ChatGPT, Perplexity, Gemini: which LLM converts

What happened

Per Search Engine Journal, an expert panel convened to answer the question every CMO is now asking their analytics team: when AI search sends traffic, which assistant actually drives conversions. The panel compared ChatGPT, Perplexity, and Google's Gemini across referral behaviour, intent quality, and downstream pipeline impact.

The headline finding from the panel: ChatGPT sends the largest volume of referral traffic, but Perplexity referrals convert at materially higher rates because users arrive with citation context and a narrower question already framed. Gemini sits in a different category entirely, often resolving the query inside Google's surfaces before a click happens at all.

The panel's framing matters because it moves the conversation past raw visibility ("are we cited?") to commercial outcome ("does the citation produce a buyer?"). Those are not the same problem, and they do not have the same solution.

Why it matters for your brand

The first implication is that "share of LLM citations" is now an insufficient KPI. A brand can be cited frequently in ChatGPT and still see weaker pipeline contribution than a brand cited half as often in Perplexity. For a B2B marketer at an asset manager, a reinsurer, or an industrial OEM, this changes how you brief agencies and how you report to the board. The metric is not citations. The metric is qualified sessions per citation, by assistant.

Second, the three assistants reward different content shapes, and that has direct consequences for editorial planning. ChatGPT tends to surface brands inside broader explanatory answers, which means the user is still in learning mode when they click. Perplexity, because it shows sources inline and encourages the user to verify, sends visitors who already trust the citation. Gemini, embedded in Google's stack, often answers without a click. For financial services brands explaining a product like a private credit fund or a structured note, ChatGPT visibility builds category awareness; Perplexity visibility closes the loop with analysts and intermediaries doing due diligence. Both are needed. They are not interchangeable.

Third, distribution strategy now needs to be assistant-aware. A multilateral such as the World Bank or a UN agency publishing policy research will see Perplexity drive the highest-quality inbound traffic from journalists, researchers, and policy staff, because those users are already in citation-checking mode. ChatGPT will drive bulk awareness traffic that looks weaker on conversion dashboards but seeds the model's future answers. Treating ChatGPT traffic as "low quality" because it converts less well per session is the wrong read. It is a different funnel stage.

Fourth, for industrial groups and philanthropic institutions with long sales or grant cycles, Gemini's zero-click behaviour is the biggest strategic risk. If the answer resolves inside Google's AI Overview or Gemini's response without a referral, the brand has effectively been disintermediated from its own expertise. The defence is to build content that the model cannot summarise without quoting you by name: proprietary data, named frameworks, original benchmarks. Generic thought leadership gets absorbed into the answer. Specific, attributable claims survive as citations.

Fifth, attribution infrastructure has to catch up. Most enterprise marketing teams are still measuring AI referral traffic in aggregate or as "other." The panel's findings only become actionable if you can split ChatGPT, Perplexity, and Gemini sessions in your analytics, tie them to CRM stages, and report conversion rates per assistant. If your team cannot do that today, that is the project for this quarter. Without it, every strategic conversation about AI visibility is happening on vibes.

The signal in context

This piece lands at a moment when the industry is collectively realising that AI search behaves less like a single channel and more like a fragmented set of channels with distinct user intents, retrieval logics, and commercial profiles. ChatGPT skews toward exploratory and conversational use. Perplexity skews toward research and verification. Gemini skews toward Google's existing query base, which is dominated by short, transactional, or navigational intent. Treating them as one bucket called "AI traffic" is the same mistake marketers made a decade ago when they lumped paid social and organic social together and wondered why the numbers did not add up.

The deeper shift is that brand-building inside LLMs is bifurcating into two jobs. One is being present in the training data and retrieval index so the models know who you are. The other is being the source the model picks when a high-intent user is one click from a decision. The first job is won by volume, consistency, and authoritative third-party coverage. The second is won by specificity: the named statistic, the proprietary methodology, the quote that cannot be paraphrased without losing meaning. Most enterprise content programmes are optimised for neither. They are optimised for the old SEO funnel, which assumed a single dominant retrieval system and a click as the unit of value. Both assumptions are now wrong.