site, then recheck. On the managed plan this runs every month.

Tracked searches

What we check: Your position for up to 100 searches that bring customers, and which page ranks for each.
What we do about it: Write or update the page that should rank, and fix what holds it back.

Business Profile and listings

SearchAI answers

What we check: Your Google Business Profile, Apple Maps and Yelp: hours, categories, services, photos, and whether they agree with your site.
What we do about it: Keep them current and consistent, and help you answer reviews.

AI answer share

AI answers

What we check: Fifty questions your customers ask, run through ChatGPT, Gemini, Google's AI Overviews and Perplexity: whether you are named, what is said about you, and which sources they cite.
What we do about it: Publish the facts they quote (prices, hours, services, policies) where they can read them, and correct the listings they cite.

Every item, and the so what

Documented means the platform says so in its own documentation. Observed means we watched it happen in our tests and have the receipt. Inference means it follows from how the platforms describe their systems, and we're still measuring it.

Bot wall on the homepage

Blocks rung 0Documented

What it means: Your server answers a plain browser request from a datacenter with a challenge page, a 403, a 503, or an empty page that only fills in after a JavaScript test.
Why it matters, and how sure we are: Documented. A browsing assistant gets nothing and leaves. A search crawler that gets a 403 drops you from that index, and OpenAI states that sites its search crawler can't reach don't appear in ChatGPT search answers.
The fix: Exempt the homepage and the services, pricing and booking pages from the challenge, and keep the protection on everything else. Where you can, use verified-bot allow rules instead of puzzles.
Cost and risk: It costs little, about an hour with whoever runs the site. The risk is more scraper traffic on the exempted pages, which rate limits handle better than walls do.

robots.txt rules for AI bots

Blocks rung 0Documented

What it means: This is which crawlers you allow. Search and retrieval bots (OAI-SearchBot, Bingbot, Claude-SearchBot, PerplexityBot) feed the answers, training bots (GPTBot, ClaudeBot, Google-Extended) feed model training, and user-triggered fetchers (ChatGPT-User, Claude-User) act when a person asks for something.
Why it matters, and how sure we are: Documented by each platform. Blocking a search bot removes you from that assistant's answers, while blocking a training bot changes nothing about search. In the Austin census, no practice blocked a search bot, so this is rarely the problem for a local business.
The fix: Allow the search and retrieval bots and the user-triggered fetchers, and decide about the training bots on their own merits.
Cost and risk: It takes five minutes. There's no risk to search, and the only trade-off is whether you opt in to training.

Listings that agree

Blocks rungs 0 and 1Inference

What it means: Your name, address, phone and hours match across your website, Google Business Profile, Apple Maps, Yelp and social profiles, and your site links to those profiles.
Why it matters, and how sure we are: Inference, backed by how the assistants describe matching a business across sources. When listings disagree, the assistant is less sure that two of them are the same business.
The fix: Fix the mismatches at the source, and link your profiles from the site, listed as sameAs in the structured data.
Cost and risk: It takes an afternoon, and there's no risk.

Business structured data, and what "complete" means

Blocks rung 1Inference

What it means: This is a LocalBusiness (or Dentist, Plumber, etc.) block in your page code. Complete means it carries your address, phone, opening hours, and links to your other profiles. Price range and area served are the next two to add.
Why it matters, and how sure we are: Google says it isn't required for AI features, and we treat it as inference. It is the one part of the page written for machines, and it is how an assistant can answer "open Saturday?" without guessing. Our appearance study is testing whether complete data correlates with being recommended.
The fix: Add the block or complete it. Most platforms generate a partial one, and the usual gaps are hours and sameAs.
Cost and risk: It takes an hour or a plugin setting, and there's no risk.

Prices as text

Blocks rung 1Observed

What it means: At least one price, or a starting price, written on an HTML page and not inside a PDF or an image.
Why it matters, and how sure we are: Observed. When ChatGPT is asked what a visit costs and the site has no price, it says so and moves on. In the Austin census, fewer than half of readable sites had a price in text.
The fix: Publish a starting price for the service people ask about most, and mark it up as an Offer. A range is fine.
Cost and risk: It takes an hour. The risk is that competitors can see the price, although most of them already assume it.

Hours and policies as text

Blocks rung 1Observed

What it means: Your opening hours, and your cancellation, deposit or refund terms, on a page that a machine can read.
Why it matters, and how sure we are: Observed. "Open Saturday" and "same day" are among the most common things people ask for, and an assistant booking on someone's behalf looks for the cancellation terms before it commits.
The fix: Put your hours in the text and in the structured data, and publish your policies on a page linked from the footer.
Cost and risk: It takes an hour, and there's no risk.

Content that needs JavaScript to read

Blocks rungs 0 and 1Documented and inference

What it means: The raw page is nearly empty, and the words only appear after scripts run.
Why it matters, and how sure we are: Documented for crawlers, and inference for agents. Simple fetchers see nothing, and browsing agents see the content but more slowly.
The fix: Render the main content on the server, or at least the key facts.
Cost and risk: It depends on the platform and can be a job for a developer.

The booking path

Blocks rungs 2, 3 and 4Observed

What it means: We follow your "book" or "order" link in a real browser and record what an automated visitor finds there, which could be real availability, a cart, a page hosted by a vendor, an embedded widget, a request form, a login wall, a phone number, a scripted button, or a bot challenge.
Why it matters, and how sure we are: Observed, with a receipt for each attempt. This is the check that decides rungs 2 to 4. In the Austin census, 68 of 117 practices stop at rung 3 and none reach rung 4.
The fix: Show live availability from your booking system and link to it directly, give scripted buttons a real URL, allow guest booking, and move any CAPTCHA to the confirmation step.
Cost and risk: This can be anything from a setting change to a switch of vendor. The risk is spam bookings, which verification at the confirmation step handles better than a gate at the front door.

CAPTCHA on the booking page

Blocks rung 3Documented

What it means: A human-verification puzzle that appears before the form or the times are shown.
Why it matters, and how sure we are: Documented. Assistants don't solve them, and a quarter of Austin booking pages had one.
The fix: Move it to the final confirmation step, or use a verified-agent check instead of a puzzle.
Cost and risk: It takes minutes. The risk is slightly more junk submissions, which get filtered out at confirmation.

Google reviews and profile

Blocks rung 1Documented

What it means: Whether your Google Business Profile exists, what its rating and review count are, and whether its website matches the site we scanned.
Why it matters, and how sure we are: Documented that assistants read reviews. In Austin dental it isn't what separates practices, because every practice had a profile and the average rating was 4.7.
The fix: Keep the profile claimed and current, and respond to reviews. This is usually not where the work is.
Cost and risk: It's ongoing and low in cost.

llms.txt, UCP profile, FAQ markup

Blocks nothing on its ownLow impact

What it means: Optional files and markup written for machines.
Why it matters, and how sure we are: Low impact. Platforms are now generating these automatically, so they don't set you apart, and we have no evidence that they change recommendations.
The fix: They're fine to have, but don't pay for them.
Cost and risk: None.

What we can't tell you yet, and what we do instead

Nobody can see how many people asked an assistant for a business like yours and were sent to a competitor, because the platforms don't publish it. So we don't guess at a number of lost customers, and you should be wary of anyone who does.

What we can measure directly is the appearance rate. We run thirty realistic questions that a customer in your area would ask through ChatGPT and Gemini, and record which businesses come back. The finding then sounds like "you appeared in 2 of 30, and the practices that appeared in 20 or more all publish a price, show Saturday hours, and have a booking page an assistant can read." That describes a pattern across your actual competitors, and it is not a promise. It's the study we're running on the Austin census now.

To see today's volume, look in your own analytics for referrals from chatgpt.com, gemini.google.com and muse.ai. The number may be small today, and tracking it month to month is how it becomes a trend you can act on.

Two things we never do

We don't certify anything. There's no badge, because a badge would imply that a platform agreed with us, and none has. We report what we observed on a given date, and we retest after a fix.

We don't promise placement, because Meta, Google and OpenAI haven't published how their assistants choose between businesses. Everything on this page is about removing the reasons an assistant can't pick you, and whether it then does is up to the assistant and the question it was asked.

See where you stand

All we need is your website and an email address, and a person does the rest.