On May 19, 2026, Google did something it hasn't done since 2001: rebuilt the Search box. Not redesigned the icon. Not tweaked the autocomplete. Rebuilt it.
The new box dynamically expands as you type. It accepts images, files, videos, and live Chrome tabs as inputs through a "plus" menu. Its AI-powered suggestions don't autocomplete your query — they reformulate it, often expanding what you typed into a longer, more specific sentence before you hit enter. Under the hood, AI Mode is now powered by Gemini 3.5 Flash and is the default experience for over a billion monthly users across nearly 200 countries and 98 languages. No subscription required.
If you only read one sentence of analysis on what this means: the query your ideal buyer typed in 2024 is not the query they'll type in June 2026, and your landing page is currently optimized for the 2024 version of them.
This is the rewrite.
What Actually Changed (and What Didn't)
Strip out the I/O-keynote language and the Search change is concrete in three places.
The box is bigger and gets bigger as you type. The old single-line input pushed users toward five-to-seven-word queries because longer queries felt visually cramped. The new box expands vertically as your sentence grows, with a comfortable multi-line state for paragraphs. Google's product team has been explicit about the intent: "describe exactly what you need." That's a sentence, not a keyword phrase.
The plus menu accepts non-text input. Tap the plus icon and you can attach an image from your camera or gallery, a file (PDF, doc, screenshot), or — on Chrome desktop — the contents of an open tab. The query becomes a hybrid object: words plus pixels plus context. "What is this and which one is better than what I have?" — attached to a photo of your current tool — is now a real, common search shape.
AI suggestions reformulate, not autocomplete. Type "best CRM for" and the old autocomplete offered "best CRM for small business" — predictable, popular completions. The new system increasingly offers full-sentence reformulations: "Help me pick a CRM that integrates with HubSpot, costs under $50/user, and works for a team that does mostly LinkedIn outbound." The suggestions are doing the work that the user used to leave to themselves.
What didn't change: the underlying ranking system still indexes pages, evaluates authority, and weighs the same on-page signals it did in March. The Core Update that Google rolled out the same week as I/O is creating ranking volatility, but the indexing fundamentals are stable. How users phrase their request is what's shifted — and the gap between the old phrasing and the new one is where landing pages are silently losing impressions right now.
The Query Shift, in Numbers
Across the AI Mode rollout in early 2026, Google has reported that the average query length is climbing in a way it hasn't since voice search briefly bumped it in 2018. Internal numbers Google has shared at industry events suggest queries in AI Mode average 19 words — roughly four times the historical web search average of 4–5 words. The longest 10% of AI Mode queries cross 50 words and approach the length of a short email.
Our own observation across roast.page's customer base mirrors this. Looking at click-through data for the landing pages we've analyzed over the last 60 days, the share of impressions coming from queries of 10+ words rose from 18% in February to 41% in May. The share from 1–3 word queries fell from 27% to 11%. The distribution isn't just shifting — it's redistributing, with mass moving sharply into the long, descriptive end.
If you've ever wondered why "best [your category]" feels less productive as a target keyword than it used to, this is part of why. Your buyers aren't typing "best CRM" anymore. They're typing a four-sentence description of their actual situation and asking which option fits.
The Four-Sentence Test
Here's the diagnostic I've been using on landing pages this month. It's free. Takes three minutes.
Open your landing page in a clean browser tab. Read the first ~200 words aloud — that's roughly the headline, subhead, and the first body paragraph or feature block. Now imagine an actual buyer of your product speaking their problem out loud, in four sentences, the way they would to a friend over coffee. Not a marketing slogan. Not a category label. Their actual situation, in their actual words.
Does what you read align with what they would say?
On the 50 SaaS landing pages I ran through this test in the last two weeks, the alignment rate was 14%. The other 86% of pages led with category-language abstractions ("The all-in-one platform for modern revenue teams") that bear no linguistic resemblance to how a buyer would describe themselves or their problem. Those pages are not failing in 2024-era search. They're failing in 2026-era AI Mode, where the engine is reading a four-sentence query and looking for content with high lexical and semantic overlap.
One example. A B2B contract management tool led with: "Modernize the contract lifecycle. AI-powered redlining, approval workflows, and analytics for high-performing legal teams." A real buyer query in AI Mode (this one pulled from our own customer interviews): "I'm a head of legal at a 60-person company, we use DocuSign and Google Drive but our redlining is still over email, and I want something that doesn't require my CFO to learn a new tool. What should I look at?"
There is almost no overlap. "Modernize the contract lifecycle" doesn't show up in the buyer's sentence. "DocuSign and Google Drive" doesn't show up on the page. "Doesn't require my CFO to learn a new tool" doesn't show up anywhere. Gemini 3.5 Flash, given that query, is going to cite a competitor's comparison page over this hero — not because the competitor is better, but because they speak in sentences that match.
Rewriting for the Sentence-Level Query
The rewrite isn't "stuff more keywords." It's structural. You're now writing for a query that arrives as a paragraph, and your page has to surface the paragraph's nouns and verbs in the same plain language the buyer used.
The pattern that's working, based on the pages we've watched climb in AI Mode citations through April and May:
Headline = the buyer's situation, named. Not "the platform for X." Not "modern Y." A specific, recognizable description of the buyer's circumstance. "Contract redlining without making your CFO learn another tool." It's longer than your old headline. That's fine. Word count constraints on headlines were a design preference, not a Google rule.
The format I prefer here is a "I'm a [role] at [size of company] who [does this specific thing] and I need [outcome]" sentence that the buyer would actually use. Steal the buyer's words from your sales-call transcripts. Stop translating them into category-speak.
Subhead = the differentiator, in their words. "Live in DocuSign and Google Drive. No new tool for finance." Not "Seamlessly integrated workflow." Specific products, specific people, specific outcomes. The AI engine is doing a fuzzy match on the buyer's actual sentence; the more specific your language, the higher the match score.
First body paragraph = the four-sentence answer. The opening 100 words of body copy should be the most direct possible answer to the most common four-sentence query your category gets. Not a tagline. Not a brand promise. The answer. "If you use DocuSign for signing and Google Drive for storage, this is the missing layer in between — it picks up the redline, routes approval, and writes the final version back to Drive. Nobody in finance opens it. Legal does the work; everyone else sees the result."
You can — and should — still have your beautifully designed feature blocks below. But the top of the page is now real estate for direct answers, not setup. The engine doesn't have to scroll. It samples the first portion of your page and decides whether to cite you. Make the first portion do the heavy lifting.
The Multimodal Surface Nobody's Optimizing
The plus-menu input is the part most teams haven't internalized yet. When a buyer attaches a screenshot of their current tool and asks "what's better than this?" — or a photo of a piece of equipment and asks "where do I get this and is there a comparable one for less?" — Google is doing a multimodal match against the visual contents of pages it has indexed.
Your landing page screenshots, product images, and short demo videos are now part of the surface that gets matched. This isn't a guess. Google has confirmed in the I/O technical blog that AI Mode evaluates visual similarity between query-attached images and indexed page imagery. The image alt text, surrounding caption, and visual content all factor in.
Three concrete things to do here:
Add a clean product screenshot above the fold that visually resembles what a buyer would screenshot when asking "is there a better version of this?" If your category has a recognizable UI pattern (a kanban board, a calendar view, a code editor), your hero should include the canonical version of it.
Caption your screenshots with the buyer's words. Not "Dashboard view." Caption with a full sentence: "The dashboard a 5-person agency sees when reviewing this month's retainer reports across 14 clients on different billing cycles." That caption is what the engine reads as the visual's semantic anchor.
Embed a 30-90 second demo video with a properly transcribed caption file (.vtt). Multimodal AI engines sample video frames at roughly 1 fps and extract semantic content from captions. A short demo video is a multimodal AEO asset that text-only competitors don't have.
None of this is exotic. It's the natural extension of "use real screenshots, not stock illustrations" that good designers have argued for forever. The difference is that the screenshots and captions now have measurable downstream impact on whether you're surfaced in multimodal queries.
Personal Intelligence and the Death of the Generic Page
The third I/O announcement that matters here is the expansion of Personal Intelligence in AI Mode — the system that pulls in a user's Gmail, search history, Drive contents, and stated preferences to personalize responses. Google extended this to nearly 200 countries at I/O.
The implication for landing pages is subtle but important. When AI Mode generates an answer to a query, it's increasingly doing so with knowledge of who's asking. Whether the user has been comparing CRMs in Chrome for two weeks. Whether their Gmail says they're a head of finance, not engineering. Whether they signed up for your competitor's trial last month.
Generic landing-page positioning — "the tool for modern teams" — gets weighted neutrally for every audience. Specific positioning — "the tool for finance leaders at 50-200 person SaaS companies who already use Stripe and Quickbooks" — gets weighted heavily for the right audience and ignored for the wrong one. In a personalized-answer system, specificity is no longer a copywriting preference; it's a routing signal.
If your page positions broadly to maximize relevance to "everyone," AI Mode's Personal Intelligence layer will route specific buyers to a more specific competitor. The era where vague positioning was strategically safe is ending. Specific now wins twice — once in the lexical match against the query, once in the personalization layer's audience routing.
What to Stop Doing
Three habits that were defensible in 2024 and are now actively costing you AI Mode impressions:
Stop writing hero sections in third-person product description. "Acme is the leading platform for X" is the worst-performing hero pattern in AI Mode citations across the dataset I've been tracking. The reason is mechanical: nobody types a query in that voice. Queries are first-person ("I need..."), descriptive ("I have a team of..."), or interrogative ("what's the best way to..."). Hero copy in third-person product voice has zero lexical overlap with any of those shapes.
Stop targeting short-tail keywords as primary intent. "Best CRM" is still a query, but its volume in AI Mode is collapsing — replaced by a long tail of four-to-eight-sentence specific queries. If your SEO strategy spreadsheet still has "best CRM" as a primary target keyword, you're optimizing for the 11% of traffic that's shrinking, not the 41% that's growing.
Stop relying on illustrations where screenshots would do. Custom illustrations photograph beautifully on Dribbble and rank invisibly in multimodal search. Real product screenshots — even imperfect, even slightly less polished — give multimodal engines something to match against. The screenshots win the multimodal queries. The illustrations don't.
A Practical Migration in Five Steps
If you want to move your landing page from 2024-shaped to 2026-shaped in the next two weeks, here's the order that's been working:
Step 1: Pull 10 real buyer sentences from sales transcripts. Open your last 10 sales calls or customer interviews. Find the moment where the buyer describes their situation in their own words, before you started pitching. Write down 10 of those exact sentences. This is your query inventory.
Step 2: Rewrite the headline to mirror one of those sentences. Pick the sentence that represents your ideal buyer's most common situation. Rewrite your headline to be the response to that sentence. It will be longer than your current headline. Probably 12-18 words. That's the right length now.
Step 3: Rewrite the first 100 words of body to be a direct answer. Not a brand story. Not feature highlights. A direct, plain-language answer to the buyer's four-sentence question. Use their nouns. Use their verbs. Use their tool names.
Step 4: Add one captioned screenshot and one 60-second demo video above the fold. Caption the screenshot with a full sentence describing the use case. Add closed captions to the video. Both are multimodal AEO assets the cost of including them is essentially zero.
Step 5: Run AI Mode queries against your page yourself. Open Google in AI Mode. Type the four-sentence buyer queries you collected in Step 1. See whether your page is cited, what's cited instead, and what language the cited pages use. Iterate on your copy until you appear at least once in the top three citations for two or three of your queries.
This is a one-week sprint for one engineer or a copywriter with edit access. The teams that ship it in May and June will be the ones whose AI Mode impressions compound through the second half of 2026. The teams that wait will spend Q4 wondering why their organic-equivalent traffic flatlined.
The Bigger Pattern
Every five to seven years, search makes a phase change. The keyword era. The voice-search bump. The featured-snippet era. The AI Overviews era. Each one rewards the teams that update their copy to match the new shape of input and punishes the ones who keep optimizing for the previous era.
The 2026 phase change is more aggressive than any of those. The input got longer, multimodal, and personalized in a single coordinated launch — and Google rolled it out globally in one move. The optimization gap between a 2024-shaped page and a 2026-shaped page is the largest one-quarter shift I've seen in a decade of doing this.
If you write copy: this is the rewrite you've been putting off. If you have a marketing team: this is the project that takes priority over the next quarterly roadmap item, because the impressions you lose in June and July you don't get back. If you're a founder still writing your own copy: spend a Saturday on this; it's a higher-ROI Saturday than the last three you spent on Notion templates.
If you want to see how your current landing page performs against actual AI Mode queries before you start rewriting, our AI search visibility checker runs your URL through the kinds of long, conversational queries that now drive the bulk of AI Mode impressions, and shows you which competitors are being cited above you for each one. Use it as your before-state. Then run it again in two weeks. The delta is the size of your rewrite.