Foundations

Google will let your readers pin you into their AI answers. It does nothing for strangers.

By Arnav Mukherjee, founder of TofuBofu · September 2, 2026

The entire implementation is two lines of HTML. A script tag from news.google.com, and an empty div. Add them and Google renders a button that lets a reader mark your site as a preferred source, after which Google says your content can be highlighted with a "preferred" badge inside AI Mode and AI Overviews for that person, until they change their mind.

I went to check whether our own domain was eligible before writing this. Google's documentation says the proof is whether your site turns up in the source preferences tool, so I opened the tool. It redirected to a sign-in page. That is the shape of this whole feature in one click: Google will tell you plenty about it, and the only thing that settles it is a person with an account making a choice.

Which is also why it is the most interesting thing Google has shipped for AI visibility this year, and why almost nobody selling AEO will describe it accurately. Every tactic in this category is about persuading the engine. This one is about being chosen by the person. That makes it genuinely new, and it also puts a hard ceiling on it that the people writing "add this to your site now" posts this week are not going to mention.

What Google actually shipped, in its own words

Preferred Sources already existed for Top Stories. On 27 May 2026, Duncan Osborn, a Product Manager on Google Search, wrote that "starting today, it's coming to AI Overviews and AI Mode, so you'll be able to easily spot links in AI responses from the sources you've already selected". The implementation documentation puts it as the mechanic: "In AI Mode and AI Overviews, your content can be highlighted with a 'preferred' badge for users who have selected your site as a preferred source."

On eligibility, Google is about as broad as Google ever is: "Any website that publishes fresh content is eligible." That sentence is the one that decides whether a managed IT provider or a B2B software company can play at all, and it is Google's, not an inference from a summary. I could not verify it for a specific domain, for the reason above, and I am not going to upgrade Google's claim into a measurement. Go and type your domain into the tool. It takes ten seconds and it is the only real answer.

Google gives two numbers with it: "people are twice as likely to click through to a Preferred Source, and people have already selected more than 345,000 unique sources". Both are Google's own, published in a product post rather than a study, with no denominator for the 2x and no window on either. I am quoting them because they are the only figures anyone has, and I am telling you what they are worth. The 345,000 counts distinct sources that somebody somewhere picked. It is not a count of people, and it is certainly not a count of anyone picking you.

One more line from the documentation, because it is the one that keeps this honest: "It's not required to do them in order to appear as a preferred source." The button is a prompt, not a registration. You are not adding yourself to an index. You are making it one click easier for somebody who already reads you.

The part that is genuinely new: an audience becomes an AI visibility asset

Think about what every other lever in this category requires. Schema, so a parser can read you. Comparison pages, so an engine has something to quote when someone asks who else does this. Review profiles, so third parties vouch for you. Crawlability, so the thing can reach your pages at all. Every one of them is an argument aimed at software.

Preferred Sources is aimed at a human being, and it converts assets a B2B firm usually cannot use for AI visibility at all. A newsletter list. Customers who already log in. A community, a podcast audience, a conference talk, the four hundred people who opened your last update. None of that has ever moved an AI answer before. Now, one click each, it moves theirs.

And it is cheap in a way almost nothing in this field is. Two lines of HTML, no schema audit, no content programme, no crawl budget argument, no waiting for an engine to re-index anything. If you have an audience and a head tag, you are done in an afternoon.

And the ceiling, which is the half that will not be in the other posts

Read the mechanic again: the badge is for users who have selected your site. So it changes the answers of people who already know you, already rate you, and already went to the trouble of picking you out of the web. It does exactly nothing for the buyer who opens an AI engine and types the name of your category.

That buyer is the entire reason this field exists. They are three weeks into a purchase you do not know about, they are asking an engine which vendors to shortlist, and they have selected nobody, because they do not yet know there is anybody to select. Preferred Sources is a loyalty lever wearing AI clothing. It rewards an audience you already earned. It cannot introduce you.

There are two smaller limits worth knowing before you plan around it. It is Google only: Top Stories, AI Overviews and AI Mode. Of the six engines we query on every scan, that is one. ChatGPT, Claude, Perplexity, Bing Copilot and Gemini are untouched, and no Search setting reaches them. Gemini is Google's own model and this is still not a lever on it.

And it is domain or subdomain only. Google spells this out: "the subdirectory https://www.example.com/blog isn't eligible". If your writing lives at yourfirm.com/blog, you are asking a reader to prefer your whole company, pricing page included. If it lives at blog.yourfirm.com, the publication can be preferred on its own. Nobody chose their blog's URL structure with this in mind, and it now has a consequence.

What our own data says about the row this does not touch

We pulled every citation out of our own stored AI answers and counted them. Across 882 distinct citation sources gathered from 55 scans, the number of times an engine answering a buying question in a company's own category cited that company's own domain was 14. One point six percent.

Three caveats travel with that figure and they are not fine print. It counts two engines, Google AI Mode and Bing Copilot, because those two attach a reference list we can store. Eleven of the 55 scans produced any citation at all, and four of those supply 95.8% of the sources, so this is a concentrated sample. And it is a floor for firms that already suspected they were invisible, not an industry average, because that is who runs a first scan.

Even read that conservatively, the shape is unmistakable. When an engine describes you to a stranger, it is overwhelmingly reading somebody else's page about you. Preferred Sources acts on the 1.6%, in front of readers who already chose you. It is the other 98.4% that decides whether a buyer who has never heard your name ends up on a shortlist.

What to actually do this week

One. Check eligibility yourself. Sign in to Google, open the source preferences tool, type your domain into its search box. Google's documentation names that as the test. Ten seconds, and it beats every blog post guessing on your behalf.

Two. If you are there, add the two lines. Script tag in the head, div where you want the button. Put it where a reader who already likes you will see it: the end of a post, the footer of your newsletter, the thank-you page after a download. Not your homepage hero, where it will be seen by people who have no idea who you are and will not click it.

Three. If your CMS cannot run the script, use Google's deeplink instead. It works in a text link, an image, an email or a social post, which is arguably a better fit anyway, because the newsletter is where your actual audience is.

Four, and this is the one that matters. Go and find out what an engine says about your category to somebody who has never heard of you. That answer is unaffected by anything above, it is the one your next customer is reading, and most firms have never seen it once.

See the answer a stranger gets. We ask six AI engines the buying questions your category actually types, and show you who they name instead of you.

Run a free AI visibility scan

The judgement, and it is a judgement

Do it. It costs an afternoon and it is the only mechanism in this field where your existing goodwill converts directly into AI visibility. If you have a list, a community or customers who like you, that is an asset which until now did precisely nothing for how an engine answers, and now does something.

Then be clear about what you bought. You made yourself easier to prefer, on one engine, for people who already prefer you. You did not become findable. Anyone selling this to you as an AI visibility strategy is selling you a retention tactic with the wrong label on it.

The position we have held since we started still stands, and this feature is a good illustration of it rather than an exception to it. Search rank is the floor. AI visibility is a distinct layer you can win even when your Google rank does not move, and it is decided by what the engines find when a stranger asks. Preferred Sources sits on top of an audience you already have. Everything that matters happens underneath it.

Frequently asked questions

What are Google Preferred Sources?

Preferred Sources is a Google Search setting where a person chooses the websites they want to see more of. Google brought it to its AI surfaces on 27 May 2026, writing that it is "coming to AI Overviews and AI Mode, so you'll be able to easily spot links in AI responses from the sources you've already selected". Google's documentation states that "In AI Mode and AI Overviews, your content can be highlighted with a 'preferred' badge for users who have selected your site as a preferred source". The important word is selected. Nothing happens until a specific human picks you, and the effect exists only in that person's results.

Is a B2B site eligible to be a preferred source?

Google's announcement says "Any website that publishes fresh content is eligible", which is as broad a statement as Google makes about anything. We could not independently confirm it for a specific B2B domain, and we will not pretend otherwise: Google's documentation says the proof is whether your site appears in the source preferences tool, and that tool requires a signed-in Google account, which we confirmed on 2 September 2026. So the honest answer is that Google says yes and the check takes you ten seconds. Sign in, open the source preferences tool, and type your domain into its search box.

How do I add the Preferred Sources button to my site?

Two lines of HTML, and Google publishes both. A script tag, preferably in the head: <script async src="https://news.google.com/swg/js/v1/publisher.js"></script>. Then a div wherever you want the button to appear: <div google-add-preferred-source-btn></div>. Google renders a localised, styled button from that, with data-theme and data-lang attributes if you want to override the look or the language. If your CMS cannot run the script, Google documents a deeplink alternative you can put in a text link, an email newsletter or a social post: https://www.google.com/preferences/source?q=example.com with your own domain. Google also states that neither is required: "It's not required to do them in order to appear as a preferred source."

Does Preferred Sources help you get cited by AI when someone has never heard of you?

No, and this is the whole limit of the mechanism. Google's badge is for users who have already selected your site. A buyer who does not know your name has selected nobody, so your answer to them is unchanged. That buyer is the entire problem AI visibility work exists to solve. Preferred Sources is a loyalty lever wearing AI clothing: real, cheap, worth doing, and structurally incapable of introducing you to a stranger asking an engine which vendor to shortlist.

Which AI engines does Preferred Sources affect?

Google's own, and only Google's own. The announcement and the documentation name Top Stories, AI Overviews and AI Mode. Of the six engines we query on every scan, that is one: Google AI Mode. ChatGPT, Claude, Perplexity, Bing Copilot and Gemini are outside it entirely, and no Search setting reaches them, Gemini included, even though Gemini is Google's own model. If you are treating this as an AI visibility strategy rather than as one Google feature, you are planning around a sixth of the surface.

Can I make one page or my blog a preferred source?

No. Google is explicit that eligibility is domain and subdomain only: "Only domain-level and subdomain-level sites are eligible to appear in the source preferences tool. For example, https://www.example.com/ and https://code.example.com/ are eligible for preferred sources, but the subdirectory https://www.example.com/blog isn't eligible." So a firm whose writing lives at yourfirm.com/blog is asking readers to prefer the whole company domain, marketing pages included. A firm that publishes at blog.yourfirm.com can be preferred as the publication on its own. That is a hosting decision most B2B firms made years ago for unrelated reasons, and it now has a consequence.

How many people actually use Preferred Sources?

Google says "people have already selected more than 345,000 unique sources" and that "people are twice as likely to click through to a Preferred Source". Both figures are Google's own, published in a product announcement rather than a study, with no denominator for the 2x and no window given for either. Note what the 345,000 counts: distinct sources selected, not people doing the selecting, and not selections of you. Treat it as evidence the feature is being used at all, which is genuinely useful, and not as a size estimate for your own audience.

Sources and further reading

Google, "New ways to find your favorite sources and original content in AI Search", by Duncan Osborn, Product Manager, Google Search, 27 May 2026. The announcement bringing Preferred Sources to AI Overviews and AI Mode, and the source of the 2x click-through and 345,000 figures.
Google Search Central, "Help your readers find your site through preferred sources". The implementation documentation: the badge in AI Mode and AI Overviews, the domain and subdomain eligibility rule, the two HTML lines and the deeplink alternative.
Which websites do AI engines actually cite? We counted 882 of them. Our own citation count, including the 14 of 882 own-domain figure used above and the sampling caveats that go with it.