Most work research still looks the same. Fifteen tabs. Three unfinished searches. A half-written question you abandoned because the context got too messy to carry.
Google's new AI Mode is trying to collapse that into one thread. I ran three real work tests during May 2026 rollout week to find out what actually changed — and where it still breaks.
Here's what happened.
Test 1: research across open tabs
What I tested: Instead of reading multiple tabs separately and piecing together an answer, AI Mode now accepts your open Chrome tabs directly as context. Hit "+" in the search box, select open tabs, and the synthesis happens inside the thread.
The exact setup: I opened two hiking pages in Chrome — Kirby Cove and the Angel Island Perimeter Loop — then asked:
"I liked these two hikes I did recently, what are similar ones near Monterey that could also be good for kids?"
What happened: This was the first test that made AI Mode feel genuinely different. Instead of bouncing between regional blogs and Reddit threads manually, Google mapped the patterns from both open tabs and surfaced alternatives that matched the original context — kid-friendly terrain, similar pacing, similar scenic profile. The recommendations were actually useful.
It stopped feeling like search results and started feeling like someone already did the comparison work.
The honest caveat: AI Mode currently treats your attached tabs as ground truth. If the tabs contain outdated information, affiliate bias, or low-quality content, Google synthesizes that bias directly into the answer. It's good at synthesis. It's not skeptical about sources. That distinction matters before you trust anything it pulls.
Test 2: upload a screenshot, skip the translation step
What I tested: Most work information is trapped inside screenshots — dashboards, PDFs, campaign reports, exported data. Normally you still have to translate all of that into text before AI can engage with it. AI Mode now removes that step.
The exact prompt: I uploaded a campaign performance report screenshot and asked:
"What is the primary performance bottleneck shown in this campaign report screenshot, and what are two logical next steps to fix it based on mid-market SaaS benchmarks?"
What happened: The moment you upload a file or screenshot, Search routes directly into AI Mode instead of visual search. The extraction was fast. Dense information became conversational almost instantly. This was the most immediately useful test for day-to-day work — it removes one of the biggest hidden frictions in AI workflows: preparing the context before you can ask the question.
The honest caveat: There's already evidence that some sites are using SEO-heavy formatting and keyword stuffing to manipulate Google's extraction layer. Uploaded files with heavily optimized layouts can push the AI toward promotional conclusions instead of neutral analysis. The extraction layer is becoming a new surface for manipulation. Worth paying attention to early.
Test 3: one big question, multiple follow-ups
What I tested: Whether AI Mode can hold context across a real, messy strategic question — not just one lookup, but a full thread where each follow-up builds on the last without restarting from zero.
The exact prompt:
"We need to launch a defensive local marketing strategy against a competitor who is aggressively targeting our core demographics in the Midwest — what are the immediate blind spots we should look for in their digital presence, and how do we exploit them on a limited budget?"
What happened: The first response was strong. Competitor analysis angles, local SEO weaknesses, content opportunities, budget-conscious positioning — and it read like an AI-generated strategic answer, not a list of links. Context carried through: business size, budget, regional focus, competitive positioning all held across follow-ups without re-prompting.
Then I asked for a specific 30-day implementation framework for the local SEO portion.
That's where it broke. Google dropped back into traditional search behavior and surfaced mostly links instead of synthesizing the answer. The entire promise of AI Mode is reduced friction. Being pushed back into tab-hunting and manual synthesis mid-thread breaks that promise immediately.
I asked one more follow-up — weekly metrics to track. The AI-generated response came back.
So the experience right now is uneven: sometimes conversational, sometimes traditional Search, sometimes neither.
Step-by-step multiple follow-ups
The honest verdict: This was still the strongest test overall. When it works, the reduction in cognitive friction is real and noticeable — you can feel the difference between rebuilding context every time and having it carry. But Google is still caught between two systems. You can feel both competing underneath the experience. The "Continue in AI Mode" bridge from standard AI Overviews works better than expected. The inconsistency in the middle is the part that matters.
One more thing I noticed: a growing number of power users are already using browser extensions or custom URL parameters to disable AI Mode entirely for basic factual lookups. They find the AI integration too aggressive for simple searches. That tells you something — Google isn't just changing Search. It's changing expectations around how people interact with information itself.
The actual shift
The tests themselves aren't the point.
The interface of research is changing. The old workflow: search → open tabs → compare → summarize manually → decide. The new one: give AI the messy context → let it synthesize → verify the result → move.
When that chain holds, the friction reduction is noticeable. When it breaks mid-thread, you feel it immediately.
The people who benefit most from this won't be the most technical. They'll be the ones who learn how to give AI useful context, how to verify what it returns, and how to move without drowning in information.
That's the actual change underneath all of this.
Tests conducted during Google's May 2026 rollout week. Some interface details may evolve as Google continues updating AI Mode globally.