We tested 100 B2B SaaS companies in ChatGPT and Perplexity. 53 were invisible.

I run AnswerManiac, a GEO agency focused on AI search for B2B SaaS. We kept hearing that every SaaS brand needed to optimize for ChatGPT, but there was very little public data showing how visible companies actually were.

So we tested 100 B2B SaaS companies across 42 buyer-intent queries in ChatGPT (GPT-4o, browsing off) and Perplexity (Sonar Pro).

Here is what we found:

• 53 companies received zero citations across both engines.
• 47 were cited at least once.
• Only 22 appeared in both ChatGPT and Perplexity.
• 25 appeared in one engine but not the other.
• The most-cited company appeared only 3 times across the entire test.
• Perplexity cited more companies per query on average than ChatGPT.

The surprising part was not simply that half the sample was invisible. It was how thinly recommendations were distributed even among the visible half.

Three practical takeaways

1. AI visibility is engine-specific.

A single combined score hides important differences. If a company appears in Perplexity but not ChatGPT, that should be reported as two separate facts, not averaged into one reassuring number.

2. Traditional marketing spend does not guarantee recommendation visibility.

Several well-funded SaaS companies received no citations. Meanwhile, companies with Wikipedia pages, strong G2 profiles, and analyst coverage were overrepresented among the cited set. That is a correlation, not proof of causation, but it suggests that third-party corroboration matters alongside owned content.

3. Measurement needs a documented baseline.

Before changing content, run a fixed set of buyer-intent prompts, record the engine and configuration, and preserve the exact outputs. Otherwise it is impossible to tell whether GEO work changed anything.

A simple baseline founders can run

1. Write 20 to 40 questions that real buyers ask while comparing products.
2. Test each question separately in the engines relevant to your market.
3. Record whether your company appears, where it appears, and which sources support the answer.
4. Separate results by engine rather than reporting one blended AI visibility score.
5. Repeat the same set on a fixed cadence and log configuration changes.

There are important limits to our study: each query was run once per engine, only two engines were included, and the results are a structured snapshot rather than a statistical estimate. We published those limitations because AI outputs are non-deterministic and a neat percentage can otherwise imply more certainty than the data supports.

The full methodology, findings, and raw-data link are in the State of AI Search: B2B SaaS report.

For founders already checking AI visibility: are ChatGPT and Perplexity giving you similar results, or completely different ones?