In plain English
AI Overviews and AI Mode are Google's AI-written answers, shown with or above the usual list of links. An opt-out lets a publisher refuse to have its content used for them. A default is the setting you get unless you change it, like the search engine a phone or browser comes with. robots.txt is a file on a website that tells automated crawlers which pages they may read, and the new shared signal is a line in that file that tells AI companies not to train on the site.
How would you feel if this happened to you: you run a small bike repair shop, and you spend a weekend writing a guide to fixing a slow puncture for your website. A month later, people who search for that fix see Google's AI summary of your guide at the top of the page, and they may not need to visit your site. Since 31 August a setting in Google Search Console can keep your guide out of that summary, and Google says it won't affect your place in the ordinary results. But Google uses your guide in the summary unless you find the setting and change it, and the setting covers only Google's search summaries. Keeping your guide out of the data AI companies train on is done separately, with a robots.txt rule for each company's crawler, and a shared rule for all of them won't work until 2027.
On 8 October, Turkey's competition authority, the Rekabet Kurumu, ordered Google to set up conditions that do not discriminate between news publishers, reflect the economic value of their content, and let publishers decide whether it is used in AI Overviews and AI Mode independently of their visibility and indexing in Google's other services. On 2 November it announced an investigation into Google's AI Overviews and AI Mode, following a preliminary inquiry that began on 27 August. Google's AI summary appears above the search results, and Google writes it from publishers' articles. Google's Search Console setting, live worldwide since 31 August, already lets publishers in Turkey keep their pages out of AI Overviews. The order also requires terms that reflect what the content is worth.
The European Publishers Council described the problem in its February complaint to the European Commission: "In practice, opting out of AI use entails a loss of search visibility that most publishers cannot afford." The Commission had opened a formal antitrust investigation on 9 December 2025, to assess whether Google has breached EU competition rules by using the content of web publishers for AI purposes. In the UK, the Competition and Markets Authority required Google to give publishers effective tools to stop their content being used in AI features in search. Google began testing the Search Console setting with some UK sites on 3 June and rolled it out to every website worldwide on 31 August. Sites that opt out "will not receive traffic or impressions" from Google's AI features, and Google says the setting "will not be used as a ranking signal" for the ordinary results.
On 30 October Google announced that it plans to support a shared AI training control in robots.txt, the file that lists which pages crawlers may read. Microsoft and OpenAI announced the same plan, and the draft standard behind it was written by Kevin Kelley at Anthropic. A site opts out with a group headed User-Agent: __usage_ai-training. A rule written for every crawler, User-Agent: *, doesn't count, so a site that blocks all crawlers today has still not opted out under the new signal. Google will read the opt-out as a block on Google-Extended, its existing training control, which covers training and answers in the Gemini app, not AI Overviews. Google aims to support it in 2027. Neither Google's post nor the draft standard says anything about content already collected or about payment.
My concern is that people aren't going to own their own voice anymore. The flip side is that if publishers don't like Google showing their results, they could just not use Google. But the trouble is, everyone uses Google. Google has its own operating system, and it is the default search engine on many things. So people don't know any better, which makes it even worse, because you have ignorant people, in the best sense of that word, being misinformed.
I don't want to go demonising tech companies. They've got enough working for them and against them.
In October 2025 the European Broadcasting Union published a study in which journalists evaluated more than 3,000 responses from ChatGPT, Copilot, Gemini and Perplexity to news questions. Forty-five percent of the answers had at least one significant issue, and 31 percent showed serious sourcing problems: missing, misleading or incorrect attributions. The study tested chat assistants. AI Overviews were not part of it, so it gives no figures for Google's summaries.
It goes back to critical thinking and verifying sources.
Publishers can build a list of readers who come to them directly, such as newsletter subscribers or paying members. They can make each AI choice on purpose: the Search Console setting for Google's AI summaries, Google-Extended and each AI company's own crawler rules for training, and the shared robots.txt group once it is supported in 2027. A site that changes none of these settings is still used in Google's AI summaries and can still be used for AI training.
Readers can click through from an AI summary to the source it names, and check any number, quote or date that matters against the original. They can subscribe directly to a few sources they trust.
Made-up example for illustration. The village, the people and the article are invented.
Check the summary
A made-up summary, written the way an AI summary reads, beside the story it summarises. Decide whether you trust each claim, then click or tap it to check.
Summary
Click or tap a claim to check it.
-
Correct. The article says exactly this, in the passage now highlighted.
-
Wrong number. The article says £48,000 a year. £31,000 is what the clerk said the council's reserves would fall to.
-
Out of date. The article says the start was first given as 6 April and then moved to 4 May. The summary kept the old date.
-
Wrong source. Councillor Tamsin Okoro said this. Councillor Vane opposed the bus.
-
Correct. The article says the trial will be reviewed in March next year.
Article
Dunmarrow council backs new evening bus route
Dunmarrow Parish Council voted 7 to 2 on Tuesday 10 March to fund a new evening bus service linking the village to Pellbrook Junction station. Matching passage in the article. The route, numbered 14, will run from 6 pm to 11 pm, Monday to Saturday. The start date was first given as 6 April, but Matching passage in the article. the council has moved it to 4 May.
Matching passage in the article. The council will pay £48,000 a year for a two-year trial. Councillor Tamsin Okoro, who proposed the motion, said: Matching passage in the article. "Young people and shift workers have had no way home after six." Councillor Gerald Vane, who opposed it, said the money would be better spent on repairing the Mill Lane footpath. "I don't think enough people will use the bus," he said.
The service replaces the evening route 9, which was cut in 2022. Ticket prices have not been set. Matching passage in the article. The trial will be reviewed in March next year. The funding comes from the council's reserves, which the clerk said would fall to £31,000.
Invented for this page. It is not a real news story.
Nothing checked yet.
2 of 5 claims are correct. 3 had an error: one in a number, one in who said a quote, and one in a date.
This page sends nothing anywhere. Your clicks stay in your browser.
Updated 5 November 2026: Google's Search Console setting for AI Overviews and AI Mode has been available to every website since 31 August and is not used for ordinary ranking, so this post no longer says opting out costs search visibility. Added the shared AI training rule for robots.txt announced on 30 October.