For fifteen years, “search intent” meant a string of text typed into a box. In 2026, that box is disappearing. It’s being replaced by a lens, a microphone, and a pair of glasses that never leave someone’s face. If you’re still modeling queries as keyboard input, you’re modeling the wrong interface.
This is a working guide for search scientists — the people who actually build the campaigns, not just theorize about them. We’ll cover what’s shipping right now in Google Ads for AR and voice, how spatial audio changes bid strategy, and a concrete setup checklist for smart glasses inventory.
Why this matters now: Google’s smart glasses partnerships (Samsung, Warby Parker, Gentle Monster) and the Gemini-powered Assistant stack pushed ambient, always-listening, always-rendering queries into mainstream hardware in 2025–2026. Query volume through wearable surfaces is no longer a rounding error — it’s a distinct inventory type with its own auction dynamics.
Table of Contents
Why "Spatial and Vocal" Is a Real Category, Not a Buzzword
Three shifts collapsed into one hardware form factor:
- Voice-first input — glasses have no keyboard. Every query is spoken, which means longer, more conversational, more entity-dense phrasing than typed search.
- Spatial context — the device knows where the user is looking, standing, and moving. “Where can I get this cheaper” now has a literal object in the field of view attached to it.
- Ambient audio delivery — responses (and ads) arrive as spatial audio anchored to a location in 3D space, not a rectangle on a screen.
For a search scientist, this means query classification, bid modifiers, and creative all need a new axis: surface type. Text search, voice search, and spatial/AR search are no longer the same auction with different UI skins — they have measurably different intent distributions and conversion paths.
We live, breathe & Dream Paid Traffic
What's Actually Live in Google Ads Right Now
| Feature | What it does | Status | Search scientist notes |
|---|---|---|---|
| Voice Campaigns | Native bidding layer for Assistant/Gemini voice queries, separate from standard Search | Beta | Requires conversational match type — broad match performs better here than in text search |
| AR Product Overlays | Shopping ads rendered as 3D objects anchored to real-world surfaces via smart glasses camera | Limited pilot | Feed quality (3D asset, dimensions, material) becomes a ranking signal, not just an image URL |
| Spatial Audio Ads | Audio creative positioned in 3D space, triggered by proximity or gaze dwell time | Early access | New metric: "dwell-to-listen" replaces CTR as the primary engagement signal |
| Gaze & Dwell Signals | Bid modifiers based on how long a user's gaze rests on a physical object matched to inventory | Testing | Treat like a viewability threshold — set minimum dwell time before an impression counts |
| Conversational Query Insights | Search terms report expanded to include full voice transcripts and follow-up turns | Live | Mine this for long-tail, question-based negative and positive keyword harvesting |
Setting Up an AR Campaign: The Checklist
Treat this like a Shopping feed migration, not a new campaign type from scratch.
- Audit your product feed for 3D-readiness. Google’s AR renderer needs GLB/USDZ assets, real-world dimensions, and material metadata. If you only have flat product photography, this is your blocker — not budget.
- Segment AR inventory into its own campaign. Don’t blend it with standard Shopping. Conversion windows are longer (users often browse in-store, convert later) and CPA benchmarks are still forming — you need clean data to set them.
- Set gaze-dwell thresholds conservatively. Start at 1.5–2 seconds minimum dwell before an impression is billable. Anything shorter is noise from someone glancing past a shelf.
- Write copy for the ear, not the eye. Spatial audio creative needs to work with zero visual reinforcement. Lead with the brand name and the single strongest differentiator in the first 3 seconds.
- Build negative keyword lists from conversational transcripts. Voice queries include filler and false starts (“um, find me, uh, something like—”). Your existing negative list built from typed search won’t catch these patterns.
Voice Search Campaign Structure That Actually Works
The mistake most accounts make: applying text-search match type logic to voice. Voice queries are 3–5x longer on average and front-load context (“okay so I’m standing in the kitchen aisle and I need…”) before the actual intent. Structure around this:
- Broad match + Smart Bidding, not phrase match. Voice phrasing variance is too high for phrase match to catch enough volume.
- Question-based ad groups. Cluster by interrogative (“where,” “how much,” “does this”) rather than by product category alone — the interrogative predicts funnel stage better than the noun does.
- Shorter headlines, longer descriptions. Voice assistants read headlines aloud; keep them under 6 words so the TTS rendering doesn’t clip or rush.
Data point to watch: Conversational Query Insights reports are currently under-sampled in most accounts because voice volume is still ramping. Don’t make structural decisions off fewer than ~500 voice search terms — the noise floor is high below that.
The Bottom Line
Spatial and vocal search aren’t a distant roadmap item — they’re a live beta you can be testing in your account this quarter. The search scientists who win this cycle will be the ones who treat AR and voice as distinct inventory with distinct data models, not as a stylistic variant of the same old text campaign. Start with the feed. Everything else follows from whether your product data is 3D- and voice-ready.
Check out more blog posts:
What Is Googlebot & How Do Search Engines Work? (2026 Guide)
