Milena Traikovich lives at the crossroads of lead generation and AI-powered search. She has spent years tuning analytics, optimizing performance, and orchestrating nurture programs that turn curiosity into qualified pipeline. In this conversation with Noah Thwaite, Milena unpacks how Microsoft’s upcoming AI Max for Search and a wave of new commerce features across Bing, Copilot Search, and Copilot Answers are rewriting the playbook. We explore how longer, richer queries change keyword strategy and creative, how to keep brand control while leaning into automation, and how new visibility tools, structured product data, and conversational shopping experiences can unlock measurable growth—from impression share lifts near 90% with real-time product feeds to twofold gains from on-site Brand Agents. Along the way, she shares pragmatic rollouts, auditing checklists, and test designs to turn AI-era discovery into dependable demand.
AI-powered search queries are getting longer and more detailed. How should advertisers rethink keyword strategy, creative testing, and landing page design to capture this intent, and what early indicators would tell you the approach is working?
Start by shifting from brittle, exact-match lists to clustered intent themes that embrace expanded query matching in AI Max, then map each cluster to modular creative assets and URL variants. Your creatives should include dynamic snippets that can personalize to the nuance in longer prompts, and your landing pages should be block-based so they can adapt to sub-intents—comparison, troubleshooting, or price sensitivity—without spinning up dozens of static pages. I watch for early signals in search term and asset reports from day one: longer-tail phrasing appearing in match logs, higher engagement with personalized assets, and meaningful lifts in qualified lead behaviors like deeper scrolls or tool interactions. When that alignment lands, you feel it in the clarity of sessions—visitors linger, explore, and convert with less friction—because the page mirrors the language they used to find you.
Expanded query matching, asset personalization, and URL routing are being bundled into AI-driven campaign types. How would you structure safeguards, set measurement baselines, and phase rollouts to balance automation gains with brand control?
Treat AI Max like a powerful but fenced garden: opt in with brand inclusions/exclusions, term exclusions, and messaging constraints pre-populated from your brand book and legal lists. Establish a clean baseline using your current search setup, then introduce AI Max in a controlled segment—one product line, one geo, or one audience—so measurement doesn’t get muddied by simultaneous changes. I phase automation in layers: first query matching, then asset personalization, then URL routing, with weekly reviews of search term mapping and asset usage to confirm brand-safe alignment before unlocking the next capability. The balance feels right when the system is clearly expanding reach into longer, high-intent queries while your constraints keep language on-brand and destinations relevant.
Opt-in controls now include brand inclusions/exclusions, term exclusions, and messaging constraints. Which combinations have the biggest impact on quality and safety, and how would you audit them weekly with concrete checklists and thresholds?
The winning trio is brand inclusions to prioritize your core entities, exclusions to fence out known low-quality themes, and messaging constraints to keep claims compliant. My weekly audit runs like a pilot’s preflight: scan new search terms for drift, confirm excluded terms are actively suppressing, validate that ad copy variants never cross your messaging red lines, and spot-check routed URLs for relevance and policy fit. I add a creative asset review against a living style guide—tone, value props, and visual cues—plus a landing-page sniff test to ensure the promise matches the page. Safety improves most when these controls move together: inclusions aim the system, exclusions clear hazards, and constraints keep the voice consistent even as assets personalize.
Search term and asset reporting are available from day one in new AI-driven campaigns. Which metrics and slices would you review first 30, 60, and 90 days in, and how would you turn those insights into budget shifts or creative changes?
In the first 30 days, I mine search term breadth versus intent fit and evaluate which assets the system favors in longer queries, then prune mismatches and double down on assets tied to high-quality behaviors. By day 60, I slice performance by conversation context—problem, comparison, or solution language—and route budget toward themes where personalized assets are clearly resonating, while building net-new variants for underperforming contexts. By day 90, I compare multi-surface performance across Bing, Copilot Search, and Copilot Answers to see where queries are most detailed and where assets get cited; that’s where I push incremental budget and spin up new landing modules. Throughout, I let the reports guide creative evolution—swap headlines that echo emergent phrasing, expand FAQs reflecting conversational asks, and align imagery with the emotional cues users bring to the query.
Automated traffic is outpacing human visits, with AI-driven sessions surging and agentic browser activity jumping thousands of percent. How should marketers adjust attribution windows, session definitions, and incrementality testing to avoid over- or under-crediting?
With AI-driven sessions nearly tripling in 2025 and agentic browser traffic up about 8,000% year over year, redefine a “session” to separate human exploration from agent activity so models don’t inflate performance. I maintain parallel attribution views: a human-only lens for conversion path clarity and an all-traffic lens to understand how AI exposures influence demand. For incrementality, I use geo or audience holdouts and rotate opt-in features like URL routing on and off in matched cohorts; if human conversion doesn’t move while agent exposures spike, you’ve likely got assistive rather than direct value. The outcome should be calmer reporting—less whiplash—and truer readouts of what actually persuades a person to buy.
Visibility into which pages influence AI responses—even without clicks—is expanding. How would you operationalize this, from mapping content to themes, to filling gaps, to tracking share-of-answer against competitors, and what KPIs would you use?
I start with a content-to-theme map keyed to buying moments—awareness, evaluation, and decision—and tag each page with primary and secondary intents. Using AI Visibility in Microsoft Clarity, I track which pages are cited in AI-generated answers and assemble a share-of-answer view: percent of responses drawing from our content versus competitor content on priority themes. Gaps become a build queue—FAQ blocks for long-tail questions, comparison matrices, and policy clarifiers that AIs love to cite because they resolve ambiguity. My KPIs are share-of-answer movement by theme, citation frequency for decision pages, and downstream conversion signals; when your pages are increasingly “present” in answers even without clicks, you feel it in steadier, more confident buyers.
When competitor pages are cited more often than a brand’s pages on key topics, how do you diagnose the root cause? Walk through a step-by-step playbook for content upgrades, schema changes, and internal linking, and share turnaround timelines you’ve seen.
First, diagnose: is the competitor winning on completeness, clarity, recency, or structure? I inventory our pages against theirs, looking for missing spec details, policy nuances, or outdated pricing language that might make an AI favor their content. Then I execute in sprints: refresh copy for precision, add structured elements and comparison tables, implement schema to make specs and policies machine-readable, and build internal links from authority pages so crawlers and AIs can find and trust the update. With focused resourcing, I’ve seen notable citation shifts in a few cycles of crawling—often within the same quarter—especially when the upgrades remove ambiguity and add structured clarity that AIs can readily cite.
Structured commerce data is becoming essential. How should teams prioritize product data cleanliness, pricing freshness, and inventory accuracy, and what governance or data contracts keep feeds reliable as catalogs and policies change?
Make a “golden record” your north star: one authoritative source for titles, attributes, pricing, and availability that all surfaces read from. Prioritize accuracy where shopper trust is on the line—pricing freshness and inventory status—then deepen attributes and enrich content so AI systems can find, cite, and match to nuanced queries. I formalize data contracts across merchandising, engineering, and marketing that define field ownership, change triggers, and validation rules, with QA gates before anything syndicates. The payoff is reliability across channels, and you’ll notice it in fewer dissonant moments—no more promising a variant or price that doesn’t exist when a user engages through Copilot.
Universal Commerce Protocol–ready feeds aim to standardize how AI agents read product data. What practical hurdles do merchants face when adopting these feeds, and how would you phase implementation across engineering, merchandising, and analytics?
The biggest hurdles are mapping legacy fields to UCP-ready schemas, normalizing inconsistent attributes, and aligning update cadences so pricing and inventory stay fresh. I phase it like this: engineering builds the translation layer and validates required fields; merchandising cleans and enriches attributes; analytics instruments downstream tracking so you can read impression share, engagement, and conversion impacts. Start with a core category to prove the pathway, then expand once syndication is smooth and QA catches edge cases. This staged approach reduces the anxiety that comes with schema changes and sets you up to participate fully as AI agents standardize how they ingest product data.
Shopify Catalog can stream real-time product data into AI shopping experiences, with some merchants seeing major impression share gains. What prerequisites, enrichment tactics, and monitoring would you recommend to unlock similar lifts?
Ensure your Shopify Catalog is pristine—accurate pricing, live inventory, and well-structured attributes—so the real-time feed reflects reality. Then enrich: consistent titles, compelling descriptions, variant clarity, and policy details like shipping and returns that AIs can surface in context. Monitor impression share within Copilot and watch for shifts like the nearly 90% growth top Shopify merchants have recorded through real-time feeds; pair that with product-level engagement to see which attributes attract conversational interest. When the data hums, you feel momentum—more eyes on the right products, fewer dead-ends, and smoother handoffs into checkout.
Copilot Checkout enables transactions inside an assistant while keeping the merchant of record. How do you integrate this without cannibalizing site conversion, and what experiments would you run to compare conversion rates, AOV, and refund or fraud outcomes?
I integrate Copilot Checkout as a complementary lane, not a detour—surface it where the user’s context is conversational or mobile, and keep your site checkout prominent for browsing-led journeys. To guard against cannibalization, I run A/B or geo-split tests that randomize Copilot Checkout availability and compare not just conversion rate but AOV, post-purchase refunds, and fraud outcomes. Expand availability when you see stable or better economics and equivalent or improved customer signals. With catalog data expanding to more than 500,000 merchants and mobile availability growing, the key is to let users choose while you measure which path delivers happier customers and healthier margins.
Loyalty linking at checkout can surface member benefits like free shipping. How would you design eligibility rules, edge-case handling, and messaging to reduce cart abandonment, and which metrics signal true loyalty lift versus promotion-driven spikes?
I define crisp eligibility rules—who qualifies, when benefits apply, and how benefits stack with offers—and build graceful fallbacks that invite account linking without blocking checkout. Messaging needs to be calm and confidence-building: show the value (free shipping, exclusive discounts) early, then reaffirm it at payment so users feel the win rather than second-guess it. To separate durable loyalty from promo spikes, I track repeat purchase cadence, member engagement with benefits, and post-promo retention alongside conversion; if behavior sustains after the perk window, loyalty is doing the work. You can almost sense the relief in the journey when benefits are clear—less hesitation, fewer backtracks, and a cleaner path to purchase.
Brand Agents bring conversational assistance onto merchant sites and reportedly double conversions in some cases. What product taxonomy, policy training, and escalation paths are required to sustain that lift, and how do you staff QA and measure agent quality?
Start with a rigorous product taxonomy so the agent can navigate variants and options without confusion, then train it on brand and policy materials—shipping, returns, warranties—so answers are precise and trustworthy. Build escalation paths for ambiguity: hand off to chat or support when confidence dips or policy stakes rise. Staff QA like you would for a high-stakes funnel—review transcripts, score intent resolution, and audit adherence to policies—because that average twofold lift in conversions happens only when quality stays high. I measure agent quality by resolution rate, policy compliance, and downstream satisfaction; when it’s working, conversations feel human—helpful, fast, and reassuring.
Offer Highlights let retailers feature benefits like free shipping or in-store pickup directly in AI conversations. Which differentiators perform best by category, and how would you test messaging sequencing across Bing, Edge, and Copilot to prove incremental sales?
Match differentiators to shopping anxieties: logistics perks like free shipping or in-store pickup for bulky or time-sensitive categories, and assurance signals like warranties or easy returns for considered purchases. I test sequencing by surface—lead with the most anxiety-reducing benefit in Copilot conversations, then reinforce specifics on product detail page ads across Edge and Bing. Use holdouts where Offer Highlights are withheld for matched SKUs and track lift in engagement and sales; it’s the cleanest way to isolate incremental impact. Best of all, you can feel friction melt when the right benefit appears at the right moment—shoppers exhale and move forward.
Audience generation turns a natural-language prompt into targeting settings. How would you validate and refine those suggested audiences, and what safeguards ensure demographic and interest signals don’t drift from brand guidelines or legal requirements?
I validate by triangulating: compare the AI-suggested segments to your existing high-value audiences, inspect in-market and custom signals for intent fit, and run small-budget pilots to gather performance and quality cues. Refinement comes from creative resonance—if messaging built for that audience lands, you’ll see engagement and qualified actions rise in tandem. Safeguards live in your governance: codified brand guidelines, legal constraints baked into prompts, and periodic audits of demographic and interest settings to prevent drift. Audience generation is a gift when guided; left alone, it can wander, so keep a steady hand on the tiller.
Do you have any advice for our readers?
Treat this wave as an operating-system upgrade for demand, not a feature drop. Lean into AI Max and the commerce stack with clear guardrails, and use the new visibility—AI-driven sessions nearly tripled, agentic traffic up about 8,000% year over year, real-time feeds driving impression share gains near 90%, and on-site Brand Agents averaging a twofold conversion lift—as motivation to modernize your data and content. Start small, measure like a skeptic, and expand what proves durable. Most of all, design for the human at the end of the conversation; if your answers are trustworthy, your data is fresh, and your promises are kept, growth follows.
