News and large publishers took 3.2% of the citations, which fixes the size of the prize

News and large publishers accounted for 3.2% of the 20,143 citations we recorded on 16 October 2026 across 1,260 readings of ChatGPT and Gemini. An earlier bank of questions, read on 15 October 2026 across 597 usable readings and 10,652 citations, put the same category at 2.5%. Two differently designed question sets, one day apart, both landing between two and four points.

That is the size of the prize, and knowing it before a retainer is signed changes the conversation from a hope to a budget line. It does not make press worthless. It makes press a thing you buy for the people who read it, with a known and small contribution to a source list.

The two readings, and the rest of the table

Kind of page	15 October 2026, of 10,652 citations	16 October 2026, of 20,143 citations
A company's or agency's own website	85.6%	82.0%
Official or platform documentation	9.7%	12.3%
News or a large publisher	2.5%	3.2%
Social or community	1.5%	1.9%
A review directory	0.8%	0.7%

The news row is third of five on both readings, and the gap between it and the row above is large on both. A plan that treats coverage as the main route into these answers has chosen the third largest share and left the largest one, which is pages on your own site, to somebody else.

What a single publisher's behaviour adds to the picture

One news organisation reached the top ten of the 16 October source table, at 197 readings of 1,260. When we fetched the robots.txt of 26 domains our audits had named, on 20 October 2026, that same host blocked eleven of the fourteen agents we tested at the root and allowed GPTBot and OAI-SearchBot. The access run of 16 October had separately recorded it returning HTTP 403 to PerplexityBot and ClaudeBot.

So the practical ceiling on a press route has two parts rather than one. The category's share is small, and the individual title may not permit the agents you are measuring. The second part is free to check and almost nobody checks it.

What to do before a placement is bought

• Fetch the publication's robots.txt. It is a public file and it takes a minute. A title that permits two of fourteen agents is a different purchase from one that permits fourteen, at the same price.
• Check the edge as well as the file. They disagree often enough that one reading is not an answer, and the method for both layers is published as a dated test of robots.txt, HTTP access and named agents.
• Write down what the placement is for. Reaching buyers, recruiting, raising money and reaching a source list are four different purchases. Press is strong at the first three.
• Ask whether the piece will carry a figure. A placement that quotes one of your numbers with a date is a different asset from one that quotes an adjective.
• Keep the measurement separate. Report press on its own terms and keep it out of a citation target, because 3.2% is the whole category.

Where press does carry weight in our own data

Not in the aggregate. In one specific city reading, the four most cited sources for Houston included a national newsroom, axios.com at 2 citations of the city's 12 readings, and in Birmingham a national newspaper site, thetimes.com at 3 of 12. Those are six question readings per city on both engines, from the separate ten city scan taken on 15 and 16 October 2026, and they say something narrow and real: on a local question with few strong local pages, a newsroom with a local desk can fill the slot.

That is the shape of a useful press expectation. It is not a category wide route into answers. It is a thing that happens where nothing local has been published well, which is also the condition that makes a local page worth writing.

It is worth saying what those two city rows are not. They record which sources an answer used and nothing about how those sources got there, so a newsroom placement remains an untested claim either way. And they are not a figure for any other city: across the whole ten city scan our own domain was cited zero times, and the only cities with numbers are the ten we read.

The limits of both readings

Two apps only, ChatGPT and Gemini, on single dates, with no reading taken from the three further surfaces our product covers. No per industry rate exists in either run, because no industry was named. The city rows are twelve readings each across ten named places and extend to no other city. And every figure counts a domain in a source list rather than a business named in the answer text, which is a separate measurement with a separate denominator.

We sell AI visibility measurement and the publishing work that follows, so the number that does us no favours is on the page: 6 of those 1,260 readings had a page of ours in the source list. The Google Search Console side of our own record, with both windows and the export date printed on it, is our proof page.

Questions about press coverage and AI citations

What share of AI citations went to news publishers?

3.2% of 20,143 citations on 16 October 2026, and 2.5% of 10,652 on the differently written bank of 15 October 2026.

Is public relations a waste of money for this?

For reaching readers, no. As a route into a source list it is a 3.2% category, which is a budget fact rather than a verdict, and it belongs in a different line of the plan from the citation work.

Did any news site reach the top of your table?

One, at 197 readings of 1,260, and it is the only news publisher in that top ten. The same host blocks eleven of the fourteen agents we tested in its own robots.txt.

Do local newsrooms help on local questions?

In two of the ten cities we scanned, a newsroom site was among the most cited sources at 2 and 3 citations of 12 readings. That is a reading about those cities on those six questions and it does not extend to a city we did not measure.

Which engines and dates produced these shares?

ChatGPT and Gemini on 15 and 16 October 2026, with 102 then 210 questions per market. The first run's method and full source mix are published as 597 readings across ChatGPT and Gemini in three markets.

Buy the coverage, size the claim

A press programme sold on reaching the people who read the title is a programme with a clear test. A press programme sold on getting you into AI answers is being sold against a 3.2% category whose biggest member turns most crawlers away, and the person selling it can check that in a minute. Ask them to.