AI citations this week: what ChatGPT, Gemini, and Perplexity quoted
A weekly snapshot of which press releases got cited by which AI engines — and what the citing prompts had in common.
Every week we monitor a rotating set of high-intent prompts across ChatGPT, Gemini, Perplexity, Claude, Grok, and Copilot, and log every URL each engine cites in response. Here's what stood out in the last seven days.
Headlines
- Perplexity continues to cite original press releases far more often than any other engine — 3.2× more than ChatGPT in our sample. If you can only optimize for one surface this quarter, optimize for Perplexity.
- Gemini is cycling source diversity faster than it was three months ago. URLs that were cited weekly in February are now appearing roughly twice a month for the same prompt. Freshness signals matter more, not less.
- ChatGPT is leaning harder on Reddit and Wikipedia for general queries and on first-party press releases for product-specific queries. The split is sharp at around the "what does X do" / "how does X compare" boundary.
What the cited releases had in common
Pulling the URLs Perplexity cited this week and looking at the underlying pages, three patterns repeat:
- The first sentence directly answers a prompt. Every cited release leads with the what + who + when + where. None of them open with "Today we're thrilled to announce..."
- Named quotes outnumber unattributed claims. Cited releases averaged 2.4 first-and-last-name attributions; uncited releases in the same vertical averaged 0.8.
- They ship
FAQPageJSON-LD. Roughly 80% of the cited URLs include an FAQ block with schema; the comparison group is under 15%. This is the cheapest, highest-leverage change you can make today.
For the longer-form version of this argument see Anatomy of a press release that gets cited by Perplexity.
The prompt that surprised us
"What's the best AI press release service for SaaS startups?" — Perplexity returned a four-source answer this week, and three of the four sources were independent blog posts, not vendor pages. That's a long-tail prompt with explicit purchase intent and the vendor pages are losing it to commentary. If you sell into this market, the right move is to publish (or get cited inside) the commentary, not to keep tightening your own landing page.
We'll be back next week with the next snapshot. If you want to monitor your own prompts across all six engines, PressGEO does that — and you can see the live receipts on the proof page.