I Tested 10 Websites for AI Visibility - Here's What I Found

We're in the middle of a quiet but significant shift in how people find information online. ChatGPT, Perplexity, Claude, and Google's AI Overviews now answer questions directly — pulling from websites they can read, understand, and trust. Sites that aren't optimized for AI engines are effectively invisible in this new layer of search.

So I ran an experiment.

I used AEO Tester to audit 10 real websites across five categories — SaaS products, e-commerce platforms, content blogs, major news sites, and a local business. The goal was simple: find out how "AI-ready" the web actually is in 2026.

The short answer? Most sites are not ready. And some of the most well-known brands on the internet are making surprisingly basic mistakes.

Here's everything I found.

The Methodology

Each site was scored against 20 AEO criteria covering technical setup, structured data, content signals, and AI crawler accessibility. The maximum score is 110 points. Each check falls into one of three buckets: passed, warning, or critical failure.

The 10 sites tested were:

I included AEOTester itself as a reference point — not to be self-congratulatory, but to show what a fully optimized site looks like in practice. Every other site was chosen because it represents a category that real website owners belong to.

The Results at a Glance

Average score across all 10 sites: 77.2% (85.0/110 points)

That sounds decent until you look at the distribution:

Only 1 out of 10 sites achieved "Ready for AI" status. Just 1 more came close. The remaining 8 sites — including household names like Notion, Shopify, TechCrunch, and The Verge — have meaningful gaps that could be limiting their visibility in AI-generated answers.

Finding #1: 70% of Sites Are Missing llms.txt

The single most common failure across all 10 sites was missing an llms.txt file.

7 out of 10 sites don't have one. That includes HubSpot, Etsy, Backlinko, Neil Patel, TechCrunch, The Verge, and the local dental clinic.

llms.txt is a simple plain-text file you place at your domain root — similar to robots.txt but designed specifically for AI systems. It gives language models a structured, concise summary of what your site is about, what your key pages are, and how your content should be understood. Think of it as an introduction letter you hand directly to every AI model that visits your site.

It was proposed in late 2024 and has been gaining traction ever since. Yet the majority of the web hasn't adopted it.

The sites that do have it — Shopify, Notion, Backlinko (partially), and AEOTester — are sending a clear, machine-readable signal about their content. The others are leaving AI systems to guess.

Finding #2: Not a Single Site Has FAQ Schema

This one genuinely surprised me.

FAQPage schema is one of the highest-value structured data types for AI visibility. It maps your questions and answers directly into the format that AI engines use to answer user queries. When an AI is deciding whether to pull from your site to answer "how does X work," FAQ schema is a strong signal that your content directly answers questions.

0 out of 9 tested sites (excluding AEOTester) had FAQPage schema.

Not HubSpot. Not TechCrunch. Not Backlinko — which otherwise scored the highest of any real-world site in this test.

Most sites had partial structured data. HubSpot had Organization and a broken Product schema. Shopify had only an Organization schema. Backlinko had Article, BreadcrumbList, and Person schemas. The Verge had 15 Person schemas. But none had FAQ, which typically adds up to 8 points on its own and — more importantly — directly maps to how AI engines surface answers.

Notion had zero structured data at all. A $10B+ company with no JSON-LD on its homepage.

Finding #3: News Sites Are Actively Blocking AI Crawlers

This was the most striking finding of the study.

TechCrunch is blocking 12 AI bots. The Verge is blocking 13. Both explicitly block anthropic-ai (Claude's crawler), ChatGPT-User, Bytespider (ByteDance/TikTok), CCBot (Common Crawl, used to train many models), and Applebot-Extended.

In other words, two of the most-read tech publications on the internet have made a deliberate decision to be invisible to AI search systems.

This is a calculated bet — and it may make sense for publishers who monetize through ad-supported page views. If ChatGPT answers a question using your content, that user never visits your site. No visit means no ad revenue.

But it also means that when someone asks an AI assistant about a tech story, TechCrunch and The Verge are systematically excluded from the answer pool. Newer outlets without these restrictions will increasingly fill that space instead.

Notion blocks just one bot — Amazonbot — which is much less significant and may simply be an oversight.

Skip the manual checking — AEO Tester scans all AI bot permissions in one click.

Finding #4: 60% of Sites Have Image Alt Text Failures

6 out of 10 sites had image alt text issues — either flagged as a critical failure or a warning.

The worst offenders:

• Shopify: Only 25 out of 68 images have alt text (37%)
• Notion: 36 out of 74 images have alt text (49%)
• OmniaDental: 2 out of 8 images have alt text (25%)

Alt text is one of those "unsexy" optimizations that falls through the cracks because it feels like a minor detail. But for AI systems that parse pages visually and contextually, images without descriptions are invisible. A product screenshot, a team photo, a chart — all meaningless to an AI if they're not labeled.

For e-commerce and SaaS sites in particular, this is a significant missed opportunity. Product images are core content, not decoration.

Finding #5: Content Freshness Is a Blind Spot for SaaS Homepages

Half the sites tested had content freshness issues.

Notion, Shopify, and Etsy all received critical failures for having no publish or modified dates anywhere on the page. TechCrunch and The Verge had partial dates (modified only, missing publish date).

From an AI visibility perspective, freshness signals matter because AI engines deprioritize content that can't be verified as current. A page with no date could have been written in 2018 or 2025 — the AI can't tell, and it will weight more confidently-dated content higher.

This is a particular pattern for SaaS homepages. They're written as "evergreen" landing pages and never treated as dated documents. But the lack of date metadata is a silent penalty.

How Industries Compare

Looking at average scores by category (excluding AEOTester as the benchmark):

Blogs outperform everyone else. This makes intuitive sense — sites like Backlinko and NeilPatel were built by SEOs who understand meta tags, structured data, and content signals at a deep level. Their sites reflect that expertise.

E-commerce performs worst. Etsy scored 65% — the only site to receive "Major Gaps" status alongside The Verge — with 6 critical failures including no llms.txt, no sitemap, and no H1 tag on the homepage. For a platform hosting millions of sellers, these are foundational issues.

SaaS is a mixed bag. Shopify scores 80% and nearly matches Backlinko, largely because its global homepage has solid technical fundamentals — all bots allowed, valid schema, llms.txt in place. But Notion sits at 71% with zero structured data, pulling the category average down. The gap between the two shows how much individual team priorities vary even within the same industry.

The Most Surprising Single Finding

The Verge's homepage title is 9 characters long.

Not 9 words. 9 characters.

The AEO audit flagged it as 2 separate issues: too short and missing page-specific context. A 9-character title on a major publication's homepage is almost certainly the result of a template issue or a CMS bug — but it's gone unnoticed long enough to still be there.

It's a good reminder that even high-traffic, well-resourced sites have basic technical issues hiding in plain sight.

What Good Looks Like

For reference, here's what a fully optimized site looks like based on AEOTester's own score (107/110, 97%):

• All AI bots allowed to crawl (126/126)
• Full structured data with FAQPage, Organization, and supporting schemas
• All schemas valid
• llms.txt present
• Sitemap present and referenced in robots.txt
• Content freshness dates on all content
• Alt text on all images
• Proper H1-H6 heading hierarchy
• Open Graph and Twitter Card tags complete
• SSL, language tags, viewport config — all in order

The only warning was an E-E-A-T note about adding more specific author credentials. That's genuinely a 97% job, not 100%.

The gap between that and the average site in this study — 76.4% — is meaningful. It represents dozens of signals that AI engines use to evaluate, trust, and cite content.

Key Takeaways

For any website owner

The single highest-ROI action right now is creating an llms.txt file. It takes 15 minutes, it's free, and 70% of your competitors haven't done it yet.

The second priority is adding FAQPage schema to your most important pages. Zero sites in this study had done it. That's an open field.

For SaaS companies

Your engineers know how to add JSON-LD. Your marketing team probably just hasn't asked. Run an AEO audit on your homepage — the results will surprise you.

For news publishers and bloggers

Your content freshness metadata is almost certainly incomplete. Publish dates and modified dates need to be machine-readable, not just visible to human readers.

For e-commerce sites

Product image alt text is not optional anymore. Every unlabeled image is a product that AI systems can't understand or recommend.

Where Does Your Site Fall?

The gap between an AI-ready site and an average site is not a technical gulf — it's an awareness gap. Most of the fixes are straightforward. Most of the failures are invisible unless you know what to look for.

You can check your own site in about 30 seconds.

This study was conducted in March 2026 using the AEO Tester browser extension. All scores reflect a single-page audit of each site's homepage (or a representative article page for content blogs). Scores may have changed since publication.