Published October 16, 2026 · 7 min read
Meltwater analyzed 9.5 million AI citations across six AI tools. Here is what makes content quotable, and how to build a page those tools can actually read.
People do not search the way they used to. They ask a question and act on the one answer they get back. That answer is assembled from a handful of sources, and which sources get picked is not a mystery.
Meltwater analyzed 9.5 million AI citations across ChatGPT 5, Copilot, Claude Sonnet 4, Gemini, Google AI Mode, and Google AI Overviews. The report is aimed at B2B marketing teams, but most of what it found applies to a single person with a profile, a portfolio, and something to say. The rules describe what gets quoted.
Here is what the data says, and what you can do about it this week.
This is the part that matters most if you are one person rather than a department. 51% of the cited creators in the study have fewer than 10,000 followers, and the biggest share of citations came from accounts in the 1,000 to 10,000 range.
The models are not counting your reach. They are looking for a credible person answering a specific question with specifics: your job title, your company, your industry, your examples, your numbers. That is a profile you can build today.
The same report found that content from executives performs well per article (CEOs 8.2%, founders 7.5%), but the report is clear that expertise matters more than job title. A specialist who writes about their own domain, with details only they would know, is exactly the source these systems are looking for.
The study looked at the 24 most cited LinkedIn articles and counted what they had in common.
| Element | Share of top-cited articles |
|---|---|
| Bulleted or numbered lists | 100% |
| Clear H2/H3 headings | 92% |
| Named companies, tools, or products | 75% |
| Hard numbers and data | 67% |
| Comparison or evaluation frameworks | 50% |
| "How to choose" decision guidance | 33% |
Format matters as much as structure. The most cited formats were:
Written content carries almost all of the weight. Plain text posts made up 72% of cited content and articles another 12%, for 83% combined. Video came in at 11%. A well produced video can build an audience, but the citation goes to the paragraph that answers the question directly.
An article that gets quoted tends to look like a buyer's guide: a clear question, the answer near the top, criteria, ranked options, a section on how to choose, and an FAQ. Vague opinion is the one format that consistently underperforms.
48% of cited content was published within the previous three months. Only 12% was more than a year old. Reposting does not count for much either: 72% of cited content was original.
The practical read is that one great article from two years ago is not a strategy. Updating it is. Add the new numbers, rename the tools that changed, keep the page current, and write the next one. Meltwater's own recommendation to teams is two to three posts per week per expert, with longer structured pieces publishing on a slower cadence.
Platforms you do not own take the largest share. User-generated sites like LinkedIn, Reddit, and YouTube accounted for 47.5% of citations, review sites like G2 and Capterra took 15.0%, and company websites took 18.7%.
So the honest answer is layered. You should publish where the conversations happen, because that is where models look first. You also need a page you own, because it is the only source where you decide the structure, the entities, the numbers, and how current the facts are.
That is what ProfilePilot is for. A profile at profilepilot.net/yourname is built from twelve modules you toggle on and off: experience, education, skills, portfolio, and more. It includes a blog at your own URL with categories, tags, and series, and a resume that generates as both a PDF and an ATS-friendly web page at /[yourname]/resume using semantic HTML and structured data.
The pieces that map to this research:
Free to start, no credit card: all twelve modules, three basic themes, three layouts, the blog, the Network for peer reviews, a basic resume template, share links, and data export.
Do I need a large following to show up in AI answers?
No. 51% of cited creators in the study have fewer than 10,000 followers, and accounts in the 1,000 to 10,000 band produced the largest share of citations.
What kind of content gets cited most often?
Structured, specific, decision-useful content: "Best X" lists, side-by-side comparisons, and how-to-choose guides. Most cited articles used lists, headings, named products, and hard numbers.
Does video or imagery help?
Written structure is what gets quoted, since text is easiest for a model to parse and reuse. Images and video help with engagement, which helps distribution.
How often should I publish?
Consistently and recently. 48% of cited content was published in the previous three months, so a steady cadence beats occasional big swings. Two to three short posts a week per expert is a realistic starting point.
Does my own website matter if platforms get cited more?
It matters, and it is the part you fully control. Company sites took 18.7% of citations. Your own page is where your facts live in the format models can read, and where they stay current.
Being found used to mean ranking. Now it means being quoted, and the things that get quoted are boring in the best way: clear structure, real names, hard numbers, a direct answer, and recent updates from a person who knows the subject.
You do not need a content team to do that. You need a page you own, a habit of answering real questions, and the discipline to keep it current.
