Blog · By industry

AEO for B2B SaaS

Published 1 May 2026 · Updated 12 October 2026 · 9 min read

AEO for B2B SaaS
Photograph: Arlington Research / unsplash

The short answer: AEO for B2B SaaS is mostly not about your marketing site. Buyers ask comparative questions, engines answer them from third-party roundups and review pages, and your own domain gets quoted for a narrower job: confirming price, limits, integrations and compliance facts. So the work splits in two. Make the facts on your domain machine-readable and reachable, then get the pages that already rank for comparisons to describe you correctly.

That split is the thing most software teams get wrong, and it costs them twice.

The question a buyer asks is not the question you optimised for

Software marketing is built around informational content. 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.

An engine answering that does not necessarily read one page. Google documents the mechanism for 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 and data sources before composing a response. That is Google's description of Google. We have no comparable published mechanism for the other engines, so read it as one engine's documented behaviour, not as a law of the category. Your page competes for one of those sub-queries, not for the whole question. 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 comparisons live on somebody else's domain.

This is the part of the work that sits outside marketing's direct control, and it is the reason who owns AEO inside a company is a real question rather than an org-chart detail. The pages that decide your shortlist membership are maintained by analysts, review platforms and independent writers you do not employ.

Three failure modes we see almost only in SaaS

The managed bot rule that took you out of ChatGPT. This is the one we hit most often in audits of software companies: 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. They are not, at least for one vendor we can cite. OpenAI's published bot documentation separates GPTBot/1.4, which crawls for model training, from OAI-SearchBot/1.4, which surfaces sites in ChatGPT's search features, and states that sites opted out of OAI-SearchBot are kept out of the answers ChatGPT's search produces, while remaining eligible to show up as plain navigational links. That is OpenAI's documentation of OpenAI's crawlers. Other engines document their own fetchers differently, and some do not document them at all. Legal asked to be excluded from training. Infrastructure ticked a box. Marketing lost a surface and nobody connected the two events for a quarter.

The substance is behind a login. The most quotable material a SaaS company owns is its documentation: rate limits, SSO support, data residency, the integration list, the actual API shape. Repeatedly in the audits we run, some meaningful part of that 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. We have not counted how often across a defined sample, so take that as a pattern we keep meeting rather than a rate. The marketing site is public and says nothing checkable. The checkable material is private. Whatever an engine cannot fetch, it cannot quote from you.

Pricing that cannot be stated. When pricing reads "contact sales", an engine asked about cost has nothing from you to use. In the pricing prompts we run during audits, 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. That is what we see in our own prompt runs across five engines, not a claim about how every system behaves. A public starting price, a public seat definition, or even a published range with the conditions attached gives the model something it can attribute to you. This is a commercial decision, not a technical one, but the cost of the decision is now visible in a way it was not in 2023.

What schema actually buys a software company

Google is explicit that there is no special structured data required to appear in AI features, and that you do not need new machine-readable files or markup for them. Take that at face value. Then note what it does not say: markup still resolves identity, and identity is what decides whether four mentions of your product across four domains are understood as one company.

So mark up for resolution rather than for a rich result. SoftwareApplication carries applicationCategory, operatingSystem, offers, featureList and aggregateRating, and Organization carries the sameAs links that tie your site to your profiles elsewhere. The rich-result payoff on Google is now nil. Google restricted FAQ rich results to well-known authoritative government and health sites in a changelog entry dated 14 September 2023, then deprecated the feature outright: its changelog entry of 8 May 2026 says the FAQ rich result would stop appearing in Google Search from 7 May 2026, and the FAQ structured data page has since been removed and now redirects to the entry of 15 June 2026 recording the removal. So a B2B SaaS site adding FAQPage markup for the Google snippet is optimising for something that no longer renders there. Add it because it labels a question and its answer clearly for anything that parses the page, which is why this post still carries it. We cannot tell you what the other engines do with that markup, because none of them publish a parsing specification we could check.

The deeper version of this argument, including where the three terms stop being interchangeable, is in AEO vs GEO vs SEO.

Four checks you can run this week

None of these need a tool.

  1. Request your pricing page and one docs page with a crawler user agent string and no

    2. JavaScript. If the price or the rate limit is missing from the raw HTML, it is missing for anything that fetches without executing JavaScript. We cannot see which engines render JavaScript on your pages, so treat the raw HTML as the floor of what is available to them rather than the whole story. curl -sA "OAI-SearchBot/1.4" yoursite.com/pricing | grep -i '\$' is the whole test.

  2. Read your own robots.txt and your WAF rule list side by side. Confirm which search

    3. crawlers are allowed by name, not by category. The two files disagree more often than anyone expects, because one is in the repo and the other is in a console.

  3. Ask four engines your three hardest comparison prompts, five times each, and record which

    4. domains were cited rather than which brands were named. You are building a list of the handful of third-party pages that decide your category. In the audits we have run, that list has usually been shorter than ten, though we have not held the prompt set constant across clients, so do not read it as a measured figure.

  4. Open those pages and check what they say about you. Wrong pricing tier, a feature you

    5. shipped last year listed as missing, a dead integration. Correcting a factual error on a page that is already being cited is usually cheaper than trying to displace that page with a new post on your own blog. We cannot attribute a change in answers to any single fix, for the reasons in the limits below.

If you would rather work from a structured version of that, the self-check is ten checks in the same spirit, and an audit is the same exercise run across about 45 prompts and five engines with the evidence kept.

Why this is different from other verticals

The comparative shape of the question is what makes SaaS distinctive. A clinic or a law firm is usually answering a local or an eligibility question, where the brand's own pages are the natural source. A software buyer is running a bake-off, and bake-offs are documented by third parties. AEO by industry sets out where else the work diverges.

Regulated software sits awkwardly between the two. If you sell into hospitals or insurers, your claims carry review obligations that a growth team cannot sign off alone, and the sequencing advice in AEO for healthcare applies to your content even though you are a software company.

Honest limits

This post does not establish that the three failure modes are common in B2B SaaS generally. They are the ones that recur in the audits we run, which are a self-selected sample of companies that already suspected a problem, and we have not counted their frequency against a defined population. We cannot connect a mention in an answer to a closed deal, because these surfaces pass almost no usable referrer, and self-reported source fields on a demo form are better than nothing but are not measurement. We also cannot see inside these systems: where this post says a thing is absent, it means we could not observe it from the outside, not that it does not exist. On stability, Pew Research Center's analysis of 68,879 Google searches made in March 2025 by 900 US adults who agreed to share their browsing activity found those users clicked a result in 8 per cent of visits where an AI summary appeared, against 15 per cent where one did not. That is a snapshot of one engine, one country and one month, and anyone quoting it as a constant is overreading it, and so would we be.

FAQ

Does AEO replace SEO for a SaaS company? No. The crawl, the render and the index are shared infrastructure. What changes is the target: a passage that can be lifted and attributed, and a set of third-party pages that describe you accurately, rather than a position on a results page.

Should we block AI crawlers to protect our documentation? Separate the two kinds first. Blocking a training crawler and blocking a search crawler have different consequences, and OpenAI documents that for its own crawlers the search one controls whether you appear in ChatGPT search answers. Check each engine's own crawler documentation rather than assuming the split works the same way everywhere. For most software companies, public docs are an asset in these systems, not a leak.

How many prompts do we need to track? Fewer than most teams assume, and run more times each. Three or four genuine buying questions per segment, run repeatedly across engines, tells you more than two hundred prompts run once.

We are not in the review-site roundups. What now? Treat inclusion as the project. Get listed where listings are earned, supply accurate product facts to the people who maintain those pages, and fix the errors on the ones that already mention you. It is slow, and it is the part of the work with the longest memory.

Sources

Related reading

AEO by industry: where the work differsAEO for healthcare, with the compliance in mindAEO vs GEO vs SEO: what the three terms actually meanWho owns AEO inside a companyAEO for ecommerce, when the product is the answerAEO inside an enterprise, where approval takes longer