Google’s link expansion is a confession dressed as a feature

Google announced five new link surfaces inside AI Overviews and AI Mode this week. Inline citations now sit beside the claims they support. There's a new "Explore new angles" module at the end of long responses. More links, in more places, more visible.

Read the announcement and you'd think this was a generosity play. A thoughtful product team responding to publisher feedback. Hema Budaraju, VP of Product Management for Search, framed it as helping users dig deeper.

Read it alongside the last two years of Google's public language about clicks, and it looks like something else entirely. A company that has cycled through three separate explanations for why publisher traffic is down, and has now quietly skipped to a fourth: stop arguing about clicks, ship more link surfaces, change the subject.

That's worth saying plainly, because the SEO industry is about to spend a fortnight writing optimisation guides for the new link slots. Most of those guides will miss the more interesting story, which is what Google's communications strategy tells you about what Google actually knows.

The four phases of Google's click defence

Search Engine Journal's Matt G. Southern traced the rhetorical evolution this week, and it's worth sitting with the sequence because it's almost too tidy.

Phase one, May 2024 to May 2025: *no data to share*. AI Overviews launches in the US. Publishers immediately report traffic collapses. Pew tracks 68,000 queries and finds users click results 8% of the time when AI Overviews appear, versus 15% without. Asked for counter-evidence at Google Marketing Live, Google representatives say they have none.

Phase two, late 2025: *the clicks that remain are higher quality*. By now the publisher data is harder to dismiss. DMG Media tells the UK CMA that CTRs dropped up to 89% on certain queries. Digital Content Next measures a 10% median year-over-year decline. Google's response: the lost traffic was low-value anyway. No data accompanying the claim.

Phase three, October 2025: *bounce clicks*. Liz Reid gives the argument a name in the Wall Street Journal. The clicks AI Overviews replaced were users who arrived at a page and bounced. Removing them makes the rest look healthier. Repeated on Bloomberg. No supporting data attached.

Phase four, this week: *here are more links*. Five new link surfaces announced, with no new click data.

Each phase replaced the previous one. Each was offered without supporting evidence. Each became inconvenient and was quietly abandoned. The bounce-clicks framing in particular took a substantial hit in February when a randomised field experiment removed AI Overviews from a subset of queries and found organic clicks rose 38% with no measurable change in user satisfaction. If the lost clicks really were low-value bounces, you'd expect satisfaction to suffer when they returned. It didn't.

That's the kind of finding that breaks an argument. So the argument got changed.

The pattern is the message. When a company cycles through four explanations for the same phenomenon in two years, the explanations are not the point. The pattern is.

What "more links" actually does

Let's be fair about the new features. Inline citations next to the claim they support is genuinely better UX than the old footer-cluster of source links. Proximity probably does increase click intent on queries where the AI response doesn't fully satisfy the user. The "Explore new angles" module gives publishers a click surface for adjacent intent that didn't exist before.

A product team confident in its impact data leads with the data. A product team that needs to change the subject leads with the feature.

These are not nothing. If you run a publisher, you should care about how your content appears in those slots, because they're real visibility — visibility you weren't getting last month.

But notice what the announcement carefully does not include. No new click-through data. No segmentation by query type. No publisher-side traffic numbers. No update to the "bounce clicks" framing. No engagement with the February field experiment. Nothing that would let an outside observer verify whether the new link surfaces actually move clicks, or whether they're cosmetic placement of links nobody's pressing.

That's the tell. If Google had data showing the new placement materially increases click-through from AI Overviews, that data would be in the announcement. The announcement would be the data, with the link features as the artefact. Instead, the link features are the announcement, and the data is absent.

A product team confident in its impact data leads with the data. A product team that needs to change the subject leads with the feature.

The actual problem the rhetoric is dancing around

Here's what nobody at Google can say out loud. AI Overviews and AI Mode are not designed to maximise outbound clicks. They're designed to satisfy the query inside the Google surface. When the AI response fully answers the question, the click is structurally redundant — the user already has what they came for. No amount of link placement reshuffling changes that. You can put the link inline, in a footer, in a popup, in a "related angles" module, and if the user's question is answered, the link doesn't get clicked.

This is not a bug. It's the entire point of generative search. The whole product thesis is that synthesised answers are more useful than ten blue links. Google is competing with ChatGPT, and ChatGPT does not send traffic to publishers either. If Google preserved the old click-through economics, they'd lose the AI search race to a product that doesn't have to.

The honest version of Google's position would go something like: *Yes, AI Overviews reduce outbound clicks for fully satisfied queries. That's the product. We're trying to design link surfaces that capture click intent on the queries where users want more, which is a smaller set than before. Publisher traffic from Google Search will, in aggregate, decline. We're sorry. We can't do anything about it without losing to ChatGPT.*

Nobody at Google is going to say that. The antitrust optics alone make it impossible — Penske Media's federal court memorandum in February already argued Google had "shattered the longstanding bargain" between publishers and the search engine, and a public statement matching that framing would land as legal evidence. So instead we get a rolling sequence of euphemisms. *No data. Higher quality. Bounce clicks. More link surfaces.*

Each one buys six months. Each one gets quietly retired when the data catches up. The sequence will continue.

What this means for how you should read Google's communications

This is the part that matters for anyone running an actual business that depends on search.

Google's public language about AI search performance is now an unreliable source for planning purposes. Not because Google is lying — I don't think they are, exactly — but because the language is structured around legal exposure and competitive positioning, not around helping you forecast traffic. The signal-to-noise ratio is low and falling.

Treat Google's public framings the way you'd treat a company's investor relations script. Useful for understanding what they want you to believe. Useless for understanding what's actually happening. The actual happening lives in independent data: Pew, Chartbeat, Ahrefs, the Penske filing, the field experiments. Calibrate against those.

When the next phase of Google's rhetoric appears — and it will, probably within six months — the question to ask is not *is this true?* but *what's the data this is meant to deflect from?* That's the more reliable read.

For most of the businesses I work with, the practical implication is simple. Don't plan around Google's framing. Plan around the independent data on traffic decline by publisher size, query intent type, and category — and adjust your investment in Google Search accordingly. For commercial-intent queries, Search is still the most valuable surface on the internet. For informational content, it's a declining asset, and pretending otherwise because Google's PR is reassuring is how businesses end up over-invested in a channel that's quietly contracting under them.

The deeper thing the link expansion exposes

Step back from the announcement and look at what it implies about Google's internal model.

Google's product team does not believe its own communications team's bounce-clicks framing. If they did, they wouldn't be redesigning link surfaces.

If Google believed AI Overviews were already sending high-quality clicks to publishers, they wouldn't need to ship more link surfaces. The bounce-clicks argument, if it held, would be sufficient — *the clicks you're getting are the clicks worth having, and the rest were noise*. Case closed. Move on.

The fact that Google has shipped five new link features anyway is an admission that the bounce-clicks argument doesn't hold internally either. Somebody inside Google knows the click suppression is real, knows it's becoming a regulatory problem, and knows the previous rhetorical positions are not durable. So they're trying a different tack: make the link surfaces more prominent, hope publishers feel placated, hope the conversation moves on.

This is the most useful thing the announcement tells us. Not the features themselves, but what their existence implies. Google's product team does not believe its own communications team's bounce-clicks framing. If they did, they wouldn't be redesigning link surfaces.

That's a useful tell. The next time Google publishes a confident-sounding framing of AI search performance, ask yourself: does Google's product roadmap behave as if this framing is true? Often the answer is no, and that gap is where the real story lives.

Where I'd actually push back on this read

Two honest concessions.

First, link surface design genuinely is hard, and it's possible the inline citation work and the "Explore new angles" module are the result of patient UX iteration that started long before the bounce-clicks rhetoric. Not every product decision is a strategic deflection. Some of it is just a team shipping the thing they've been working on for a year. I've assigned more intent to the timing than the timing alone can support.

Second, more links visible *is* better than fewer links visible, even without new click data. If Google had announced the same features with click-through numbers attached and the numbers were modest, the SEO industry would still treat the placement as a positive development. The absence of data isn't proof the data is bad. It's just the absence of data.

Both of those caveats matter. They don't change the broader read — Google's public language about AI search is increasingly unreliable for planning — but they do mean the link expansion itself is a real product change, not pure theatre. Optimise for it. Just don't mistake it for evidence that the click economics have shifted.

The thing to actually do

Three things.

Audit which of your pages are showing up in inline citations versus footer citations versus the "Explore new angles" module. The placements have different click economics, and you want to know which surfaces your content lives on. This is straightforward log-file and SERP-tracking work, and most agencies haven't started yet.

Stop treating Google's public framings as planning inputs. Build your traffic forecasts from independent data. The Pew, Chartbeat, and Ahrefs numbers are reliable enough to use as base rates. Adjust them for your category and intent mix.

Watch the next rhetorical phase. It's coming. When it lands, the test is whether Google's product roadmap behaves as if the new framing is true. If it does, take the framing seriously. If it doesn't — if Google keeps shipping features that imply the framing is wrong — you've got your answer.

The link expansion is not a gift to publishers. It's not an attack on publishers either. It's a company managing a difficult situation by changing the subject, and the subject will keep changing for as long as the underlying numbers refuse to cooperate. That's the loop. The features are real. The framing isn't. Plan accordingly.