Custom Proposal
Most AI visibility studies and statistics tackle citations for broad, informational queries. That’s useful data; broad queries are where category education happens. They're what a problem-aware buyer asks before they investigate solutions or brands, and they help us understand terms like “AI SOV” and “brand authority.”
But broad, informational queries don’t tell the whole story, because not all citations are created equal. A Reddit thread, a third-party review, and a product page all might count as citations, but they represent very different potential value for a B2B company.
In B2B, buyers use AI to help evaluate integrations, use cases, specific tools, and other concrete pieces of the buying decision. These are the evaluations that turn an MQL into a Closed-Won deal. We wanted to know what AI surfaces at those moments.
The answer is one-sided: By a wide margin, AI search cites pages that marketing teams produce and control.
This pattern comes from a citation data from Peec AI across mid- to bottom-of-funnel prompts in our B2B and SaaS client base: queries like "Best CRM for a 12-person sales team," "Pipedrive vs. HubSpot for sales-led companies," and "How does X handle SOC 2 reporting." Our findings hold across clients, verticals, prompts, and use cases.
Key takeaways
- Owned and controllable content dominates AI citations at the consideration stage of the funnel: product pages, articles, comparison guides, listicles, and how-to content together account for the bulk of citation volume.
- Product pages led the dataset, accounting for roughly a quarter of all citations.
- Community-driven sources (Reddit, YouTube, forums) accounted for roughly 4% of citations at this stage.
- Comparison-format queries made up about 20% of the prompts but drove roughly 27% of total citation volume, generating citations more efficiently than the average prompt.
What we tracked, and how
The dataset spans companies in Ten Speed's B2B SaaS and professional services client base, across verticals from fintech and physical security to hospitality and IT automation. We deliberately included look-alike competitor brands alongside each client, so the citations reflect the full set of players surfacing in head-to-head queries, not just our clients' own pages.
Across these accounts, we ran a set of evaluation-stage prompts through Peec AI, which monitors how URLs are cited in responses from large language models including ChatGPT, Perplexity, Claude, and Gemini.
For our purposes, a citation is a URL an AI response surfaced as a linked source. We analyzed just over 7,000 citations for this piece. That figure counts total appearances, not unique URLs: a single source cited across many responses contributes to the count each time.
Our team wrote the prompts to reflect mid- to bottom-of-funnel intent: direct product comparisons, evaluation questions, best-of roundups, and how-to queries tied to specific product use cases. We grouped these prompts by buyer intent:
How the evaluation-stage prompt set breaks down
Every prompt was classified by buyer-intent category. The set was weighted toward comparison, alternatives, and use-case queries that signal active evaluation (n = 220 prompts).
Source: Ten Speed BOFU citation study prompt set. Each prompt hand-written to reflect mid-to-bottom-of-funnel buyer intent.
We excluded informational and top-of-funnel queries by design ("what is a CRM," "how does sales pipeline software work"). Citation patterns for top-of-funnel and informational queries can look quite different, and a lot of existing research already covers them (although we may touch on this soon with the same data set).
What evaluation-stage AI prompts actually look like
A sample of the mid-to-bottom-of-funnel queries we tracked, spanning very different industries and buyer questions.
Source: Sample of evaluation-stage prompts from the Ten Speed BOFU citation study. Brand names removed.
Where the citations actually go
The dominant pattern in the data is one-sided. At this funnel stage, AI overwhelmingly cites content B2B teams are already producing.
By a wide margin, the majority of citations went to a relatively small set of page types: product pages, articles (blog posts, news, and PR), comparison content, listicles, how-to guides, and homepages.
Brand-controllable content dominates BOFU AI citations
Page-type breakdown of all cited URLs in evaluation-stage prompts (n = 7,387 citations across 170 prompts).
Source: Ten Speed client and look-alike B2B SaaS data. Citations tracked via Peec AI across mid-to-bottom-of-funnel prompts.
Product pages led, at roughly a quarter of all citations. That's higher than we'd seen in broader citation studies, and it makes intuitive sense for evaluation-stage queries. AI often summarizes what a product is, who it's for, or how it handles a specific use case. The page purpose-built to answer those questions is, unsurprisingly, the one AI cites.
Article-format content (blog posts, news pieces, and PR) accounted for roughly a sixth of citations. Comparison and listicle content each contributed around an eighth. How-to guides made up just under one in ten.
AI uses your homepage to define your brand
Homepages appeared in around 8% of citations, higher than you'd expect for evaluation-stage queries. Similar to a product page, the data suggests AI pulls the homepage when it needs a baseline definition of the company: who you are, what you sell, who it's for. Ask an assistant "what does X do" and the model often reaches for the homepage to ground its answer.
That means your homepage is doing entity-definition work for AI, not just conversion work for humans. State plainly what category you're in, what you integrate with, what you're known for instead of leaning on clever positioning over clear description and you hand the model less to work with.
Your G2 and Capterra profiles are content, not just ratings
Third-party profiles (G2, Capterra, SoftwareFinder) accounted for about 7% of citations. On the surface it might look like a surface companies can’t control. However, the profile copy, category, feature tags, and description are substantially yours, and they're the structured, factual content AI leans on when a buyer asks for options in a category.
That makes directory profiles an indexable surface for reputation, and citations. Beyond watching your star rating and responding to reviews, treat the profile text as something you actively manage. Check that you're listed in the right category, that key integrations are present, and that the product description is current with what you offer. A profile that's miscategorized, missing integrations, or written in stale language misinforms the model at the moment a buyer is building a shortlist.
The Reddit and YouTube question
For about a year now, much of the AI visibility conversation has converged on the same advice: be on Reddit, be on YouTube, be where the community is. That advice captures something real, but it doesn't hold at the bottom of the funnel.
Across the evaluation-stage queries we tracked, community-driven sources (Reddit, YouTube, forums, and similar) accounted for roughly 4% of citation volume. Reddit was present but modest. YouTube was nearly absent, at around 1%.
Community sources are a small slice of BOFU AI citations
Despite the prevailing "be on Reddit, be on YouTube" advice, community-driven sources make up a fraction of citation volume at the evaluation stage.
Source: Ten Speed client and look-alike B2B SaaS data. Citations tracked via Peec AI across mid-to-bottom-of-funnel prompts. Third-party profile pages (7.2%) and unclassified content (0.2%) not shown.
The per-URL pattern is worth flagging. AI cited the Reddit threads that did appear in our pool more aggressively, on average, than other page types. When a Reddit thread surfaces, AI tends to lean on it heavily. The pattern is statistically detectable in our data, and it's consistent with broader observations on Reddit's prominence in AI citations (see ALM Corp's analysis, which makes the case for the kind of query-led evidence our funnel-stage filter provides).
The question we consistently ask is whether Reddit is the right place for B2B marketing teams to invest. We've written separately on why we think Reddit is a red herring for B2B marketers. The TL;DR: broad-citation studies don't filter for B2B commercial intent, the agency tactics behind most Reddit marketing services come with real reputational risk, and LLMs are likely to apply quality and intent filters to Reddit threads the same way Google eventually did. The 4% number in our dataset is consistent with that view. It isn't an argument for chasing those citations.
What this suggests about content strategy
A few directional implications emerge from the data.
Product pages are doing more AEO work than teams credit them for. A product page that clearly answers what the product does, who it's for, and how it handles specific scenarios is a primary citation surface for evaluation-stage queries, with a role that reaches beyond conversion. That argues for thinking about product page copy with AI retrieval in mind: structured, specific, and aimed at the questions a buyer would pose to a chatbot when comparing options.
Comparison content gets less investment than its citation share warrants. "X vs. Y" prompts generated about 27% of citations in our dataset from roughly 20% of the prompt volume. B2B brands produce comparison content somewhat reluctantly, or defensively, and often only against their two or three most direct competitors. The pattern in our data is consistent with a case for producing more of it, producing it earlier, and not waiting until a sales team starts asking for it.
Comparison prompts punch above their weight at BOFU
"X vs Y" queries generate a disproportionate share of AI citations relative to how often they appear in the prompt set.
Outsized influence
Comparison prompts produce a third more citations than their share of the prompt set would predict.
Source: Ten Speed client and look-alike B2B SaaS data. Citations tracked via Peec AI across mid-to-bottom-of-funnel prompts. Comparison-format prompts identified by "vs," "versus," "compare," or "comparison" in prompt text.
Review site and directory presence shows up in a different light, too. Most teams treat G2, Capterra, and similar platforms as a review management problem. The data suggests they're also a content surface for AI. Profile copy and category placement shape what AI retrieves, which means keeping that content accurate, current, and well-categorized has citation implications that reach beyond star ratings.
Reddit and YouTube may still matter at an earlier point in the buyer journey, where buyers are asking awareness and consideration questions. Allocating content investment as if community participation replaces owned editorial would be reading the pattern selectively.
The pattern won't be identical for every brand or vertical. It's strong enough across our client base to be worth a closer look at your own data.
Limitations and open questions
A few things to scope appropriately.
The dataset reflects Ten Speed's B2B SaaS and professional services client base. It doesn't include consumer, e-commerce, or media brands, and patterns there may look quite different. Our team wrote the prompt set to reflect evaluation-stage intent; a different prompt frame would surface different content. Citation count is also not a downstream outcome. Being cited isn't the same as being read, clicked, or converted on, and we'd caution against reading these findings as a revenue argument.
AI citation behavior is moving fast. The patterns here reflect a single point-in-time pull. A recent CMSWire analysis flagged sharp month-over-month volatility in community citation share, and what holds at the moment of any one pull may shift by the next platform update.
The single question we'd most want answered next: do these patterns differ meaningfully across AI platforms? ChatGPT, Perplexity, Claude, and Gemini almost certainly aren't behaving identically at the bottom of the funnel, and the distinction is commercially relevant. We don't have a clean platform breakdown in this dataset. That's the gap we intend to close in the next read.
Methodology
Data source. We pulled citation data from Peec AI in a single point-in-time export. Peec captures brand and URL appearances in responses from large language models, including ChatGPT, Perplexity, Claude, and Gemini. Our team wrote the prompts to represent mid- to bottom-of-funnel buyer intent: the types of queries a B2B or SaaS buyer might run when actively evaluating a purchase. These include direct product comparisons, evaluation queries, best-of roundups, and how-to questions specific to product use cases.
Scope. The data reflects Ten Speed's B2B SaaS and professional services client base. The analysis covers mid- to bottom-of-funnel queries only; we excluded informational and top-of-funnel queries by design. Results may not generalize to other verticals, query types, or company sizes. The client base is not a neutral or random sample of B2B brands.
Classification. We tagged cited URLs by page type. We applied a reclassification pass to URLs initially labeled "Other," using URL structure and page title to assign categories. The final dataset contains 11 classified page types plus a small residual of genuinely unclassifiable pages (under 1% of citations by volume). We rounded all figures in this piece to reflect the directional nature of the findings.
Framing. This analysis is practitioner intelligence based on a specific client base. The findings have not been peer reviewed. We tested statistical patterns, including citation rate comparisons across page types, using nonparametric methods appropriate for non-normal distributions. Read the findings as directional observations about what we're seeing in our data at this funnel stage. They are not definitive industry benchmarks.
Discover how we can help
Book a call with us and we’ll learn all about your company and goals.
If there’s a fit, we will put together a proposal for you that highlights your opportunity and includes our strategic recommendations.