12.3% of the citations are closed, not contested, and the difference decides a plan

Of the 20,143 citations we recorded on 16 October 2026 across 1,260 readings of ChatGPT and Gemini, 12.3% landed on official or platform documentation. The number is not large. What makes it worth a page is that this share is closed rather than contested, and almost every content plan written in this category treats the two the same way.

A contested share is one you are losing and could win. A closed share is one where the publisher is the subject of the question, so the best page anybody else writes is still second hand. Mistaking the second for the first is how a quarter gets spent writing against the manufacturer of the thing being asked about.

The three signs a share is closed

These are general tests, and they are the ones that sorted the documentation share in our own data.

• The publisher is the subject. The question is about a thing, and the page is published by whoever makes that thing. Nobody else can be the primary record.
• The page is the authority by definition. When the question is what a setting does or what a rule says, the document that defines it is the answer, and a description of that document is strictly worse.
• Corrections only run one way. If the page is wrong, the only party who can fix it is the publisher. You cannot compete with a document you also cannot correct.

All three hold for the documentation rows at the top of our source table. Four Google hostnames and two OpenAI hostnames sit in the top ten of that run, and the questions they answer are questions about Google and OpenAI products.

Finding the closed share in your own category

Every category has one and the owner is rarely a search engine. Run your buying questions through this and the closed set falls out quickly.

Who owns the record	The kind of question it closes	What to publish instead
A regulator or a government department	What the rule is, what the threshold is, which form applies	What the rule means for one specific situation, dated, quoting the source
A standards body	What the specification says	A worked implementation, with the parts that caught you out
A manufacturer or platform	What the product does, what the setting controls	What it costs you to run, what it does not cover, what you chose instead
A professional body	Who is qualified, what the code of conduct requires	How to brief and judge one, which nobody official publishes

The right hand column is the pattern. Against a closed record you do not write a better version of the record. You write the thing the record structurally cannot contain, which is a decision made in a real situation with its constraints named.

Why the share is small and still worth naming

12.3% is the second largest row in the table and it is a long way behind the first, which is vendor and agency own sites at 82.0%. Two more rows sit below it: news or a large publisher at 3.2% and social or community at 1.9%, with review directories at 0.7%. So the documentation share is not the obstacle most people think it is. The obstacle is that it is concentrated in the questions people most enjoy writing about, which are the mechanical ones.

The earlier run of 15 October 2026, a different bank of 102 questions per market over 597 usable readings, put the same documentation row at 9.7% of 10,652 citations. Two banks, two figures a few points apart, same conclusion about which kind of question goes which way.

The discipline this asks for

• Label each question before you brief a page. Closed, contested, or open. Three words, written in the brief.
• Do not measure yourself on closed questions. Your absence there is not a result and tracking it adds noise to the questions that matter.
• Still write the closed ones if readers ask. Judge them on whether your readers used them, and leave them out of the citation target.
• Re-check the label when the owner changes. A record moving from closed to open is the single best moment to publish, and it is visible in the publisher's own dated announcements.

Who does that labelling, and whether the writing is done inside or bought in, is a separate call with its own failure modes, and judging the output before signing is the part most often skipped. The five kinds of supplier and the questions that separate them are in the five kinds of content generation company and how to judge output.

What this reading does not cover

Two apps were read, ChatGPT and Gemini, on one date, across 630 questions in three markets. The three further surfaces our product covers were asked nothing, so no sentence here extends to them. The run holds no per industry rate, since no question named an industry, and no city figure, which came from a separate scan of ten named places. A citation here records a domain in a source list, not a business named in the answer text.

We sell the measurement and the publishing that follows it, so the figure that goes against us is on the page too: our own domain was in the source list of 6 of the 1,260 readings. The search console side of our own record sits on our proof page.

Questions about documentation as a closed citation share

How much of the citation pool is official documentation?

12.3% of 20,143 citations on 16 October 2026. The earlier bank of questions, read on 15 October 2026, put it at 9.7% of 10,652.

Can I outrank platform documentation on its own subject?

That is the wrong target for the work. The document defines the thing being asked about, so the position available to you is the applied decision beside it rather than the record itself.

How do I tell a closed share from one I am losing?

Ask who owns the record and who can correct it. If both answers are the publisher of the page that got cited, the share is closed and your absence from it is not a result.

Does this mean technical content is a waste of time?

Technical content earns readers, and readers are a reason on their own. What it will not reliably earn, on these counts, is a citation on a question whose answer is defined by somebody else's manual.

Which engines and which date?

ChatGPT and Gemini, 210 questions per market, read on 16 October 2026. The earlier reading, with its method and full source mix published alongside, is 597 readings across ChatGPT and Gemini in three markets.

Write the label in the brief

One word at the top of each content brief, saying whether the question is closed, contested or open, changes what gets commissioned more than any amount of tooling. The closed ones get written for readers and taken off the scoreboard. The contested ones get the dates, the figures and the method. The open ones get written first, because being first is the only advantage that does not have to be taken off anybody.