Discussion
  • AI Search
  • News SEO

Preferred sources now reach AI Mode and AI Overviews, and there’s an embeddable button. Have you seen measurable lift after promoting it?

Published 1 reply

Discussion starter: a publisher and community SEO question, posted with an answer from Niraj Raut to open the thread. If you have dealt with this on a site, reply with what you saw, especially where it differs.

Disclosure: Niraj Raut, who posted this answer, runs the SEO consultancy linked at the end of it.

1 reply

  • admin
    Admin
    Original poster

    No public controlled test shows site-level lift from promoting the preferred sources button, and I would be sceptical of anyone claiming a clean number. Google’s own figure is that readers are about twice as likely to click through to a source they have marked as preferred. That is a statement about the users who chose you, not about what promoting the button adds. Search Console has no preferred-source dimension, so most “lift” reports you will see are before and after comparisons muddied by the news cycle, core updates and AI feature changes.

    The important distinction is between the per-user effect and the site-level effect. The per-user effect is plausibly real. The site-level effect depends on how many readers complete the selection and how often they then search for topics where you are eligible to appear. For most publishers that number is small relative to total Google traffic, which means the lift may exist and still sit inside normal daily noise. The decision should rest on cost and placement trade-offs first, and on a test designed for a small effect second.

    What preferred source status actually changes

    Google’s preferred sources documentation sets out the mechanics, and a few details matter for measurement:

    • It is per user. A reader who selects you changes their own results. Nobody else’s Top Stories or AI answers change.
    • It works at domain or subdomain level. www.example.com and news.example.com are eligible. A subdirectory such as example.com/news is not. A brand that runs its newsroom in a subfolder of a corporate domain gets the whole domain preferred or nothing.
    • Surfaces: Top Stories in all languages where Google Search operates, and AI Mode and AI Overviews wherever those features exist. The AI surfaces were added in May 2026 and the embeddable button followed in August 2026, per the TechCrunch coverage.
    • AI surfaces have an extra condition. The documentation says the site needs to be included in Search generative AI features in Search Console. If you are weighing the AI features opt-out control (Search generative AI control), which Google first tested with UK publishers and rolled out to all websites worldwide on 31 August 2026, note that Google says sites that opt out “will not receive traffic or impressions from our generative AI features”. Check how the two interact before assuming the preference carries into AI answers.
    • The wording is “more likely to appear”. Google does not claim a ranking change in general web results, and the button is described as optional. Readers can also add you through the preferences page directly.

    Why “twice as likely to click” is not the lift you will see

    The users who select a source already like it. They would click you more often than average whether or not the preference existed. So the twice-as-likely figure mixes the effect of the feature with the loyalty that caused the selection. It is useful as a signal that the feature changes behaviour, but it is not an estimate of incremental clicks.

    The button makes this selection bias stronger, not weaker. It sits on your own pages, so the people who see it are people who already reached you. Many of them search your brand or come back directly. The incremental value only shows up on non-branded news and AI queries where, without the preference, they would have clicked another publisher or not seen you at all.

    Hypothetical example:

    A regional publisher gets 3 million Google sessions a month. After adding the button to article templates, it records 9,000 button clicks in the first month. Suppose half of those complete the selection, and each of those 4,500 readers runs 10 news searches a month where the publisher is eligible. If preference adds one extra click for every 20 of those searches, that is roughly 2,250 extra sessions a month, under 0.1% of Google traffic. Those numbers are invented to show the arithmetic, but the shape is typical: even a genuine effect is far smaller than normal day-to-day variation in news traffic.

    That arithmetic is the reason aggregate before and after charts rarely prove anything here. A spike the week after launch is more likely to be a big story than the button.

    What you can and cannot observe in the funnel

    Reader sees the button on your page (observable: impressions of the element)
    ↓
    Reader clicks it (observable: click event in GA4 or your analytics)
    ↓
    Reader completes the selection with Google (not observable to you)
    ↓
    Reader later searches a topic where you are eligible, and Google shows you in Top Stories, AI Mode or an AI Overview (not observable per user)
    ↓
    Reader clicks through (observable only as google / organic, mixed with everything else)

    The two middle steps are the ones that matter, and neither is reported to publishers. Google’s documentation does not mention any Search Console reporting for preferred sources. TechCrunch reported over 345,000 unique sources selected as of May 2026, which tells you the feature is used widely across the long tail but says nothing about your own count.

    Track button clicks as an event anyway. A click is intent, not a confirmed selection, but it gives you the upper bound on how many readers could be affected. If that number is tiny, you already know no aggregate test will detect anything.

    Test designs that can detect a small effect

    Preference applies to the whole domain, so a page-level A/B test does not isolate it. A reader who adds you from a sport article gets you preferred for politics too. These designs work better:

    Design What it needs Strength Weakness
    Logged-in cohort comparison First-party user IDs (subscribers or registered readers) and button click events tied to those IDs Compares Google-referred visit frequency before and after for clickers against matched non-clickers who saw the button. Controls for baseline loyalty. Clickers are more engaged, so their trend may differ anyway. Only covers logged-in readers.
    Country holdout Meaningful Google traffic in two or more countries, and the ability to geo-target the button Show the button in one country and not another, then compare Search Console clicks and impressions by country on non-branded news queries. News cycles differ by country. Needs enough volume in both.
    Interrupted time series on a query segment Search Console query exports, a clear launch date, and a stable query set Cheapest. Watch impressions and CTR on non-branded news queries, plus AI impressions in the generative AI performance report. Weakest causally. Core updates, story mix and AI feature changes all confound it.

    For the time-series option, the Search generative AI performance reports are worth including. They show impressions only (no clicks or queries) for AI Overviews and AI Mode by page, country, device and date, and reached all eligible sites on 31 August 2026. If the preference makes you more likely to appear in AI answers for your readers, AI impressions are where it could show up first. Expect the signal to be faint.

    Whichever design you use, write down the hypothesis and the minimum effect you care about before launch. Run it for at least six to eight weeks, avoid launching inside a core update rollout, and annotate the launch in Search Console so the date is not lost later.

    What promoting it costs

    The button itself is cheap. The costs are elsewhere:

    • Placement competes with CTAs that have measurable value. An article footer slot used for the preferred sources button is a slot not used for newsletter sign-up, registration or subscription. Those convert to first-party relationships you own and can measure. This is the real trade-off for most publishers.
    • The recommended implementation loads a Google script (publisher.js from news.google.com). Check it against your consent setup and page performance budget. The documented deeplink format (google.com/preferences/source?q=yourdomain) gives you a plain link with no third-party script, which suits newsletters, social posts and lightweight templates.
    • Prominent placements on every article add clutter for readers who will never click it. Test placement frequency, not just presence.

    Where the upside is larger:

    Publishers with a loyal readership that also competes on high-volume, non-branded news topics have the most to gain. Their readers already value them, and the preference helps them win clicks against larger outlets in Top Stories and AI answers. Niche publishers whose readers mostly arrive by brand search or directly have less incremental room.

    What result would change the decision

    • If button clicks are below a few thousand a month, treat the button as low-cost hygiene. Put it in low-value slots and the newsletter, and stop trying to measure lift.
    • If the logged-in cohort test shows clickers increasing Google-referred visits faster than matched non-clickers over two months, you have a defensible case for better placement.
    • If a country holdout shows no difference in non-branded news impressions or clicks after eight weeks with adequate volume, do not give the button placements that currently drive subscriptions or registrations.
    • If AI impressions for your key sections rise in the treated period without a matching rise in the control, that is the most direct evidence the preference is reaching AI Mode and AI Overviews for your readers.

    If this were my site, I would add the deeplink to newsletters and the button to one low-competition template slot, track clicks from day one, and set up the cohort comparison if we have logins. I would only move it into prime placements once the cohort data shows Google-referred visits rising for readers who clicked it, because until then the button is competing for space with CTAs whose value is already proven.

    Need help with this on your own site?

    Niraj Raut works with businesses in Nepal, Australia, the UK, Europe and the US on technical SEO, ecommerce SEO, local SEO and AI search.

    See SEO services

Add a reply

Replies are open to approved members. Log in or apply to join.