We Scanned 410 Local Business Websites for AI Visibility. Most Have the Same Blind Spots.

In June 2026 we ran an automated AI-visibility scan across 440 local service business websites in our prospect research set: HVAC contractors, law firms, and med spas, most of them in Texas metros. The scan checks the same things an AI engine has to resolve before it can confidently recommend a business: structured data, crawler access, answer-shaped content, contact consistency, and machine-readable site summaries.

Here is what came back.

The headline numbers

30 of the 440 sites did not load at all. Roughly 7 percent of working local businesses have a website that is down, broken, or unreachable to a standard crawler. Before AI visibility, before SEO, the front door is shut.
The 410 reachable sites averaged 71 out of 100. A passing grade with the easy points, but the misses cluster in exactly the places AI engines care about most.
51 sites scored below 55. For these businesses, an AI assistant asked “who should I call” has very little to verify and very little to quote. They are effectively invisible to the fastest-growing way customers pick providers.
138 sites scored 80 or above. Strong fundamentals are common too, which is the competitive point: in most metros at least one of your competitors is already easy for AI engines to recommend.

17 sites are actively blocking the AI engines

The most surprising result: 17 of 410 sites have robots.txt rules that block GPTBot, ClaudeBot, PerplexityBot, or Google-Extended, sometimes all of them. Several were law firms and med spas, businesses whose clients heavily ask AI assistants for recommendations.

Most of these blocks look accidental: a wholesale disallow rule, a security plugin’s default, or a developer copying a template. The effect is not accidental. An engine that cannot read your site cannot cite it. Those businesses have opted out of AI recommendations without knowing it.

If you check one thing on your site this week, check robots.txt.

The most common gap, by far: no question-and-answer content

386 of the 410 scanned sites showed missing FAQ or question-shaped content among their biggest gaps. This is the single highest-leverage fix in GEO, because AI engines answer questions, and they prefer to assemble answers from pages that already ask and answer them cleanly.

The fix is not a widget. It is a section on your key pages that asks the questions your customers actually ask, in their words, and answers them directly in two to four sentences.

The vertical breakdown

HVAC (31 sites): average 75. The strongest vertical, likely because HVAC marketing agencies have standardized decent schema. FAQ content still missing almost everywhere.
Law firms (179 sites): average 73. 7 of the 17 AI-crawler blocks were law firms. Legal sites also had the most structured-data gaps in absolute terms: 88 of 179 flagged for missing or generic schema.
Med spas (101 sites): average 68. The weakest vertical. Thin content, missing schema, and 5 active AI-crawler blocks.

What was almost universally missing

llms.txt: at least 245 of 410 sites have no machine-readable summary file. It is an emerging standard, so absence is normal, which is precisely why having one is still an edge.
LocalBusiness schema: 183 sites flagged for missing or generic structured data. This is decade-old technology that AI engines now lean on to verify identity before recommending anyone.

Methodology, honestly

The scan fetches each site’s homepage, robots.txt, and llms.txt, and scores ten weighted checks: HTTPS, title and meta description, heading structure, structured data, FAQ content, llms.txt presence, AI crawler access, content depth, visible contact information, and answer-shaped formatting. It is the same engine behind our free GEO Score tool. It is a fundamentals scan, not a full visibility audit: it measures whether AI engines CAN read and verify you, not yet whether they DO recommend you. The full audit tests the second question with live buyer queries across ChatGPT, Perplexity, and Gemini.

Score your own site

The same scan, free, takes 30 seconds: showupseo.com/tools/geo-score. If you score under 70, the gaps it names are your starting list. If you score over 85 and still do not show up when you ask ChatGPT who to call in your city, that is the audit conversation.