What 57 B2B SaaS sites taught us about AI-readiness in 2026

We ran the first 57 B2B SaaS websites through Organicus — a full crawl plus grounded answers from GPT-5.4 and Gemini. It's a small sample and an honest one, but the pattern is already loud: most B2B SaaS sites are close to invisible to AI answer engines by default. Here is what the data says.

The average site is barely half-ready

The mean Organicus Score across the 57 sites was 48.8 out of 100. The typical site had about 15 crawlable pages, averaged roughly 508 words per page, and carried around 82 distinct issues — 4,703 in total across the sample. "Half-ready" isn't a rounding error; it's the median experience.

Almost nobody guides the AI crawlers

This was the most striking finding. Only 3.5% of sites served an llms.txt, and only 10.5% had a robots.txt at all — meaning roughly nine in ten give GPTBot, ClaudeBot, and PerplexityBot no guidance whatsoever about what to read or whether they're welcome. Zero used hreflang, so multilingual sites were routinely mis-read. The cheapest wins in AEO are the ones almost nobody has claimed.

The most common gaps are structural — and mechanical

Ranked by how often they appeared across the sample, the top issues were:

• Missing canonical tags (621 pages) — the single most common issue, leaving AI unsure which URL is authoritative.
• Missing structured data / schema (407) — no machine-readable description of what the page is.
• Poor entity clarity (345) — the site never plainly states what the company is, in a way a model can extract.
• Thin content (284) and missing meta descriptions (280).
• Titles too short (219), missing article schema (217), meta too short (195), no subheadings (166).
• Missing author markup (146), missing H1 (126).

What it means

None of these are creative problems. They're structural, well-understood, and mechanical to fix — which is precisely why they can be done for you rather than becoming a project for a team you don't have. The gap between a 49 and a 75 is rarely a rewrite; it's canonical tags, schema, a clear entity definition, a few comparison pages, and getting listed where AI already looks.

The practical conclusion is that AI readiness starts with clear crawler access and machine-readable site structure. The robots exclusion protocol is a formal internet standard defined in RFC 9309, and OpenAI’s crawler documentation explains how publishers can use robots.txt to manage GPTBot and OAI-SearchBot access separately.

An llms.txt file serves a different purpose: it is an emerging proposal for giving language models a concise, Markdown-formatted guide to a website’s most useful content, rather than an established web standard. The official llms.txt proposal describes the file as a way to help models access important information without navigating an entire site.

Crawler instructions alone are not enough. The Sitemaps protocol lets a site identify canonical URLs and provide crawl-related metadata, while Google’s structured data guidance explains that standardized markup supplies explicit clues about a page’s meaning. Together, accessible pages, clear information architecture, sitemaps, structured data and accurate crawler directives create a stronger technical foundation for both traditional search engines and AI-driven discovery.

Nine in ten B2B SaaS sites give AI crawlers no guidance at all. The cheapest wins in AEO are the ones almost nobody has claimed.