AI discovery
What kind of content AI assistants actually cite
AI assistants tend to cite content that answers a specific question clearly and can be checked: explainers, comparisons, how-to guides, definitions, original data, reviews and detailed discussions.
One sponsored study of LinkedIn citations found the most-cited articles used lists and clear headings and often included statistics, while opinion pieces performed worse. Formatting alone appears to matter less than having a clear, sourced answer.
The pattern in the evidence
Several analyses of AI citations point in the same direction, although most come from vendors that sell tracking tools.
- A Meltwater and LinkedIn analysis of 9.5 million AI citations, published as sponsored content in August 2026, found that the most-cited LinkedIn articles all used bullets or numbered lists, 92% used clear headings, and 67% included statistics or data. Practical, decision-led formats such as how-to guides, comparisons and buyer explainers performed best. Opinion pieces performed worse on their own.
- Digiday, using Meltwater data, reported that long YouTube videos made up a much larger share of citations than Shorts, and that chapter markers and timestamps appeared to help.
- On the other side, a summary by Papercrane describes a Sprinklr study of 252,000 controlled trials that found formatting had a negligible effect on citation likelihood.
One way to reconcile these: structure correlates with citation because well-structured pages usually answer questions more clearly. Adding headings to a weak page is unlikely to help.
Formats that are commonly cited
| Format | Why it gets used |
|---|---|
| Direct explainer ("What is X?") | Answers a definitional question in the first paragraph |
| Comparison (A vs B, ways to do X) | Matches recommendation and evaluation questions |
| How-to guide | Matches procedural questions, easy to quote step by step |
| Original data or research | Gives the assistant a fact it cannot get elsewhere |
| FAQ with specific answers | Short, quotable answers to common questions |
| Detailed reviews and discussions | Third-party view, often preferred for recommendations |
| Long-form interviews and talks | Specific claims from named experts |
Why opinion alone underperforms
An assistant answering a question needs a statement it can attribute and that fits the question. A broad opinion piece rarely answers a specific question. The same opinion becomes citable when it is tied to a specific question, explained with reasons and supported by an example or data.
This creates a two-layer approach for founders. One layer builds trust with people: the founder's views, experience and judgment, usually on LinkedIn or in a newsletter. The other makes the same knowledge citable: specific question-and-answer pages, comparisons and definitions on the company site. Both can come from the same interview.
How founder expertise becomes citable
- Record the founder answering real buyer questions.
- Pick answers that contain something specific: a number, a criterion, a trade-off, a named failure mode.
- Write one page per question, with the answer first.
- Add the evidence: an example, data with method, a source.
- Credit the founder by name, with a link to their profile, so the expertise is attributable.
- Link the page from related product pages and articles.
What to avoid
- Pages that restate what every competitor says
- Statistics without a source
- Long introductions before the answer
- Bulk-generated pages. Google's spam policies treat many pages generated mainly to manipulate rankings as scaled content abuse.
Measuring whether it works
Track whether the new pages are cited for the questions they answer, using repeated runs across assistants. See how to measure AI visibility.
Sources
- New research reveals 75% of LinkedIn AI citations come from individual profiles, Social Media Today (sponsored by Meltwater), 2026-08-17. Sponsored content. Methods are only partly disclosed.
- In graphic detail: LLMs keep citing YouTube in search results, Digiday, 2026-09-23. Uses Meltwater citation data.
- GEO and AEO: what the evidence supports, Papercrane. Summarises Ahrefs llms.txt data and other studies. Secondary source.
- Spam policies for Google web search: scaled content abuse, Google Search Central.