State of LLM search, Q1 2026

This is our first quarterly state-of-the-market report on LLM search, drawn from Promptive's monitoring data across Q1 2026. The numbers come from 1.2M brand mentions across 420 tracked brands, sampled across ChatGPT, Claude, Gemini, and Perplexity on daily or better schedules. We're publishing this because the market is making strategic decisions on incomplete data, and the patterns we're seeing are material to any brand with AI visibility on its roadmap.

The headline: Gemini is gaining ground, Perplexity is deepening

The four-engine landscape in Q1 2026 is not static. ChatGPT still commands the largest share of user queries across our tracked categories, roughly 38% of the prompt volume we sampled originated in ChatGPT-adjacent surfaces. But Gemini's AI Overview integration has materially changed the calculus: brands that rank poorly in Gemini are now losing Google SERP real estate in addition to AI answer presence. That's a double penalty that wasn't a factor six months ago, and it's accelerating the ROI case for brands that have been on the fence about GEO investment.

Perplexity showed the most significant quality increase of Q1. Its citation footprint expanded, more unique domains cited per answer on average, and its willingness to name specific brands in product recommendation answers increased noticeably across B2B software categories. Perplexity remains the most citation-honest of the four engines: when it recommends a brand, it almost always surfaces a source URL, which makes its sentiment data a more traceable and therefore more actionable signal.

Claude maintained its characteristic careful qualification style throughout Q1. We observed fewer unqualified top-of-list recommendations from Claude than from any other engine. Its default pattern, multiple options named with use-case specificity, makes Claude answers harder to "win" outright, but it also means consistent presence in Claude's nuanced multi-option answers signals genuine category authority, not just citation luck. For brands trying to own a category in the AI layer, Claude presence is the hardest to game and the most durable to hold.

Share of voice distribution: how concentrated is it?

Across our tracked categories in Q1, the top three brands by AI share of voice captured approximately 71% of all brand mentions in category-level prompts. That concentration is higher than in equivalent SERP click-share data for the same categories, which typically shows a top-three capture rate of 55-60%. The winner-take-most dynamic we've described in previous analyses is real, measurable, and intensifying.

The categories showing the most contested share-of-voice fights in Q1: AI visibility monitoring (our own category), revenue operations tooling, data observability, and security posture management. All four are relatively new categories, less than five years old at meaningful scale, where no player has established the kind of multi-year training data dominance that incumbents enjoy in mature markets. These are the highest-leverage categories for GEO investment: the narrative is still being written, and citations earned today influence model outputs for months.

The categories showing the most entrenched positions: CRM, marketing automation, and cloud infrastructure. In these spaces, incumbents with 15+ years of training data history have what amounts to a structural first-mover advantage in AI answers. Challengers in these categories consistently report that Perplexity, the most RAG-forward engine, least dependent on historical training data, is the fastest to update, while ChatGPT and Claude are slowest to revise their priors about category leaders.

Sentiment trends across the quarter

Aggregate sentiment across our tracked brand set held at +0.28 on our −1 to +1 scale for the first half of Q1, then declined to +0.22 by mid-March. The decline was not uniform, it was concentrated in specific categories and driven by two identifiable causes.

• Product quality conversations on Reddit and Hacker News in several B2B software categories gained sustained upvote momentum in February and March. The engines absorbed this as a negative signal that showed up in Promptive sentiment scores before it registered in traditional brand monitoring tools, validating the value of AI-specific sentiment tracking as an early warning system.
• Two significant model updates, one to GPT-4o and one to the Gemini base model used for AI Overviews, shifted both engines toward more cautious, qualified language across almost all categories simultaneously. This produced a systematic sentiment decrease that our calibration loop caught and flagged to customers as an engine-level event, not a brand-level problem. Without that flag, many customers would have spent time investigating a "decline" that was actually a measurement artifact.

The brands that maintained or improved sentiment through Q1 shared one trait: active content programs generating fresh, citable material at regular intervals. Freshness matters most in Perplexity and Gemini, where retrieval weight is higher relative to training weight. Brands that slowed content production in H2 2025 were running on stale training data by Q1 2026, and the engines' language about them reflected it.

Citation source shifts: what's changing in the trust stack

The citation source landscape shifted meaningfully in Q1. Reddit's share of citations in our monitored set increased from 28% in Q4 2025 to 34% in Q1 2026, the single largest citation-source shift we've observed across any quarter. The primary driver appears to be a Perplexity algorithm adjustment that weights community discussion content more heavily in product recommendation answers.

G2 and Capterra maintained their positions as the most-cited review platforms, together accounting for approximately 16% of total citations, but the type of G2 content being cited changed. In Q4 2025, G2 category and listicle pages dominated. In Q1 2026, specific review pages with detailed pros/cons narratives were cited more frequently than aggregate lists. The engines appear to be improving at extracting attributable claims from primary review content, not just using category pages as general authority anchors.

First-party brand pages increased their citation share from 18% to 22% across the quarter. This is encouraging and counterintuitive, conventional wisdom has held that engines prefer third-party sources over brand-owned content. What changed: the first-party pages getting cited in Q1 are structurally different from those consistently ignored in 2025. They lead with clean definitions, name competitors explicitly, and make specific bounded claims. Pages written in generic brand voice continue to be ignored. The citation share gains are going to brands that have rewritten for extractability.

Three things to watch in Q2 2026

First: Gemini's AI Overview integration is still expanding to new query types and geographies. Brands with strong traditional SEO positions that have not engaged with GEO are likely to see AI Overview cannibalization of their organic click share increase in Q2. This will make the ROI case for GEO more urgent for SEO-first teams.

Second: agentic AI applications are creating a new citation surface that our current monitoring doesn't fully capture. When an AI agent is tasked with finding the best tool for a workflow, it typically queries one or more of the four engines we monitor, or accesses their APIs directly. Brands that dominate the direct-query answer layer will increasingly dominate agentic task outputs, compounding their advantage into automation surfaces that traditional marketing attribution can't see at all.

Third: we're expecting multi-engine model updates in Q2, and we're pre-positioned to detect calibration events faster than we did in Q1. We'll publish an engine-level event notice within 48 hours of any detected systematic sentiment shift so customers can distinguish real brand changes from measurement artifacts. Our Q2 state-of-the-market report will be out in early July, with updated citation source rankings, per-category share-of-voice data, and a look at how the agentic layer is reshaping the competitive surface.

Methodology

All data in this report comes from Promptive's monitoring infrastructure. Brands included are those on continuous monitoring for the full Q1 period (January 1 - March 31, 2026) with at least 10 tracked prompts and daily or better sampling frequency. The 1.2M mention figure covers all brand mentions recorded across all tracked brands, prompts, and engines during the period. Citation percentages are calculated from the subset of answers that included at least one source URL. Sentiment figures use the scoring methodology published at getpromptive.ai/docs.