|
Listen to this article
Getting your Trinity Audio player ready...
|
BLUF: Earned media drives 84% of AI citations. I tracked my own brand for five months: 1,461 AI mentions, and more than 80% were me vouching for me. Hungry people get quoted; thirsty people quote themselves. Here’s a 15-minute check to see which one you are.
By Nicola Ziady | Published: October 10, 2026 | From Data to Insight
Your Own Website Is Now Your Weakest Citation
Earned media AI citations are mentions of you in sources you don’t own or pay for, and they make up 84% of what answer engines cite. Here’s what happened when I measured mine, and a 15-minute check you can run on yours.
I have a rule. Before I teach anything, I test it on myself first.
So in April I started tracking my own name in AI answers with Otterly. Five months later, I wanted to know how much of my visibility came from earned media AI citations, and how much came from me.
The growth looked great. 46 brand mentions back in April. 1,461 by September. Coverage up from 37% to 58% of the answers I track.
I nearly framed it. Then I opened the source list.
More than 80% of those mentions were me. My website. My LinkedIn. My YouTube channel. Answer engines weren’t citing me. They were reading my diary back to me.
That’s an awkward thing to learn about yourself. It’s worse to learn it after months of teaching other people how to fix it.
The case study :: 46 to 1,461 AI mentions in five months
My AI mentions grew more than 30 times in five months, and more than 80% of them came from my own channels.
| Metric | Day 0 (April 2026) | Day 150 (September 2026) |
|---|---|---|
| Brand mentions in AI answers | 46 | 1,461 |
| Brand coverage | 37% | 58% |
| Average position when mentioned | 1.00 | 1.38 |
| Domain citations | 0 | Thousands |
| Share of mentions that were me | Not measured | More than 80% |
Thirty times the mentions. More coverage. A slightly lower average position, which is what happens when you start showing up in harder, broader answers.
Every number on that table went the right way except one. And that one is the only one an answer engine really cares about.
Earned media AI citations: the 84% you can’t write yourself
Earned media drives 84% of AI citations, so the authority answer engines trust is the authority you can’t publish yourself.
Here’s what I should have checked first. According to Muck Rack’s What Is AI Reading? report (May 2026), earned media accounts for 84% of all citations in ChatGPT, Claude, and Gemini answers.
The sample: more than 25 million links across 17 industries. Journalism alone makes up 27%.
Paid and advertorial content came in at 0.3%.
So the thing you can buy is worth almost nothing. The thing that counts is the one thing you can’t produce yourself.
Muck Rack defines earned media broadly: journalism, academic research, government sources, encyclopedic sites, third-party corporate content. Strip it back and it means one thing. Anyone who isn’t you.
In plain terms, every source that mentions you falls into one of three buckets:
- Owned: what you publish. Your website, blog, newsroom, LinkedIn, YouTube channel.
- Earned: what others choose to publish about you. A trade article, a podcast episode, a conference agenda, an analyst report, a colleague’s post.
- Paid: what you buy. Sponsored posts, advertorials, pay-to-publish bylines.
Answer engines lean almost entirely on the middle one.
Ready to start?
Try this today
Ask ChatGPT and Perplexity “Who are the leading experts on [your topic]?” Ignore the names. Look at the sources cited for them. That’s the earned media your category runs on, and the list of places you need to be.
Why 1,461 Mentions Weren’t Enough …
One exception worth knowing: not every engine reads the same way.
Muck Rack, citing OtterlyAI data, reports that Google’s AI Overviews lean toward brand-owned domains, at 59.8% of citations. And Peec’s analysis of 30 million sources (March 2026) found ChatGPT’s most-cited domain is Wikipedia, while Gemini and Perplexity lean hardest on Reddit.
So your owned pages still matter. They’re the address. They just can’t be the only voice.
Your self-published content is the guy trying too hard at the party
Self-published content reads as a claim, not proof, because answer engines look for a second source that confirms what you say.
You’ve met him. Laughs a beat too loud. Agrees too fast. Tells you twice how well work is going.
Nothing he says is false. That’s not the problem. The problem is he’s the only one saying it.
Answer engines have met him too. A page that exists to tell the world you’re an authority is a claim. Ten pages saying it are ten claims from the same witness.
I’d built a very handsome echo chamber. Good content, well structured, properly marked up. It just had nobody standing behind it.
There’s a word for content like that. Thirsty. Built to look authoritative, rather than confirmed as authoritative by someone else.
Gaping Void put it best: be hungry, not thirsty. Hungry people get quoted. Thirsty people quote themselves.
Try this: Pick your three most-shared owned pages. For each one, find a single outside source that says the same thing. Where you can’t, you’ve just written your pitch list.
Mention count measures volume; citation concentration measures authority, and scattered citations mean no single page is building it.
The second number hurt more than the first. My most-cited page captured just 3% of my total citations.
Hundreds of mentions, and not one place where they gathered. That isn’t reach. It’s confetti.
Volume told me I was visible. Spread told me I had no center of gravity. The source list told me why. When you’re your only witness, your authority scatters across every page you’ve ever published, because nothing outside is pointing anywhere in particular.
That’s the shift hiding in plain sight. For a decade, you asked “Can they find us?” Answer engines ask something else: “Who else says so?”
Try this: Choose one canonical page for each topic you want to own. Point every new internal link, guest post, and interview bio at it. Authority needs an address.
How do you fix self-citation in AI search?
The fix for self-citation is confirmation, not more content: one canonical page to collect your authority, and outside sources to vouch for it.
I didn’t write more. I built one canonical home for my frameworks, so the authority I do earn has a single place to land.
Now I’m going after second witnesses. Editors who decide what they publish. Journalists who need data more than opinions. Research that other people cite for years.
Paying for a byline is off the list. At 0.3% of citations, that’s buying a megaphone for an empty room.
None of this is new thinking. It’s the T in the CITE Framework: triangulate. Your ideas need to show up in places you don’t control, saying what you say, in words you didn’t write.
I’d been teaching it for months. I just hadn’t been doing it.
That’s the point of testing on yourself first. You find out which parts of your advice you were only saying.
Your company has the same problem, at scale. If your company’s AI citations come mostly from your own site, newsroom, and press releases, your brand has the same problem I had.
01
I ran this on myself because I could. Your company is running the same experiment right now. It just isn’t reading the results.
The brand version of my diary is your newsroom. Your blog, your press releases, your product pages. It’s the only content marketing fully controls, so it’s where most budgets go. And it’s exactly what answer engines discount.
02
Do the math on your content plan. If 84% of AI citations come from earned media, every extra page you publish is fighting over the other 16%. That’s the case to take to your CFO: less spend on volume, more on the sources that vouch for you. Press. Analysts. Partners. Research other people cite.
Then look at where your authority lands. When an answer engine cites you on your most important topic, does it point to one strong page, or twenty thin ones? That’s not a content problem. It’s a structural one.
03
And there’s one second witness you can actually control: your own people. A journalist won’t write about you on demand. Your experts can publish under their own names tomorrow, and they read as independent voices in a way your newsroom never will. The question is whether your organization lets them.
Here’s the Triangulation Check, named for the T in CITE. It takes about 15 minutes:
- Export the sources AI engines cite when they mention your company, from Otterly, Semrush, or whichever AI visibility tool you use.
- Sort them into owned, earned, and paid. Your owned share is your thirst level.
- Check concentration by topic. For your top three topics, find what share of citations your single best page captures. Low numbers mean your authority is scattered across pages nobody outside your company points to.
How to read your result.
There’s no published benchmark for owned share yet, so don’t chase a magic number. Track the direction instead. Run the check every quarter: your owned share should fall and your best page’s share should rise. For reference, mine was more than 80% owned, with my top page holding just 3%.
FAQ: earned media AI citations
Any source you don’t own or pay for. Muck Rack’s definition includes journalism, academic research, government sources, encyclopedic sites, and third-party corporate content.
Yes, as the destination. Your site should be the canonical home that earned mentions point to. It just can’t be the only voice vouching for you.
Barely. Muck Rack’s May 2026 data puts paid and advertorial content at 0.3% of AI citations.
Export your cited sources from an AI visibility tool such as Otterly or Semrush. Sort them into owned, earned, and paid. Your owned share is your thirst level.
Yes. Swap your company for your name. Your profile and website are owned sources; a former manager’s post, a conference agenda, or a quote in a trade article are your second witnesses. The math is the same: if you’re the only one vouching for you, answer engines notice.
Thirsty content is content built to look authoritative rather than confirmed as authoritative by someone else. Your own blog, newsroom, and social posts all count when they’re the only sources saying you’re an expert.
No. Muck Rack, citing OtterlyAI data, reports that Google’s AI Overviews lean toward brand-owned domains, at 59.8% of citations. Peec’s March 2026 analysis found ChatGPT’s most-cited domain is Wikipedia, while Gemini and Perplexity lean hardest on Reddit.
About the author
Nicola Ziady tests the next shift in marketing on her own brand first, then turns what works into frameworks leaders can use. She’s a software engineer turned CMO, with two decades leading marketing at Cleveland Clinic, St. Jude Children’s Research Hospital, and a major research university. When she found that more than 80% of her own AI mentions were her quoting herself, she published the numbers. The Invisibility Paradox names the problem. CITE fixes it. The 5 Shifts make the fix stick.
