The B2B AI Discovery Playbook: Lessons from 380k Answers with Ross Simmonds

Before a B2B buyer books a demo, they ask an AI assistant which vendors belong on the shortlist. Ross Simmonds of Foundation joined Josh Spilker, Head of Search Marketing at AirOps, to unpack what those answers cite. Ross also shared data on how brands earn a place in them, including findings Foundation had not published anywhere else.

With joint research from Foundation and AirOps, the team analyzed over 380,000 answers and over 3 million citations across more than 80 B2B and enterprise categories. The data spans six surfaces: ChatGPT, Gemini, Perplexity, Claude, Google AI Mode, and Google AI Overviews.

Why Should Every AI Visibility Plan Start with a Goal?

Ross framed the research as a recipe, where the goal comes first and the right ingredients depend on the prompts buyers use in your category.

• Pick one primary goal, such as pipeline or sentiment, and judge every new page, review push, or PR placement against it.
• Show experience, expertise, authority, and trust (EEAT) with real author pages in place of "Acme team" bylines, plus phrasing like "we analyzed this."
• Give PR a new brief: place stories in the publications large language models (LLMs) already cite, then check whether answers pull from them.

"Start with your goal. If that is the only message that I got across to enterprise brands today, I would be very happy." - Ross Simmonds

The Two Internets: How Do Google Surfaces and AI Assistants Cite Sources Differently?

The study splits AI discovery into two internets: Google's AI Overviews and AI Mode on one side, and the assistants on the other. On Google, Reddit appears in 39% to 52% of answers and YouTube in 38% to 47%, while review platforms stay under 10%.

• Assistants send 52% to 72% of citations to web pages and 16% to 21% to review platforms, and Claude cites no Reddit or YouTube.
• Follow the deals, as Ross does: Google licensed Reddit content in 2024, and Ross links Reddit's lawsuit against Anthropic to Claude skipping Reddit.
• Reddit fell from 1 in 5 ChatGPT answers to 1 in 30 in three weeks, so learn how ChatGPT retrieves and cites pages first.

Which Page Types Earn the Most Citations in AI Search?

Ross named six page types worth building on your own site: product, solution, comparison, alternative, industry, and segment pages. HubSpot publishes comparison pages against its competitors at scale, and a Foundation analysis found that publishing velocity correlates with citation rates.

• Build solution pages by use case, the way Asana runs pages for marketing and finance teams, so prompts naming that team surface your page.
• Keep alternative and comparison pages separate: a coffee vs tea shopper knows tea, but a coffee alternatives shopper may never have heard of matcha.
• Write for long, specific prompts, like a revenue intelligence platform for a team onboarding five new sales reps, where coined category language wins.

How Do You Build a Review Engine That AI Models Trust?

Review sites earn over 15% of citations in ChatGPT, Perplexity, and Claude, and they matter most on bottom-of-funnel commercial prompts like "what is the best." G2 led with citation rates of 20% on Claude, 19% on Perplexity, and 15% on ChatGPT, with Capterra at 15% and TrustRadius at 10%.

• Ask for a G2 or Capterra review right after a customer gives a high net promoter score (NPS) on a key product action.
• Keep reviews arriving all year, because Ross says LLMs count how many reviews you have and how recently customers left each one.
• Assign one owner for G2, Capterra, and TrustRadius, and report review volume and recency next to your other pipeline inputs each month.

How Can You Counter a Negative Narrative in AI Answers?

In one case from the study, Claude kept citing a competitor's neutral to negative page to argue that the brand was not worth working with. Ross's fix is to dilute that single page by scaling your review engine and contributing useful content to the other sites models read.

• Connect the person who owns AI visibility with product marketing, so in-product moments trigger NPS surveys, review requests, and creator partnerships.
• Write blog posts and reports for integration partners, since Ross sees many partner teams with nobody available to write that content for them.
• Brief your content team to publish and contribute pieces on fixed product gaps, and send press releases announcing each fix.

"You cannot allow one URL to negatively skew your efforts." - Ross Simmonds

How to Put B2B AI Discovery Insights Into Practice

Google still handles far more search volume than ChatGPT, so Ross tells Foundation clients to protect the channel where most of their pipeline comes from.

Demand is moving among the assistants: Claude search demand rose 385% while ChatGPT's fell 35%, and Anthropic captures 40% of US enterprise LLM spend. Claude also leans more negative on B2B prompts than Perplexity, which matters for any brand selling to enterprises.

Start this quarter by tracking AI visibility on your own prompts, with Google surfaces and LLMs reported separately. It’s simple to do this within an AirOps dashboard.

In managed databases and continuous integration tools, Reddit outranks review sites, so pick answer engine optimization strategies that fit your category's citation mix.

What Comes Next for B2B Discovery in AI Search?

Ross predicts LLM ads will outgrow today's search and social ad markets, noting Facebook earned $150 million in 2007 and Meta $196 billion last year. He also expects every unicorn B2B company to own a media property by 2035, following moves like Semrush buying Backlinko and Spotify buying The Ringer.

Stories worth citing, placed where the models already look, will decide which B2B brands reach the shortlist before any sales call happens.

Want to see how AirOps helps teams create a strategy for AI search? Work with us.

‍