AI search visibility for a software company is whether an assistant names your product when a buyer describes their constraints rather than your category. That question is comparative, and comparisons mostly live on domains you do not own.
A buyer researching with an assistant does not ask what a data warehouse is. They ask which one handles semi-structured events under a two-person data team, on a budget, with SOC 2 already in place. The page that decides shortlist membership for a question like that is usually a roundup, a comparison or a review page maintained by somebody you do not employ. Your own domain gets quoted for something narrower and more specific: confirming price, seat definitions, rate limits, integrations and compliance facts.
So two things have to be true at once. The facts on your domain have to be fetchable and stated in text, and the pages that already answer the comparison have to describe you correctly. A company can own every informational keyword in its category and still be absent from the answer, because the sub-query that decides the shortlist is a comparison and the comparison is hosted elsewhere. Most software teams have done neither half, because neither sits inside the content calendar.
Google documents the mechanism on its own surfaces: its AI features guidance says AI Overviews and AI Mode may use a query fan-out, issuing multiple related searches across subtopics before composing a response. That is Google describing Google. The other engines publish no comparable mechanism, so read it as one engine's documented behaviour rather than a rule of the category.
The site sits behind a WAF, and the WAF offers a one-click rule that blocks AI crawlers as a category. Such a rule treats training crawlers and search crawlers as the same thing. OpenAI's published bot documentation separates GPTBot, which crawls for model training, from OAI-SearchBot, which surfaces sites in ChatGPT's search features, and states that a site opted out of OAI-SearchBot is kept out of the answers that search produces. Legal asked to be excluded from training. Infrastructure ticked a box. Nobody connected the two events.
The most quotable thing a SaaS company owns is its documentation: rate limits, SSO support, data residency, the integration list, the real API shape. Some meaningful part of it usually sits behind auth, behind a demo gate, or inside a docs app that renders client side and arrives as an empty shell to a fetcher on a tight time budget. The marketing site is public and says nothing checkable. The checkable material is private. Whatever an engine cannot fetch, it cannot quote from you.
When pricing reads "contact sales", an engine asked about cost has nothing of yours to use. In our own pricing prompts across five engines, the answer rarely comes back with a gap in it. It carries a number from somewhere else, usually a third-party page of unknown age. A public starting price, a public seat definition, or a published range with its conditions attached gives a model something it can attribute to you. That is a commercial decision, not a technical one, and the cost of it is now visible in a way it was not in 2023.
Google is explicit that no special structured data is required to appear in its AI features. Note what that does not say: markup still resolves identity, and identity decides whether four mentions of your product across four domains are understood as one company. Mark up for resolution. SoftwareApplication carries applicationCategory, offers and featureList; Organization carries the sameAs links that tie your site to your profiles elsewhere. The rich-result payoff is a separate question: Google's changelog entry of 8 May 2026 says the FAQ rich result stopped appearing in Google Search from 7 May 2026.
On evidence, we would rather be plain than flattering. Our clients are D2C brands in hair care and in skin care, which is ecommerce rather than software, so we will not offer them as proof that we know your category. The posts below and the sample audit are the work we would ask you to judge us on, and neither carries a client name or a client result.
Eleven posts, none of them gated, none of them a pitch.
Everything else is in the blog index.
Four of the failures above are visible from the outside in an afternoon. The ten-point self-check is free and asks for no email: it covers crawler access, whether your pricing exists as text, whether your entity resolves to one thing, and what survives with retrieval switched off.
When you want it settled rather than sampled, the audit runs 45 prompts across five engines, with retrieval on and retrieval off, and attaches the evidence to each finding. It is $500 and it lands within 7 working days. The only deliverable is the report. The sample audit shows what that report contains, section by section, before you decide, and the contact form is the way in.
No supplier can promise an engine will recommend a brand, and nothing here does. What an audit gives you is a measurement you can re-run and a list of defects you can check.