Google Preferred Sources: how to activate and measure reader adoption

A reader may value your publication without knowing how to find it again on Google. Preferred Sources provides a way to express that preference. For an editorial team, the useful task is not promising better rankings: it is making a voluntary choice easier and establishing which parts of the journey can be measured.

This guide is for audience, newsletter and editorial product teams. You will finish with a chosen implementation, a test matrix and a record that separates invitations, clicks, confirmed preferences and subsequent exposure. The examples are hypothetical; the protocol is a Mentio proposal.

If your problem is getting Google to cite a page that a reader does not yet know, start with the general AI Overviews guide. Here we focus on the relationship between a publication and its audience, not a ranking recipe.

What changed with the August 20 button

On August 20, 2026, Google announced an interactive button that helps publishers facilitate source selection and return readers to their page. The announcement connects it with Top Stories, AI Overviews and AI Mode. A current guide should therefore cover more than the link that opens preferences. Source: Google's official announcement.

That change reduces integration work for some websites, but it does not remove your operational questions. Where is the invitation displayed? What happens without a signed-in session? How does the reader return? Which event can your analytics observe? Define those answers before reporting a business outcome.

Separate reader value from the company's hoped-for benefit. “Find our publication more easily” describes a utility. “Click to make us number one” turns a personal preference into a promise you have not demonstrated.

Check the host before designing the campaign

Google's guide covers domains and subdomains, not folders, and first asks you to check whether the site appears in the selector. Top Stories availability is global; availability in AI experiences follows those features' supported locations. Source: Google Search Central.

Write down the exact host representing the publication. If your content lives at example.com/magazine, do not design a campaign as though that folder were an independent source. Do not migrate content to a subdomain merely to force an option either: that would be a much larger architecture change than this invitation.

In Mentio's check on September 7, 2026, mentio.tech appeared as an option in the selector. The checkbox was not selected and the account was not changed. That evidence establishes availability in the inspected session, not a saved preference, an audience count or universal availability.

If the host is missing, retain the date, URL and visible state. Do not display a success message on your website. Google Help notes that sources without regular updates may be unavailable; that does not establish the cause of every missing source. Source: Preferred Sources help.

Choose the interactive button or a deeplink

The documented standard integration uses these two elements; the second determines where the button appears. This is a code example for your team, not an active button inside this guide. Source and reference code: Search Central.

<script async src="https://news.google.com/swg/js/v1/publisher.js"></script>
<div google-add-preferred-source-btn data-lang="en"></div>

If your CMS or channel cannot support that integration, the alternative format is https://www.google.com/preferences/source?q=example.com. Replace the domain and verify the destination. The link leads to a selection the reader must confirm; it is not an automatic subscription URL. Source: deeplink implementation.

Start with one placement and a technical owner. Avoid loading the library repeatedly or introducing a global widget without reviewing performance, website policies and mobile behavior. Do not invent SDK methods: use only those documented for the version you implement.

Offer the invitation after delivering value

A campaign for returning readers does not need to interrupt every visit. Test a discreet invitation after a useful article or in an editorial newsletter. Define your own frequency cap and a way to dismiss the message; persistence is not the same as interest.

The copy should name the action and destination. For example: “Choose our publication as a preferred source on Google.” Do not make selection a condition for reading, getting support or accessing a purchase already made. Avoid rewarding artificial saves or asking people who do not read the website to inflate the signal.

Do not combine a paid subscription, newsletter signup and Google preference into one invitation. They are different choices. If a reader cannot explain what they accepted, the campaign is not ready, even when the button works.

Keep Digital PR strategy separate too: earning external coverage and facilitating choices by your own readers have different audiences, owners and outcomes.

Test the journey with a state matrix

Google Help lets users manage preferences through Search personalization; settings saved while signed in remain consistent across browsers using that account. Do not assume one session describes all the others. Source: user controls.

Use authorized accounts. For an inspection without permission to change preferences, stop before selection. To test a save, record the previous state and final evidence in the test account. The team must not mark a step as passed when it deliberately did not execute it.

Also check keyboard navigation, control names, contrast and the absence of overlaps. If a custom integration fails, first reduce it to the supported example; do not add JavaScript layers to conceal the actual state.

Measure a funnel without inventing saved preferences

Design the dictionary before the dashboard. The following names are internal examples, not official Google events. Every signal needs a definition, a denominator and a limitation.

When using a third-party button, do not assume your click detector can observe its internal interaction. Inspect the actual integration and report “unavailable” when a signal is missing. Do not read private content from another window or try to infer the reader's account either.

Calculate activation rates only with compatible units. Clicked sessions divided by exposed sessions can be useful if both are deduplicated by session and placement. Dividing all newsletter clicks by impressions of a website block produces a number without a coherent meaning.

Example: 96 clicks are not 96 new preferred readers

Suppose a hypothetical test exposes 2,400 sessions to an invitation and 96 sessions activate the deeplink. The observed activation rate is 96 / 2,400 = 4%. That figure describes this placement, period and counting method.

You have no confirmation of saves on Google. The saved-preferences field must therefore say “unavailable.” Do not fill it with 96, zero or an estimate presented as fact. If five authorized participants complete an observed test, you have verified five test journeys, not a 100% conversion rate for the audience.

To compare two messages, define the assignment unit, period and decision rule beforehand. A variant with more clicks and more observed abandonment may explain the action less clearly. Do not select a winner only because its number looks better; document questions, errors and maintenance cost too.

Separate adoption, exposure and ranking

A later improvement in Search Console may coincide with new coverage, content changes, seasonality or a change in Google's product. Do not attribute the entire difference to this campaign without a design that can test that explanation.

Maintain three series: the invitation you control, exposure reported by Google and answers observed by your measurement system. The Search Console generative AI report guide explains that second layer. Do not rename its impressions “readers who selected us.”

Measuring brand mentions in AI Overviews answers a different question: what appears for specific queries. A benchmark using an account that already prefers your source must not be pooled without labeling with one using a different session.

An honest report can conclude: “The invitation works technically and receives activations; we do not have audience-wide saved-preference counts and have not established a causal effect in search.” That supports a decision without turning an observation limit into a commercial promise.

Keep a record another person can review

Retain a record for each campaign version. The following example is a template, not a Mentio result:

campaign: preferred-source-pilot-01
host: example.com
placement: article-end
implementation: deeplink
copy_version: en-v1
period: defined-before-launch
unit: exposed-session
audience_wide_confirmation: unavailable
owner: audience-team
decision: awaiting-review

Before expanding, require a checked host, a tested journey, a clear invitation, audited events and someone able to remove the integration if it fails. Retain previous versions to explain differences; do not simultaneously change copy, placement and counting method if you later want to attribute a result to one of them.

Frequently asked questions

Does Preferred Sources guarantee an appearance in AI Mode?

No. A reader preference is not a guarantee of appearance, position or traffic. Evaluate invitation activation separately from what you subsequently observe in search experiences.

Can I use a folder as an independent source?

Google's documentation covers domains and subdomains, not subdirectories. Check the host in the selector before launching the campaign, and do not assume a language folder can have its own preference.

Is the JavaScript button mandatory?

No. You can use the deeplink when the button does not fit your CMS or channel. Check the actual flow and choose an implementation you can maintain; installing an invitation does not itself make a website eligible.

Does a link click count as a new saved preference?

No. Record that click as the start of the journey. The person may leave or may already have selected the source. Without verifiable confirmation, new preferences must be reported as unavailable, not zero and not the click total.

How can I test without changing a personal account?

Inspect the destination without selecting anything, or use an authorized test account. Record the initial state and the steps. Testing an actual save requires explicit permission to change that account and evidence of the outcome.

Does an AI tracker measure how many readers selected us?

Do not assume so. A tracker observes answers under its own conditions, not each reader's private preferences. Treat its data as an independent series and require a specific source before claiming it counts saved preferences.

Connect reader relationships and visibility

The immediate action is small: verify the host, choose an invitation and test its journey. The learning should be equally concrete: who saw the offer, who activated it and which part of the outcome remains unobserved.

Explore Mentio plans to measure your brand's presence in AI answers. Combine that signal with editorial analytics without presenting Mentio as an automatic Preferred Sources counter or a guarantee of recommendations on Google.

Want to know if AI mentions your brand?

Discover your visibility in ChatGPT, Claude and Gemini in minutes.

Related articles