Google rebuilt its Search box at I/O. The new version accepts images, files, videos, and even open Chrome tabs as inputs, with AI Mode promoted from a separate destination to a primary surface inside core search. This is the single most consequential I/O announcement for anyone whose marketing strategy still treats organic search as the dominant traffic source.
The piece I wrote a few months ago about organic SEO after AI Overviews is the predecessor to this one. That essay covered the first wave — what changed when AI Overviews started occupying the top of the SERP. This essay covers the second wave: what changes when the search input itself becomes multimodal and the AI surface becomes the default mode.
This is the marketer's playbook for the next quarter.
What the New Search Box Actually Does
The redesign collapses several previously-separate surfaces into one. Three concrete changes that matter:
Multimodal inputs
The Search box now accepts images, files (PDFs, docs, spreadsheets), videos, and the contents of open Chrome tabs as inputs to a query. Users who previously had to either type a query that described what they wanted to find, or upload through Lens as a separate workflow, now do both in one box. The implication for intent: searches will become more specific, more contextual, and harder to bucket into the keyword-shaped categories that organic SEO was built around.
AI Mode as default
AI Mode (the LLM-summary-first search experience) was previously something users opted into. It's now integrated into the standard search experience for most queries. The blue-link results page didn't go away, but it's now further down the page on a meaningful share of queries — and on those queries, the share of clicks reaching any blue link at all is materially lower than it was twelve months ago.
Tab-as-context
Letting users pass their open Chrome tabs as input to a search means queries can be informed by what the user was actually looking at when they searched. "What's the difference between this product and that one" can now refer to two tabs the user has open. This is a category of intent that didn't really exist before because there was no good way to express it.
Why This Changes Intent Measurement
The framing that breaks under this change is "keyword intent." For two decades, SEO planning has been organised around the idea that users search using keywords and that the keyword reveals the intent. Both halves of that statement are now weaker.
Searches that involve an image, a file, or an open tab as part of the query don't have a clean keyword to optimise against. The intent is real but it's expressed across multiple modalities. Tools that report "this many users searched for [keyword]" capture less and less of what's actually happening on the SERP. The dashboards that have anchored marketing decisions for years are quietly becoming less informative — which is the same diagnosis that applies to most of the performance-marketing reports I've written about, now extended into organic.
The practical implication: stop measuring SEO success primarily by keyword rank. Start measuring it by cited share — how often your content appears as a citation in AI Mode answers — and by task-completion presence — whether users who complete the underlying task they were searching for ever passed through your domain. These are harder metrics to instrument. They also actually reflect what's happening.
The Three SEO Motions That Newly Matter
Three motions are now disproportionately high-leverage. Most teams should reallocate budget toward them this quarter.
Citation-first content
In AI Mode, the model writes the answer and cites its sources. The question that determines whether your domain shows up in the citation set is not "do you rank in the top three blue links" but "are you the most-cited source on this specific question." The two correlate, but they aren't the same metric. Content optimised for citation-first surfaces tends to be:
- Specific — answers a narrow question precisely rather than ranking for a broad keyword.
- Authoritative on a sub-topic — depth on a slice the model treats as canonical.
- Structurally clear — explicit Q&A structure, summary blocks, schema markup that signals what the page actually answers.
This is a different shape of content than the long-tail blog posts that dominated SEO in the previous era. Less volume, more depth, narrower targeting per piece.
Brand-as-entity work
The model's decision about which sources to cite is partly a "which entity is most authoritative on this topic" decision. That decision is informed by signals the SEO industry has cared about for years (links, mentions, structured data) but with a meaningful weight shift toward entity-level signals: Knowledge Graph presence, consistent brand identity across the web, third-party validation. The teams that built their authority signals around individual page rankings are exposed; the teams that built them around brand-as-entity recognition are not.
Workflow-of-the-user content
Multimodal queries with tab context mean the model knows roughly what task the user is trying to complete. Content that wins citations is increasingly content that helps with the whole task, not just the immediate question. "Compare these two products" gets answered by content that explains the comparison, not just content that lists features.
The right framing here is the same one I've used when thinking about budget split between Google Ads and SEO: the optimisation surface has shifted, and the winning strategies are different than they were in 2023, but the underlying question — "how do I help my prospective customer complete the task they're trying to complete" — is unchanged. What's changed is which tactics actually answer it.
What Stops Working
Three motions whose ROI is dropping fast:
Long-tail content farms
Producing volumes of mediocre content targeting long-tail keywords used to work because Google's ranking algorithm rewarded thoroughness. The new model rewards specificity and authoritativeness, not volume. A hundred mediocre posts targeting variants of the same question are now worth less than five excellent ones targeting the same surface area.
Pure rank tracking
Rank-tracking tools still report numbers. The numbers correlate less and less with actual traffic, because the SERP itself has been reshaped above the blue-link results. A team that's optimising for top-three rank without measuring share of citations in AI Mode is optimising for a metric that's moving away from the underlying business outcome.
Generic listicles
The classic "10 best X for Y" listicle pattern is being absorbed wholesale into AI Mode summaries. The user gets the list without leaving the SERP. Listicle traffic is collapsing for queries where the AI summary is comprehensive enough to obviate the click. Some categories (high-consideration purchases, complex products) remain robust; commodity categories do not.
Migration Playbook
A pragmatic re-routing of organic SEO budget for the next quarter:
- Audit your top fifty pages by historic traffic. Identify which ones are still getting impressions but losing clicks. Those are the pages most affected by AI Mode. Rewrite them for citation-first reading: explicit question framing, summary up top, sources visible.
- Instrument citation tracking. Even rudimentary tracking is better than none. Manual audits of AI Mode answers for your top twenty target queries, monthly, will tell you which of your content is being cited and which isn't.
- Reallocate budget from volume to depth. If you were producing twenty long-tail posts a month, cut to five and triple the depth. The new ranking surface rewards canonical sources, not coverage.
- Audit your entity presence. Knowledge Graph, brand mentions, structured data, third-party citations. These were always SEO inputs; they're now meaningfully heavier than they were.
- Update your reports. Stop leading with rank. Lead with cited share and task-completion presence. The reports will look uglier in month one because the metrics are new. They will be more honest.
What's Not Changed
The unchanging caveats:
- Search intent is still real. Users still want to find things, complete tasks, and decide between options. The shape of how they express it has changed; the underlying motivation has not.
- Brand still matters. A user who knows your brand is still more likely to click through to your site than one who doesn't. Brand-building budget is not optional in the new environment.
- Quality compounds. Content that genuinely helps a user is rewarded by both the AI summary layer and the residual blue-link results. The teams that have always invested in quality content are best-positioned.
- The platform risk is unchanged. Building your entire customer-acquisition engine on a single search platform is a strategic decision with the same exposure it always had. Diversify accordingly.
The Practitioner's Take
The Search box redesign is the second-to-last major reshape of organic search I expect for the next two years. (The last is when an AI agent — not the user — does the searching on the user's behalf, which is closer than it sounds.) The teams that adapted to AI Overviews twelve months ago are in roughly the right position to adapt to this. The teams that didn't are now two waves behind, and the gap will keep widening.
The marketer who's still measuring success by keyword rank is doing the equivalent of measuring website performance by page-load time on dial-up. The metric is real, but it doesn't describe what's actually happening. Update the metrics, update the content motions, and the rest of the playbook reorganises around them.
The window where this matters is now. The teams that move this quarter are the ones whose organic share holds steady through 2026 and 2027. The teams that wait are going to spend the next two quarters debugging why their content is no longer found and rediscovering, slowly, that the search interface they were optimising against is gone.