While B2B content teams debate which AI platform to optimize for, the more urgent question is who’s actually in the retrieval set for their category — and the answer is rarely them.
Most B2B content teams are now running some version of the same audit. They’re asking which AI platform to optimize for, mapping their rankings against ChatGPT versus Gemini, debating whether Perplexity matters, wondering if the platform calculus shifts again when Google rolls another update.
It’s a reasonable thing to wonder about. The answer is mostly: optimize for all of them using the same underlying work, and don’t over-rotate on platform differences that are still shifting.
The interesting question is: who is currently in the retrieval set for your category, and why isn’t it you?
What’s Actually In There
The research on this has gotten quite specific. Goodie analyzed 5.7 million LLM citations across ChatGPT, Gemini, Claude, and Perplexity for B2B SaaS queries. The top 10 domains across models accounted for more than a third of all citations. Reddit and G2 led the pack. Then came the usual suspects: Capterra, PCMag, TechRadar, Forbes, TechCrunch.
Your company’s blog was not on that list. Neither was almost any vendor’s owned content.
What this means in practice: when a buyer asks an AI “what’s the best [your category] tool for a 200-person company,” the answer is being assembled almost entirely from third-party sources — review aggregators, comparison sites, tech press, community forums — with very little input from the vendors being evaluated. The vendors are the subject of the answer, not the source of it.
Ahrefs found that only 12% of URLs cited by ChatGPT, Gemini, and Copilot also rank in Google’s top 10 for the same query. The gap between “ranking well” and “being part of what the model reads” has widened to the point where they’re essentially separate games.
This is the specific thing worth sitting with. The content strategy most B2B teams have been running — invest in topical authority on your own domain, build out a content cluster, capture organic rankings — does not automatically translate into retrieval set membership. The AI is not reading your 47-post pillar cluster when it shapes a buyer’s vendor shortlist. It’s reading G2, Reddit, and whoever got quoted in a TechCrunch piece two years ago.
The Uncomfortable Corollary
If you look at the actual composition of what’s being retrieved, the pattern is fairly consistent across categories: the retrieval set for most B2B software queries is small, structurally dominated by aggregators, and already locked in.
G2 recently acquired Capterra, Software Advice, and GetApp from Gartner. Modeling the combined entity suggests G2’s ecosystem would move from the 4th most-cited source to second place (behind only Reddit) for bottom-of-funnel queries. For proof and evidence queries — the “what do customers actually say about [vendor]” searches — their combined share reaches above 12%, nearly double the next closest domain.
That’s one company, through one acquisition, becoming the dominant voice in the AI-generated answers your buyers receive when they’re closest to a decision.
This is not a technology story. It’s a distribution story. The publishers who spent years building structured, frequently-updated, credibility-signaling content about software categories are now the entities that AI trusts to summarize what vendors do. The vendors themselves are largely downstream of that trust.
What This Actually Changes About Content Strategy
The implication isn’t that content on your own domain doesn’t matter. It does — particularly for technical buyers who click through, for SEO signals that still feed into some retrieval pathways, and for the pages that do get cited directly when the query is specific enough to reach you.
But there’s a category of investment that most content teams are chronically under-indexed on, because it doesn’t live in the CMS and doesn’t show up in the organic traffic dashboard.
Third-party presence is the retrieval surface. Research from SE Ranking found that review platforms appear in roughly a third of all AI Overview responses for commercial queries. Quoleady found that 100% of tools appearing in ChatGPT answers for software category queries had Capterra listings, and 99% had G2 profiles — not because the review counts drove placement, but because absence from those platforms was disqualifying. You can’t be selected from a source you’re not on.
The same logic extends to editorial press, industry analyst coverage, and community presence. The model doesn’t care that you published a comprehensive guide to your category on your own blog. It cares whether third parties it already trusts have written about you, in what context, and with what consistency.
The Actual Audit
The useful version of this exercise isn’t a platform comparison. It’s a retrieval audit.
Pick five or six high-intent prompts that a buyer in your category would actually type. “Best [category] for [your ICP].” “[Competitor] alternatives.” “How does [your category] work at enterprise scale.” Run them across two or three AI platforms and take notes — not on which platform is friendlier, but on which sources are being cited. Who is in there? What types of content are they citing? Are you present at all, and if so, through which third-party vehicle?
What you’ll typically find: the retrieval set is narrower than you expected. A handful of sources show up repeatedly. Your own site either doesn’t appear or appears in a narrow slice of queries where the prompt was specific enough to reach your content directly. The aggregators and press are handling the rest.
Once you see it laid out that way, the strategic question becomes cleaner. It’s not “how do I optimize my content for AI retrieval.” It’s “which of these external surfaces am I not present on, and what would it take to change that?”
That’s a different brief than anything in a typical content strategy document. It involves G2 profile management, customer review generation, digital PR, getting onto the comparison pages that are already in the retrieval set, and building the kind of external brand signal that gives a model enough cross-referenced evidence to treat you as an established category player.
It probably also involves publishing less, and making what gets published more specifically extractable — structured, direct, and positioned around the exact questions buyers are asking models.
None of that is particularly glamorous work. But it’s the work that actually connects to the surface where the buying conversation is now happening.