← Back to Resources
Benchmark

What gets cited in 2026: 12,000 AI shopping answers

Lantern Team·10 min read

Across 12,000 AI shopping answers Lantern logged in Q1 2026, one finding holds across every provider: the model is more likely to cite somebody talking about your brand than it is to cite your brand directly. Owned domains contribute 20% of all citations on average. Earned (third-party reviewers, trade titles, Sprudge-class category sites) contributes 31%. Social (Reddit, YouTube, community forums) contributes 38%. Competitor-on-competitor citations make up the remaining 11%.

The headline mix

Social (Reddit, YouTube, forums)38%Earned (third-party reviewers, press)31%Own (your domain)20%Competitor (rival brand domain)11%
Citation source-type mix, all providers, Q1 2026. Lantern 12,000-answer sample.

Per-provider divergence

ChatGPT28%34%25%13%Claude32%38%18%12%Perplexity47%28%16%9%Gemini22%34%30%14%Amazon Rufus41%25%20%14%SOCIALEARNEDOWNCOMPETITOR
Citation source-type mix by provider, Q1 2026. Lantern 12,000-answer sample.

Perplexity is the most Social-heavy surface in the sample at 47%. Gemini stands almost alone on Own (30%) — for Gemini-skewed prompt sets, Catalog work is the most direct lift. ChatGPT's Own share at 25% is structural: ChatGPT sources roughly 83% of its product carousels from Google Shopping (ALM Corp, 2026).

The Reddit story

Reddit is the single most consequential citation source of 2025–26 and the most volatile. Tinuiti's Q1 2026 Citations Trends Report puts Reddit at roughly 40% of all citations across major models. That weight has been violently unstable — SaaS Intelligence tracked Reddit's ChatGPT share dropping from roughly 60% to 10% over a two-week window in September 2025.

The Wikipedia story

Wikipedia is the citation source that did not move. Claimed Wikipedia articles appear in roughly 11% of branded-prompt answers across the sample. A brand without a Wikipedia article is invisible on definitional prompts almost regardless of how much Visibility it has on buyer-intent prompts.

Methodology
Sample: ~12,000 distinct AI answers logged across ChatGPT, Claude, Perplexity, Gemini, and Amazon Rufus in Q1 2026, drawn from the 45-prompt monitoring sets of the 200 DTC brands in the Lantern quarterly benchmark. Citation typing (Own / Earned / Competitor / Social) is deterministic per URL.