We Ran an LLM SEO Check on 188 SaaS Homepages. Here's What We Found.

There is a lot of confident advice about "optimizing for AI search" and very little data on what companies actually do. So we measured it.

On October 16, 2026, we ran our LLM SEO checker against the homepages of 195 well-known SaaS companies - developer tools, productivity apps, marketing and sales software, support desks, analytics, finance, AI, and HR platforms. Seven blocked our fetcher, which left 188 homepages.

The checker looks at a site the way most AI crawlers do: it reads the server-rendered HTML without running JavaScript, then fetches robots.txt, llms.txt, llm.txt, and ai.txt. Every number below comes from that run.

---

The short version

Overall, 76 sites (40%) scored Good, 5 (3%) scored Ready, and 107 (57%) landed in Needs work - meaning they failed at least one critical check.

---

Finding 1: llms.txt went mainstream - but a chunk of them are broken

llms.txt is a plain markdown file at the root of a site that gives language models a short summary and a curated list of links. A year ago it was a niche experiment. In our sample, 120 sites (64%) served something at /llms.txt.

Not all of those are usable. We counted a file as usable only if it had an H1 title and at least one markdown link, which is the minimum the format expects. 17 sites served a file that failed that bar - no title line, no links, or something that was clearly not the intended format. That leaves 103 sites (55%) with a usable llms.txt.

Among the usable files:

• The median file listed 64 links.
• 93 of 103 included a short blockquote summary under the title, the part a model is most likely to quote.

Adoption varied a lot by category:

Developer tools are ahead for an obvious reason: their documentation already lives in markdown, and coding assistants are a real acquisition channel for them.

The companion files are still rare. Only 12 sites (6%) had an llm.txt and 14 (7%) had an ai.txt. Both are optional; we would not prioritize them over getting llms.txt right.

What to do: if you have an llms.txt, open it in a browser and check that the first line is # Your Product Name, there is a one-paragraph > summary, and every link is a full URL in [title](url) form. If you do not have one, our llms.txt and AI crawlers guide walks through a minimal version.

---

Finding 2: Almost nobody blocks AI search - the blocks that exist target training crawlers

We checked each robots.txt against 15 AI crawlers. 172 sites (91%) blocked none of them.

Of the 16 sites that blocked at least one, most were blocking crawlers associated with collecting training data rather than answering live questions. The ten most-blocked:

Only 5 sites (3%) fully blocked a crawler we classify as high impact - the ones tied to the major assistants and AI search products.

Two more patterns worth knowing:

• Path-level restrictions are common and usually fine. Many sites allowed AI crawlers but disallowed specific paths like contact-sales forms or newsletter archives. That is a reasonable way to keep low-value pages out without disappearing from AI answers.
• Not every AI bot does the same job. Some vendors run separate crawlers for model training and for live search or browsing. Blocking one does not necessarily block the others, so check each vendor's documentation before copying someone else's robots.txt.

What to do: read your own robots.txt with fresh eyes. Many sites picked up AI blocks from a template or a CDN default and never decided to block anything. If you want to appear in AI answers, make sure you are not blocking the search and assistant crawlers by accident.

---

Finding 3: A quarter of homepages give AI crawlers no structured data at all

This was the biggest gap. 48 homepages (26%) had no JSON-LD in the server-rendered HTML.

Some of those sites may add structured data with JavaScript after the page loads. That helps Google, which renders JavaScript, but many AI crawlers read only the initial HTML. If your schema arrives late, a lot of AI systems never see it.

Among the sites that did ship JSON-LD, the most common types were:

The pattern: companies describe who they are (Organization) far more often than what they sell. Only 55 sites (29%) described the product itself with SoftwareApplication, Product, or WebApplication markup. That is the type that carries a product name, category, operating system, and pricing - exactly what a model needs to recommend you for a "best tool for X" question.

What to do: view your homepage's source (not the inspector - the raw source) and search for application/ld+json. If nothing is there, add Organization and SoftwareApplication JSON-LD to the server-rendered page. If it only appears in the inspector, it is being injected by JavaScript.

---

Finding 4: The basics of page structure are mostly fine

The fundamentals held up better than we expected:

• 80% of homepages had exactly one H1. 10% had none, and 10% had more than one.
• 76% used a <main> landmark, which helps parsers separate content from navigation.
• 90% listed their sitemap in robots.txt.
• Only 3 homepages (2%) carried a noindex directive, and none used an AI opt-out meta tag.

A homepage with no H1 is a small thing to fix and an easy one to miss, because designers often put the headline in a styled div.

---

What the best-prepared sites had in common

Only 5 sites passed every critical and recommended check. Looking at the 81 sites that scored Good or Ready, three habits stood out:

1. A real llms.txt. Every site in those two tiers had a usable one. (In fairness, our checker treats llms.txt as a critical check, so this is partly how the scoring works - but the file is also the cheapest item on this list to fix.)
2. Structured data in the HTML, not added later by JavaScript, including a description of the product, not just the company.
3. A deliberate robots.txt. Either no AI blocks, or narrow path-level rules instead of site-wide bans.

None of these requires a rebuild. For most SaaS teams, all three fit in a single sprint.

---

Run the same check on your site

The checker used for this study is free and needs no account: GEO and LLM SEO analyzer. Paste your homepage URL and you will get the same pass, warn, and fail checks within a few seconds.

If you are working through the results, these go deeper:

• LLM SEO analysis guide - how to read each check and what to fix first
• How to get cited by ChatGPT without paying for GEO
• Best GEO tools in 2026 - if you want ongoing citation tracking rather than a one-off check
• GEO for SaaS: what it is and how to measure it - the plain-English explainer

---

Methodology and limits

• Sample: 195 homepages of well-known SaaS companies across nine categories, chosen to cover a broad spread of company sizes and markets. Seven sites returned fetch errors and were excluded, leaving 188. This is not a random sample of all SaaS companies; it skews toward established brands.
• Date: all checks ran on October 16, 2026. Websites change constantly, so any individual result may already be out of date.
• What we fetched: the homepage's server-rendered HTML (no JavaScript execution), plus /robots.txt, /llms.txt, /llm.txt, and /ai.txt on the same host.
• Crawlers checked: 15 AI user agents. "Blocked" means a site-wide disallow for that agent; path-level restrictions were counted separately and not treated as blocks.
• Usable llms.txt: an H1 title line and at least one markdown link.
• Scoring: the same rules as the public tool. "Needs work" means at least one critical check failed; "Ready" means every critical and recommended check passed.
• Homepage only: we did not crawl pricing, docs, or blog pages, where structured data and content often look very different.

We are not naming individual companies in this write-up. The goal is a picture of the category, not a leaderboard - and you can check any site yourself with the free tool.