Citability

For a long time there were two questions you could ask about a piece of content. Can a person read it and understand it, and does it rank near the top of search results. The second question is the job of search engine optimization (SEO), where the work mostly came down to which words appear how often and how many other sites link back to your page.

Then a new kind of screen appeared: you type a question, an AI writes an answer in full sentences, and underneath it lists the sites it drew on. Suddenly there were outcomes the two old questions could not explain — pages that rank perfectly well in search and never once appear as a source when the model composes its answer.

Pagedeck ran into exactly that. Ravi spent a year publishing forty blog posts and doubled search traffic, but when anyone asked an AI assistant what to look for in a team document tool, the sources under the answer were all competitor posts. The writing wasn't bad and the search numbers weren't bad, so there was no way to explain it in the review meeting.

Citability is the word teams started using for that missing column. Alongside "is it readable" and "does it rank," it adds one more question: if someone wanted to repeat this claim to a third party and point at us, is there enough in the sentences for them to feel safe doing it?

When an AI answers a question it reads several documents, mixes what it found into new sentences, and then attaches the documents it leaned on. Citability is decided at that last step. Ten pages can carry the same fact and only the two or three that are easy to verify get attached.

Easy to verify breaks into four things.

1. Whether numbers carry a basis. "More companies are adopting collaboration tools" can't be repeated by anyone, while "as of the first half of 2025, 38 percent of companies under 300 employees had adopted a collaboration tool" can be repeated word for word. The sentence itself says when the figure was measured and who it covers.
2. Whether claims carry evidence. If you print a number, print where it came from in the same place — how many people you surveyed for your own research, or which organization published it in which year for outside data.
3. Whether authorship and dates are visible. A document with no author and no updated date may be perfectly correct, and a reader still can't tell whether it is correct now.
4. Whether sentences stand alone. Citing means lifting a paragraph whole, so a sentence that starts "the second approach mentioned above" needs the paragraph before it and never gets lifted.

Writing a term paper is the closest everyday version of this. You may want to reuse a line you read somewhere, but if there is nothing to put in the source column you drop it and go find another reference. The model sits in the same seat. It has to attach a link under the answer, so a page with nothing verifiable never gets picked.

That points the work in one direction: less publishing more posts, more replacing vague phrases in the sentences you already have with checkable ones. Marking up which line is a price and which is a publication date so machines can tell them apart is structured data, and listing your worthwhile documents up front is llms.txt; both of those get models to your writing, and citability is what makes them choose it once they arrive.

It comes up less when planning new posts and more when revisiting the ones already published. A team that has done content marketing for a few years has dozens of posts, and a large share of them are perfectly accurate while missing all four items above.

Three places it actually gets used.

1. Auditing the existing library. Start with the posts that still pull traffic and check the four items on each. At Pagedeck, twenty-eight of forty posts had no updated date, and nine of the twelve that contained figures gave no source.
2. Product help documentation. Help docs look like the least citable thing you own since they only talk about your own product, but for a question like "how does permission sharing work in this tool," your documentation is the most accurate source there is. Naming screens and buttons exactly and leaving a last-modified date is most of the work.
3. Publishing your own data. When you produce a number that exists nowhere else, every answer that wants to state that number points at you. Pagedeck counting the average number of editors per document from customer usage logs is one of these.

The place to check results is the AI overview, the summarized answer block at the top of search results with its list of cited sites underneath. Check regularly whether your name appears there, and feed the difference between the posts that made it and the posts that didn't into the next round of edits.

"The content is right but it has no citability. Not one of the numbers says when it was measured."

Not a request to rewrite the post, but to fill in the measurement date on every figure.

"This report runs on our own data, so it'll be the most citable thing we've published."

It repeats no one else's material and carries numbers only we have, which leaves many places for it to be attached as a source.

"Let's start by adding updated dates to the help docs. On citability that's the cheapest work there is."

A proposal to begin with the items that return the most for the effort.

At Pagedeck, Ravi runs a content audit every quarter. The team puts the post list on screen and walks the four items, and the conclusion that comes out most often is not to write something new but to fix three sentences in something old. Last quarter the entire output was adding last-modified dates and owner names to thirty help documents.

The most common confusion is with writing stiffly. Being told to add sources and measurement dates sounds like being told to write like an academic paper. The truth runs closer to the opposite. Sentences that get cited are usually short and plain, and simply happen to contain something checkable. Make them long and dense and people leave partway through, and a page nobody finishes doesn't stay in the pool that gets read.

Reading it as "just put numbers in" is the other common one, since the first of the four items is the most visible. A number without a source is worse than no number. When someone tries to verify a bare figure and can't find it, trust in the whole document drops with it, and a site filtered out that way tends not to be pulled back into later answers. Adding one number means adding one line of provenance next to it; that pair is the unit.