December 2025: Google Core Update Lands, Num=100 Issue Returns & AI Mode Expansion Continues

December had two clear themes. First, Google’s final core update of the year arrived with enough force to end any festive lull fairly quickly. Second, Google kept extending AI Mode into more surfaces, more journeys and more moments where users might otherwise have clicked through and done their own digging.

Over the past few months, we’ve talked a lot about search becoming more interface-led, more synthesised and more dynamic. December pushed that further. Google isn’t just building AI into search results; it’s building AI into the spaces around discovery, browsing and onward exploration too.

December 2025 Core Update rolls out with visible early impact

Google officially began rolling out the December 2025 Core Update on 11 December, with the company saying the rollout would take up to three weeks. As expected, it was framed in familiar language: a regular update intended to better surface relevant, satisfying content from all types of sites.

The interesting bit was the speed at which meaningful volatility showed up. Early signs suggested the update started biting hard within a day or two, with 13 December standing out as a particularly notable day for movement across the SERPs.

This was Google’s third confirmed core update of 2025, following March and June. That alone makes it notable. Google had previously talked about more frequent core updates, but the year never really played out that way. So in practical terms, December felt less like routine maintenance and more like one last broad recalibration before the year closed.

My Take: This one matters less because it was dramatic on day one, and more because of where it lands in the wider sequence. Across the second half of 2025, we’ve seen stronger signals around site quality, tighter pressure on low-value scaling tactics, and a broader shift towards interface-first search experiences. This core update looks consistent with that direction rather than a departure from it.

That’s important commercially. Businesses still hoping for a clean separation between “classic SEO” and everything happening in AI search are reading the room badly. The same broad quality, utility and trust signals increasingly sit underneath both. The packaging changes. The surfaces change. The retrieval and ranking logic evolves. But the idea that you can build thin, search-shaped assets for one environment and somehow remain resilient in the other is woefully outdated.

Also, it’s worth saying plainly: when Google calls something a “regular update”, that doesn’t make it strategically uninteresting. Quite often it’s the opposite. The bland language is a wrapper designed to keep panic at bay and to signal consistency in the direction of travel.

AI Mode starts creeping into Discover and the wider Google app journey

Google added new AI Mode actions to pages opened through Google Discover and, more broadly, within the Google app on Android. Users were shown options including “Summarize with AI Mode”, “Ask a follow up with AI Mode” and “Dive deeper with AI Mode”. In other words, a click from Discover no longer just leads to a webpage. It can now become a springboard back into Google’s own answer layer.

That’s a subtle change on the surface, but strategically it’s not subtle at all. Discover has already become one of Google’s most personalised and behaviourally driven surfaces. Adding AI Mode actions on top gives Google more ways to keep users inside its own ecosystem while still using publisher content as the raw material.

My Take: This is exactly the kind of product move that tells you where Google wants the journey to go. Search, discovery, browsing and follow-up exploration are being stitched together into one increasingly closed loop.

For publishers and brands, that doesn’t mean Discover stops mattering. Quite the opposite. It means the value of winning attention there may increasingly depend on what happens after the first click, not just whether the click happens. If Google keeps turning webpages into hand-off points for AI interactions, then the battleground shifts again: from “how do I get visited?” towards “how useful am I as source material, cited reference, brand impression and conversion destination within a compressed journey?”

The bigger strategic point is that Google seems increasingly interested in owning the full funnel in-platform. We’ve been moving in that direction for a while. December gave us another very clear nudge.

Google tests fine-tuned AI Mode to encourage clicks

Google tested updated links in AI Mode, increasing the number of inline links and adding short contextual introductions explaining why a link might be worth visiting. Alongside that, Google was seen testing very long, expandable AI-generated snippets directly in the main search results.

Taken together, these changes suggest Google is (at least ostensibly) trying to solve a tension it created itself. AI-generated answers are useful because they reduce the need to click. But if they reduce clicking too much, the open web becomes a worse deal for publishers, merchants and creators, which eventually makes the product worse for Google too.

So the latest direction seems to be: keep the synthesis, but make onward journeys feel more intentional, more guided and more justified.

My Take: This isn’t Google generously rediscovering the value of sending traffic your way. It’s Google trying to make a lower-click environment more sustainable.

The contextual intros are especially interesting because they make the referral feel curated rather than incidental. A plain citation says, “here’s a source”. A framed citation says, “here’s why this source is useful right now”. That’s a stronger editorial role for Google inside the click decision and will be read by many users as a recommendation.

For brands, that raises the bar. It’s no longer enough to be merely present. You need to be the kind of result that makes sense as the next step after an answer has already done some of the heavy lifting. In practical terms, that often means sharper page purpose, tighter information, cleaner entity alignment and fewer pages that meander before delivering the thing the user actually wanted.

Gemini 3 Flash becomes the default model in AI Mode

Less than a month after bringing Gemini 3 into AI Mode, Google began rolling out Gemini 3 Flash as the default model globally in AI Mode. The stated pitch was familiar: stronger reasoning, tool use and multimodal capability, delivered at search speed.

From Google’s point of view, that makes sense. Search products live or die on latency/speed. A more capable model only helps if it’s fast enough to feel native to the search experience.

My Take: This is one of those infrastructure stories that sounds technical but has very visible consequences. Better speed plus better reasoning usually means users push the product further. They ask longer questions, add more constraints, trust the answer layer more, and spend less time reformulating clunky prompts.

Every quality improvement in AI Mode increases the chance that more informational and mid-funnel tasks stay inside Google’s environment for longer. The headline is that Gemini 3 Flash is cleverer, but the story is whether it makes AI Mode feel fluid enough to become habit-forming, because once a behaviour becomes habitual, attribution models, content formats and SEO priorities all start bending around it.

Search Console gets messy again as the num=100 saga continues

We previously covered the disruption caused by Google’s changes around the num=100 parameter and automated scraping of search results. In December, some sites saw average position drop while impressions rose sharply, suggesting that bots and automated queries may once again be finding ways through.

Once noisy, synthetic or scraper-driven behaviour starts leaking back into Search Console data, it becomes harder to interpret trends cleanly, harder to benchmark performance, and easier for teams to overreact to ghosts in the machine.

My Take: This is the sort of thing that quietly wrecks decision-making if you let it. Not only because the reporting loses fidelity, but because it becomes easier to mistake measurement artefacts for genuine market movement.Before September, we had a fairly solid baseline. After the num=100 change, we'd hoped we would simply have a new, recalibrated basline. The reality, it seems, is that the background noise keeps changing volume, making real trends harder to isolate.

We’ve said before that modern search reporting needs more scepticism, not less. That’s even more true when Google is changing interfaces, changing query handling, changing AI surfaces and fighting scrapers all at once. Search Console is still valuable, but it’s time to accept it's no longer sacred, "straight from the horses mouth", trustworthy, directional data.

Google’s “digital mulch” line says more than the one-liner suggests

John Mueller shared a line describing a lot of SEO-driven blog content as “digital mulch”, paired with the warning that if a blog exists solely to rank, it’s living on borrowed time. The phrase wasn’t originally his, and the overall sentiment isn't even remotely a new one, but he amplified it for a reason.

Separately, Danny Sullivan and John Mueller discussed SEO for AI on the Search Off the Record podcast. The headline framing was that SEO for AI is still SEO, or perhaps a subset of SEO, even if the formats and surfaces differ.

Neither point is shocking in isolation. Together, though, they paint a fairly consistent picture of how Google wants the industry to think: stop obsessing over the label, stop chasing mechanical content patterns, and stop assuming that publishing at scale is inherently useful. Be genuine, useful, helpful, and quality-driven.

My Take: The “GEO is just SEO” line is both true and incomplete, which is why it keeps irritating sensible people.

At the highest level, yes, the core philosophy is shared. You still need to be discoverable, understandable, trustworthy, useful and preferable to alternatives. But that doesn’t mean GEO is a meaningless label any more than local SEO, video SEO or image SEO are meaningless labels. These subsets are important because the specific surfaces inherently shape and dictate exactly what is 'high quality' looks like. The inputs, outputs, optimisation levers and success metrics differ, even if those differences are nuanced and slight. In a competitive game with small margins between you and your rivals, small differences matter.

So when Google says “it’s all SEO”, the smart response isn’t to nod along too eagerly or reject it theatrically. It’s to notice the nuance. Once they get into the weeds, they usually end up conceding the point anyway – as was the case in this interview.

As for “digital mulch”, the interesting point is what kind of low-value content Google seems increasingly impatient with: pages built from familiar SEO ingredients, arranged competently enough to rank somewhere, but lacking any real viewpoint, evidence, distinction or reason to exist. That’s the commercial threat to mediocre content operations. Not that AI wrote it. Not that it’s optimised. But that it’s interchangeable. Nothing new, but that context makes John's throwaway comment more meaningful.

Final thoughts

Google is still broadening the answer layer. It’s still trying to manage the traffic trade-offs that come with that. It’s still improving AI Mode so users can trust it with more nuanced tasks. And it’s still signalling, in its usual slightly slippery way, that low-value search-shaped content is on borrowed time.

The opportunity for serious brands is still there, but it’s not in mass-producing pages and hoping the old mechanics keep carrying the load. It’s in becoming the kind of source, destination and brand that remains useful across traditional rankings, generated answers, citations, summaries and increasingly compressed user journeys.

That’s harder work. It’s also more defensible. And in 2026, that’s likely to matter even more.

‍