Fully open, still invisible

AI crawler access vs AI mentions across 94 B2B SaaS brands. September 2026, our own data, every number with its denominator.

The question we tested

Most advice about AI visibility starts with crawler access: open your site to GPTBot, OAI-SearchBot, ClaudeBot and PerplexityBot, and AI assistants will start recommending you. We sell the fix for blocked crawlers, so we wanted to know what access is actually worth. We took 94 B2B SaaS brands across five software categories, measured how readable each site is to AI crawlers with our own scanner, then asked ChatGPT, Claude and Perplexity the questions a real buyer would ask in each category and recorded which brands the answers actually named.

Finding 1: almost everyone is already readable

84 of the 94 sites could be scanned and scored. The median AI Access Score is 91 of 100, and 54 of the 84 score 90 or higher, meaning every live-answer crawler we test can fetch their pages and the pages are structurally readable. In this population, blocked AI crawlers are the exception, not the rule.

Finding 2: being readable does not get you named

Of the 54 sites that AI can fully read, 33 were never named by ChatGPT in any captured answer on their category's ten buyer questions. Ranking all 84 scored brands by AI Access Score and by how often each assistant named them produces no relationship at all: Spearman rank correlation -0.06 for ChatGPT, -0.03 for Claude and 0.01 for Perplexity. Zero means the two rankings are unrelated: a 95 site is no more likely to be named than a 75 site.

35 of the 94 brands were named by none of those three assistants. Joining a Gemini and Google AI Overviews tracking window over the same days, 30 of the 94 appeared on none of the five platforms, including brands with thousands of employees and fully open, fast, structured websites.

Finding 3: what does predict being named is reputation, not infrastructure

On the same 84 scored brands, the rank correlation between G2 review count and how often ChatGPT names a brand is 0.51. The assistants are not rewarding technically perfect websites; they are echoing the brands the web already talks about. That matches how the answers are built: the assistants read comparison posts, review sites and community threads, and name whoever those sources name.

Finding 4: the three assistants barely read the same web

Across the 50 buyer questions, the three assistants consulted 940 distinct web domains. 71.6% of those domains were consulted by only one of the three assistants, and 8.3% by all three. Averaged per question, the source overlap (Jaccard) between any two assistants is 0.072 to 0.098 depending on the pair, and across all three at once it is 0.022. Visibility on one platform says very little about the other two. One more number people assume wrong: ChatGPT fetched Reddit on 27 of the 50 questions but cited it in only 1 answer, and fetched YouTube on 21 of 50 and cited it in 0.

What this means if you run a SaaS website

Crawler access is a pass/fail gate, not a ranking. If AI crawlers cannot read your site, you are ineligible: nothing you publish can be quoted, so fixing access is the first move and it is usually under an hour of work. Once you are readable, more access buys nothing further. The work moves to the layer the assistants actually rank on: being named by the comparison pages, review platforms and communities they read. Readable is the entry ticket. Recommended is the game.

You can test the first half of that sentence on your own site right now: run the free AI Crawler Access Report. It takes about 15 seconds and shows exactly which AI crawlers can and cannot read you, with a verification command for every claim. Whether AI actually recommends you against your competitors is what the AI Search Audit measures.

Method

• Brands: 94 B2B SaaS brands across five software categories, deduplicated from a 100-row category list spanning enterprise leaders, challengers and small vendors.
• Mentions: ten buyer questions per category, 50 in total, asked once each on ChatGPT with web search, Claude and Perplexity between September 5 and 9, 2026, on neutral accounts from a US IP. A brand counts as named when its name appears in the visible answer text (whole-word match, aliases included). Gemini and Google AI Overviews coverage comes from a third-party tracking window over September 5 to 7, 2026.
• Access scores: measured by this site's own scanner (21 automated checks across 15 named AI crawlers) on September 26, 2026; 84 of the 94 sites were reachable and scorable. Scores were re-measured after a September 26 scanner correction that stopped counting identity-verifying firewalls as blocks, and the findings on this page hold on both the early-September scores and the corrected ones.
• Honesty notes: mention capture and the final score measurement are three weeks apart. Everything is a single run per assistant, so treat any individual brand's zero as a snapshot; the aggregate rates are the finding. We are an AI visibility agency, so we would love this data to say that technical access drives mentions. It does not, and we publish it anyway.
• The per-brand data behind every number on this page is available on request.