We measured which sources ChatGPT, Claude, and Perplexity cite for cybersecurity buyers — and which quietly disappear. On ChatGPT, the median citation lasts about a week. Here's how to be one of the few that stays.
A CISO at one of your target accounts opens ChatGPT. She types: "best [your category] platform for a mid-market security team."
Four seconds later, she has her shortlist. Three vendors, cited, with links. She'll spend the next month evaluating those three.
If you're not one of them, you were never in the room — and you'll never know the meeting happened. No impression. No click. No line in your analytics. Just a deal you never saw, closing for someone else.
This is not a hypothetical about "the future of search." It's happening right now, at scale. 94% of B2B buyers used generative AI somewhere in their last purchase. AI-referred traffic to B2B sites is up 975% year over year — and when it lands, it converts at four to five times the rate of a Google click. The people deciding whether to shortlist you have quietly moved to a channel your dashboard can't see.
And here's the part that should keep you up at night.
For 15 weeks, I've tracked which sources AI answer engines cite for real cybersecurity questions. About 43,000 citations, across 9,700 pages, re-checked every single week. The pattern is brutal and consistent:
A page cited by ChatGPT holds that citation for a median of one week. Then it's gone, replaced by someone else — and the cycle repeats.
I watched it happen to companies you know. SailPoint went from 62 AI citations to 17. Check Point, 60 to 17. Duo, 45 to 11 — in about eight weeks. Category leaders. None of them announced it. As far as I can tell, none of them even noticed.
Independent research agrees this is systemic: AirOps' 2026 State of AI Search found only about 30% of brands stay visible across AI answers consistently. Seven in ten don't.
Right now your feed is full of "AI search optimization." Get cited. Show up. Add the schema, write the FAQ, check the boxes. And there's a whole category of tools that will sell you a dashboard showing whether you're mentioned today.
But presence is the wrong metric. A snapshot of "are we cited right now" tells you nothing about tomorrow — and in a channel where the median citation evaporates in a week, tomorrow is the only thing that matters. You can report a healthy number to your CEO on Monday and be gone by the next board meeting, with no warning and no idea why.
The metric that actually protects your pipeline isn't whether AI cites you. It's whether AI keeps citing you — and almost no one is measuring it.
That's the entire reason Quoted exists.
I'm Dan Russell. I run Quoted, a research lab that does one thing: measure which sources AI answer engines cite for real buyers' questions, week after week, and figure out why some sources stay cited while others quietly fall out.
We exist to make AI search less of a black box — to give businesses of every size honest, accurate research on what actually earns and keeps a citation. That means a few things we take seriously: every finding ships with its numbers and its limits, we re-test every week, and we say so plainly when something stops holding. We'd rather be trusted than loud.
Not opinions. Not a checklist someone made up. Sixteen content patterns that survived statistical correction and held up when we split the data in half — ranked by how often pages that use them stayed cited.
The gap between a page that uses these and a page that doesn't is the difference between 32% and 83% — between churning out of the answers in a week and staying in them for months.
What we learned from 43,000+ AI citations: the patterns that separate the pages AI keeps citing from the ones it drops, how long each engine’s citations really last, and the handful of things worth doing first — data almost no cyber marketing team has seen. Drop your email at the top of this page and we’ll send it to your inbox.
It’ll put you ahead of almost everyone in your category. Take it and run.
The channel isn’t waiting. AI referral traffic nearly doubled again this year. Your competitors are already in those answers — and some are already falling out without knowing it. The research shows you the handful of things that actually keep you cited.
Real. Weekly runs of prompts in the cybersecurity niche — ~43,000 citations across three engines with live web search. Every finding is corrected for statistical luck and re-checked on split halves of the data before we report it. We publish our stats and our limits — and we retract in public when something stops holding.
It's a five-minute read with exactly one thing to do. It exists to save you time — so you stop chasing AEO tactics that quietly stopped working and spend your effort on the handful that actually keep you cited.
Especially for you. A small team can't afford to pour months into AI-search tactics that don't hold. This tells you where to spend the effort so it compounds instead of resetting every few weeks.
That's our full study — your brand, your prompts, your competitors, tracked weekly. A Quoted Study is your bespoke playbook, built on your data. If you're interested in creating your own, click here.
Start with our free research: what keeps content cited in AI search, with the evidence and the limits.
Citation figures are from the Quoted research corpus (cybersecurity, 15 weeks). Durability figures reflect the engines with a full measurement history. External statistics attributed to their sources (Opollo, AirOps State of AI Search 2026).