I Audited an AI Automation Agency.
AI Systems Cannot See Them.
By Lesli Rose · June 18, 2026 · 6 min read
This agency had built something real. Over 150 clients across more than eight industries. Three proprietary SaaS products it built and operates itself: an AI interviewer, a workflow tool, and an AI-powered CRM. A verified WhatsApp Business API partnership. Thousands of AI interviews delivered, with case studies carrying hard numbers. Its entire business is building intelligent systems that help other companies scale.
So I did the one check that matters most for a company like this. I asked AI tools who builds custom AI recruitment systems in their market. Competitors came back by name. This agency did not appear at all.
Then I looked at why, and the answer was sitting in the page's first few lines of HTML.
Finding 1: The Portfolio Lives on a Free Subdomain
The portfolio was deployed to a free platform subdomain, the kind that ends in the hosting provider's own domain. That host is fast and modern, so this is not a speed problem. It is an authority problem. A free subdomain carries almost no ranking weight and reads to search engines as a demo, not a production asset. Every signal the page earns accrues to the platform, not the brand. The real company domain existed separately; the showcase of its best work was parked somewhere search engines barely count.
Finding 2: Zero Structured Data, Zero Metadata
The entire descriptive layer of the page was a title tag. That was it.
<title> ... </title> <meta charset="UTF-8"> <meta name="viewport" content="..."> // no meta description // no Open Graph / Twitter tags // no canonical // 0 of 8 recommended JSON-LD schema types
No meta description, so search engines invent a snippet from random page text. No Open Graph tags, so every time the link is pasted into LinkedIn or WhatsApp it renders as a bare URL with no image and no headline. And no structured data at all: no Organization schema, no Person schema for the founder, no SoftwareApplication schema for the three products. A model scanning the page gets prose and headings, but nothing machine-readable to anchor an understanding of who this company is or what it sells.
Finding 3: robots.txt, sitemap.xml, and llms.txt All Return 404
There was no sitemap inviting indexing, no robots.txt stating a crawl policy, and no llms.txt giving AI systems a summary of the entity. For most businesses these are gaps. For a company whose product is AI, the missing llms.txt is the one that stings: it is the file that tells ChatGPT, Claude, and Perplexity what the business is and how to represent it. Two hours of work, and it wasn't there.
Finding 4: Absent From Every Roundup AI Already Cites
According to AirOps analysis of 21,311 brand mentions across GPT-5, Claude, and Perplexity, 85 percent of brand mentions in AI search come from third-party sources, and nearly 90 percent of those come from listicles, comparisons, and reviews. When I pulled the "best software in their category" roundups for their market, their direct competitors appeared in several. This agency appeared in none. No reviews on the major software review platforms, either. The pages AI reaches for when recommending a vendor simply did not know the company existed.
The Scores
45
Technical SEO
0
Schema
35
AI Discoverability
5
Earned Visibility
What Is Actually Working
The proof is all there, and it is genuine. Real case studies with specific outcomes: zero material leakage on a manufacturing ERP, a quotation engine spanning thousands of products, a hospital chatbot handling round-the-clock patient queries. Three owned products, not white-labeled tools. A clean, well-designed page that loads fast on desktop and adapts to mobile. This is not a credibility problem or a talent problem. The business is fundamentally strong.
The gap is legibility. Everything that makes this company worth recommending is written for human eyes and invisible to the machines that now answer "who should I hire for this?" The fixes are days of work, not months: a real domain, a structured-data layer, the three crawl-and-AI files, and a push to get listed where competitors already are.
The Lesson for Anyone Selling Technology
The irony is sharp, but the pattern is common. The businesses most capable of fixing their AI visibility are often the last to do it on their own site, because they are busy doing it for clients. If your product is intelligence, the fastest credibility move you have is to be legible to the systems your buyers now ask for recommendations. Be the proof, not just the pitch.
Does This Look Like Your Business?
The companies best at building technology are often the worst at making themselves findable. I run a visibility report that shows exactly what AI systems see when they crawl your site.
Run Your Visibility ReportFAQ
Why does hosting a site on a free subdomain hurt visibility?
A free deployment subdomain carries almost no domain authority and reads to search engines as a demo rather than a production site. Any signals the page earns accrue to the platform, not your brand. Moving the page onto your real domain consolidates that authority where it can compound.
What structured data does an AI company need on its own site?
At minimum: Organization schema with sameAs links, Person schema for the founder, and SoftwareApplication schema for each product. Service schema describes custom offerings, and FAQPage schema makes buyer-question content citable by AI. Structured data is how a machine recognizes a specific company rather than just reading text.
Why are third-party listicles important for AI recommendations?
AirOps found that 85 percent of brand mentions in AI search come from third-party sources, and nearly 90 percent of those from listicles, comparisons, and reviews. When a buyer asks AI for the best provider in a category, the model pulls from the roundup pages it trusts. A business absent from those pages is absent from the recommendation.
