Somewhere in late 2025, a line was crossed: surveys now find that 51% of B2B software buyers start their product research with an AI chatbot more often than with Google. Not "have tried it": start there, as the default first move. If you sell software and your go-to-market mental model still begins with a search results page, half your market exited that model while you weren't looking.
This article is about what actually changed in the funnel (which is subtler than "SEO is dead") and what a small SaaS can concretely do about it.
What the chatbot-first funnel actually looks like
The old funnel began with a query, a results page, and ten tabs of vendor content the buyer triangulated themselves. The new first step outsources the triangulation: "What are the best options for [problem], for a team like [context]? Compare them." The assistant returns a synthesized shortlist (three to five names, with reasoning), and the buyer's subsequent Googling is mostly verification of that shortlist, not discovery.
Three properties of this funnel matter enormously for vendors:
The shortlist is the battlefield. Being named in the assistant's first answer is the new page-one ranking, except there are three to five slots instead of ten blue links, and there's no position eight to limp along in. Products outside the answer don't get verified, visited, or considered. They simply don't exist for that buyer.
The assistant describes you, whether you like it or not. Buyers don't read your positioning; they read the model's summary of you, assembled from your docs, your pricing page, reviews, comparisons, and community threads. If that corpus is thin, outdated, or contradictory, the model's description of you will be too, with total confidence.
Zero-click went terminal. Many buying conversations resolve entirely inside the chat (shortlist, comparison, objection-handling, pricing summary), with the buyer's first touch on your site being the signup page. Your analytics read this as "direct traffic, instant conversion," systematically hiding the channel that produced it. (Watch for chatgpt.com and perplexity.ai referrers and unexplained branded-search lift; that's the channel's shadow.)
How the shortlist gets assembled, and entered
Assistants build product answers from two layers, both workable:
Retrieval (weeks to influence). For current "best X" questions, assistants search the web and synthesize from what they fetch: disproportionately comparison pages, directories, review aggregations, and answer-shaped content with concrete facts. This is the GEO playbook: answer-first structure, entities and numbers and dates, FAQ schema, and presence in the third-party comparison layer (category directories like ours, alternatives sites, honest competitor comparisons on your own domain; the citation goes to whoever does the comparing).
Parametric knowledge (quarters to influence). What the model knows about your category from training: accumulated docs, tutorials, discussions, reviews. Thin public footprints produce hesitant, hedge-y model descriptions; rich ones produce confident recommendations. Every piece of durable, specific public content, including a build-in-public archive, is a deposit in this account.
For developer-facing products there's a third layer: the assistant doesn't just recommend tools, it uses them. A coding agent that scaffolds your library successfully on the first try converts the recommendation into an install. That's agent discoverability, and for dev tools it may already outweigh the chat layer.
The small-vendor advantage (yes, really)
Counterintuitively, the chatbot-first funnel is currently kinder to small products than the Google funnel it's replacing:
- Synthesis rewards the best answer, not the biggest domain. Models assembling a comparison cite the clearest, most specific, most current source, and most incumbents' content is neither answer-shaped nor honest about trade-offs. A small vendor with genuinely useful comparison content regularly out-cites enterprise marketing sites.
- Long-tail context is the native query shape. Buyers give assistants rich context ("for a 3-person bootstrapped team, EU-based, needs SOC 2"), and rich context favors specialized products over horizontal giants. Generic leaders win generic queries; vertical products win the contextualized ones, which is most of them.
- The window is open. Most companies haven't restructured for citability. Early movers in each category are visibly over-represented in assistant answers right now, a first-mover return that won't last past the point where everyone's content is answer-shaped.
The practical program, in priority order: get your category's comparison content right (yours and third-party), restructure your money pages answer-first with schema, keep facts current and dated, seed the communities models retrieve from, and audit monthly by asking the assistants your buyers' questions. None of it is exotic; all of it compounds. It's the same distribution-system logic pointed at a new front door.
Frequently Asked Questions
Do most software buyers really start with AI chatbots now?
The 51% figure refers to B2B software buyers who report starting product research with an AI chatbot more often than with a search engine: a majority, and rising. It doesn't mean search is gone: buyers still use Google to verify the assistant's shortlist. But the discovery moment, where the consideration set is formed, has substantially moved into the chat window.
How do I know if AI assistants are recommending my product?
Audit it directly: monthly, ask ChatGPT, Claude, and Perplexity the five questions your buyers would ask ("best [category] for [context]", "[you] vs [competitor]", "is [you] worth it") and log appearances, descriptions, and cited sources. Indirect signals: referral traffic from chatgpt.com and perplexity.ai, and branded-search or direct-signup growth that classic attribution can't explain.
What's the fastest way to show up in AI shortlists?
Target the retrieval layer: publish genuinely useful comparison content for your category (including honest competitor comparisons), restructure key pages so each section opens with a direct, fact-dense answer, add FAQ schema, and ensure you're accurately listed in the directories and review sites assistants cite. Movement on retrieval-driven questions typically shows within four to eight weeks; the parametric layer takes quarters and rewards starting now.
Is traditional SEO dead for SaaS?
No; it's been demoted from the whole funnel to one layer of it. Crawlability, authority, and ranking still matter both for the verification searches buyers run on their shortlist and as input to what AI assistants retrieve. The work that's obsolete is content written to rank rather than to answer; the work that pays now serves both engines at once, which is exactly what GEO is.