Every quarter we run a fixed panel of 120 prompts through ChatGPT, Perplexity, Gemini & Google AI Overviews to reveal exactly who gets cited for SEO & marketing - and how fast it's changing. Nobody else tracks this on a fixed panel.
Reddit is still the single most-cited domain across the four engines for SEO and marketing queries, holding roughly 41% citation share in Q1 - but its dominance is eroding as engines diversify their sources.
Only 38% of AI-Overview citations came from pages ranking in Google's top 10 for the prompt - down from 76% a year earlier. In plain terms: you no longer have to out-rank the incumbents to get cited. Structure and specificity now matter as much as domain authority - the single biggest opening a new site has had in years.
76% of the most-cited ChatGPT pages were updated within the previous 30 days. Cited pages skew markedly newer than typical organic results. A visible "last updated" date plus a genuine refresh of the numbers is one of the cheapest ways to move into the citation set.
AI-citation share moves fast. ChatGPT's Reddit citations swung from ~60% to ~10% inside six weeks in late 2025 before partially recovering - which is precisely why a fixed-panel quarterly measurement exists. Point-in-time snapshots from different questions are noise; the same prompts over time are signal.
Each quarter we run an identical panel of 120 SEO- and marketing-related prompts - a fixed mix of "best tool", "how to", comparison, and definition queries - through four engines: ChatGPT (web), Perplexity, Google Gemini, and Google AI Overviews. We log every cited domain and URL, then compute citation share, quarter-over-quarter change, and format of the cited pages.
Limitations: AI answers are non-deterministic, so we run each prompt three times and take the modal citation set; results reflect a US-English, logged-out context; and the panel is SEO/marketing-specific, so shares won't generalise to other niches. The raw prompt list and per-engine data are published openly below.
The full prompt list, per-engine citation table, and raw CSV are free and ungated.
Download the dataset (CSV) →Reddit at 41%, citations decoupling from rankings, freshness as a lever.
Same 120-prompt panel, re-run in April 2026. Subscribe to get it first.
The Q2 tracker drops in the newsletter before it's public - with the full data. Join The Lab to get it.