When one mid-sized e-commerce brand decided to run its entire paid search budget exclusively inside Google AI Mode ads, the internal debate was fierce — but the results made the argument irrelevant. This case study breaks down exactly how the brand achieved a 3.2x return on ad spend (ROAS) within 11 weeks, covering every setup decision, bidding configuration, and measurement challenge that shaped the Google AI Mode ads e-commerce results.
Context and Challenge: The Brand, the Problem, and What Was at Stake
The brand in this case study is a direct-to-consumer home goods retailer operating primarily in the United States, with an average order value (AOV) of $148. At the start of 2026, they were running a conventional Performance Max campaign alongside standard Shopping ads, achieving a blended ROAS of roughly 1.9x on a monthly ad spend of $22,000. Profitability was marginal — their internal target was 2.8x to cover margins and fulfillment costs.
The core problem was attribution fragmentation. Users were increasingly starting product research in AI-generated answer surfaces, clicking through several AI-generated summaries, and only converting two to four sessions later. Standard last-click attribution was attributing almost nothing to upper-funnel discovery touchpoints, making it impossible to justify any budget for awareness-stage placements. The marketing team was essentially flying blind on half of their purchase funnel.
When Google began expanding ad placements directly inside AI Mode conversational results, the brand's paid search manager saw an opportunity: instead of treating AI Mode as one more channel layered on top of existing campaigns, what if they restructured entirely around it? The risk was real — pulling budget from known-performing Shopping campaigns to test an emerging format with uncertain measurement was a decision that required sign-off at the CFO level.
Strategy and Approach: What Was Decided — and What Was Deliberately Left Out
The core strategic decision was to run a clean split rather than a blended test. For 11 weeks, the brand paused its standard Shopping campaigns entirely and allocated 100% of its $22,000 monthly budget to AI Mode placements. This was not a gradual ramp — it was an intentional forcing function designed to generate unambiguous data rather than muddied cross-channel results.
What they chose NOT to do is equally instructive. The team deliberately avoided Performance Max during the test window, even though PMax can include AI Mode inventory. The reason: PMax's opaque allocation would have made it impossible to isolate AI Mode's contribution. They also chose not to run Display retargeting in parallel, resisting the temptation to prop up results with warm-audience ads.
"The hardest discipline in this test was doing less. Every instinct said to hedge — run PMax as a safety net, keep some Shopping live. Resisting that pressure was what made the data clean enough to actually act on."
Bidding strategy defaulted to Target ROAS, set initially at 2.5x — deliberately below their 2.8x profitability threshold to give the algorithm room to gather data in a new placement environment. The team planned to step the target up in increments of 0.2x every two weeks once the campaign had accumulated at least 50 conversions per week.
Audience signals were loaded heavily upfront: first-party customer match lists (past purchasers from the last 18 months), website visitor segments with 30-day lookback windows, and a custom intent audience built around 47 high-intent product-category search terms. The theory was that AI Mode queries tend to be longer and more contextual — meaning audience context would matter more than keyword matching as a relevance signal.
Implementation: Key Steps, Timeline, and Tools
Week 1 was entirely pre-launch preparation. The team audited their Google Tag Manager setup and rebuilt their conversion tracking configuration from scratch, ensuring that both Google Ads conversion tags and Google Analytics 4 purchase events were firing correctly and deduplicated. This step proved more consequential than anticipated — they discovered that roughly 18% of purchases had been double-counted in historical reporting, which explained why their baseline ROAS had looked healthier than it actually was.
Proper Google AI Mode ad conversion tracking was treated as a non-negotiable foundation. The team implemented enhanced conversions to pass hashed first-party data, and enabled consent-mode v2 to maintain measurement quality under privacy constraints. They also set up a custom GA4 exploration report to track multi-session conversion paths specifically, capturing cases where the first AI Mode click initiated a journey that converted over subsequent sessions.
Ad creative was restructured around conversational intent. Rather than the typical short-headline Shopping format, the team wrote longer, benefit-led descriptions that answered the kind of questions users were already asking in AI Mode — "which is the best material for a small kitchen?" rather than keyword-stuffed product names. Five asset variants per product category were loaded, with the expectation that Google's system would surface the most relevant variant given the conversational query context.
| Week | Key Activity | Milestone |
|---|---|---|
| Week 1 | Tracking audit, tag rebuild, enhanced conversions setup | Tracking accuracy confirmed at >98% |
| Week 2 | Campaign launch, audience signals loaded, creative variants live | First 50 impressions recorded in AI Mode placement |
| Weeks 3–4 | Bid strategy learning phase, no major changes | 52 conversions recorded; tROAS raised to 2.7x |
| Weeks 5–8 | Creative rotation analysis, low-performing assets paused | ROAS crossed 2.8x profitability threshold |
| Weeks 9–11 | tROAS raised to 3.0x, budget incrementally increased to $26,000/month | Final ROAS: 3.2x; campaign declared successful |
Results: Before-and-After Metrics with Specific Numbers
The headline number is a 3.2x ROAS at the end of Week 11, up from a baseline of 1.9x — a 68% improvement in return on ad spend. But several supporting metrics tell a more complete story about what actually changed.
Click-through rate on AI Mode placements averaged 6.1%, compared to 1.8% on the brand's legacy Shopping campaigns. Industry observers have noted that CTR figures for conversational ad placements tend to run higher because the user is already in an active research mindset when the ad surfaces — this brand's data aligned with that observation. Cost per click came in at $0.94 on average, slightly below the $1.12 average the brand had been paying on standard Shopping.
The more striking finding was in conversion rate. AI Mode visitors converted at 4.3%, compared to 2.1% on legacy Shopping traffic — more than double. The team attributed this to query intent quality: users arriving through AI Mode had already engaged with a multi-turn conversational session about their product category, arriving at the brand's product page with significantly more context and purchase readiness than a standard keyword-triggered click.
| Metric | Before (Baseline) | After (Week 11) | Change |
|---|---|---|---|
| Blended ROAS | 1.9x | 3.2x | +68% |
| Conversion Rate | 2.1% | 4.3% | +105% |
| Avg. CPC | $1.12 | $0.94 | −16% |
| CTR | 1.8% | 6.1% | +239% |
| Monthly Ad Spend | $22,000 | $26,000 | +18% |
| Revenue Attributed | $41,800 | $83,200 | +99% |
It's worth noting a methodological honesty point: the team acknowledges that some revenue lift may reflect seasonal demand increases in the period studied (Q1 to mid-Q2 2026). They applied a conservative 15% seasonal adjustment to their internal analysis, and even under that adjustment, the ROAS improvement remained significant at approximately 2.9x adjusted — still comfortably above profitability threshold.
Key Learnings: What Worked, What Failed, and What Was Unexpected
What worked: Loading rich first-party audience signals before launch was the single most impactful setup decision. During the learning phase, the algorithm appeared to use customer match data aggressively to identify high-intent users entering AI Mode sessions about relevant product categories. Campaigns that lacked these signals in informal comparisons shared by other practitioners took two to three weeks longer to exit the learning phase.
What failed: The initial creative approach — importing existing Shopping ad headlines verbatim — performed poorly in Week 2. AI Mode queries are phrased as natural language questions, and assets written as keyword-dense product titles felt tonally mismatched. After rewriting descriptions as direct answers to common purchase questions, asset performance scores improved and CTR increased by 1.9 percentage points within a single week.
What was unexpected: The multi-session conversion path data surfaced a behavior the team had not anticipated. Approximately 31% of converting users clicked an AI Mode ad, did not purchase, returned via a direct URL type-in two to five days later, and then converted. Under last-click attribution, these conversions would have been attributed to direct — not to paid search at all. The GA4 multi-session path report revealed that AI Mode was initiating nearly a third of all purchases that the standard reporting model had been crediting elsewhere.
"We were essentially undercounting AI Mode's contribution by about 30% using last-click. The channel wasn't underperforming — our measurement was."
Google's expansion of AI Mode ad placements is continuing globally. According to Google's official ads blog, the company is scaling AI Mode ads in markets including Singapore, Malaysia, and the Philippines, though market-specific performance benchmarks have not been disclosed. For brands operating in those markets, this case study's measurement methodology is likely more transferable than its raw ROAS numbers, which will vary significantly by category and competitive density.
How to Replicate It: An Actionable Checklist
These are the specific steps, in sequence, that produced the results above. Skip or shortcut any of them and the data quality — and likely the performance — degrades.
- Audit and rebuild conversion tracking first. Before any campaign goes live, verify that your Google Ads conversion tags and GA4 purchase events are firing without duplication. Enable enhanced conversions and consent-mode v2. Expect to find discrepancies — this brand found 18% double-counting.
- Set up multi-session path reporting in GA4. Create a custom exploration that tracks conversion paths across multiple sessions. Without this, AI Mode's contribution to assisted conversions will be systematically undercounted.
- Load first-party audience signals before launch. Upload customer match lists (minimum 1,000 matched users recommended), website visitor segments, and a custom intent audience built from your 40–50 highest-intent search terms.
- Start tROAS at 10–15% below your profitability threshold. Give the algorithm room to gather conversion data before you tighten the target. Raise in 0.2x increments every two weeks once you're hitting 50+ weekly conversions.
- Write creative assets as answers, not product titles. Frame headlines and descriptions around the questions your customers ask, not the keywords you want to rank for. AI Mode users have already stated their intent in natural language — your assets should match that register.
- Run the test clean. Pause competing campaigns for the duration of your measurement window. Cross-channel bleed makes attribution impossible and will cause you to either over- or under-credit AI Mode results.
- Rotate five or more creative variants per category and kill low performers at Week 4. Asset performance scores in AI Mode placements stabilize within three to four weeks. Prune aggressively to let budget concentrate on the strongest variants.
- Apply a seasonal adjustment to results before reporting upward. Quarter-on-quarter demand shifts can inflate or deflate apparent ROAS gains. Build this adjustment into your internal reporting to maintain credibility.
Frequently Asked Questions
How much budget do you need to test Google AI Mode ads effectively?
Industry practitioners generally suggest a minimum of $8,000–$10,000 per month to generate enough conversion volume for the tROAS bidding algorithm to exit its learning phase within a reasonable timeframe. Below this threshold, the campaign may remain in a perpetual learning state, producing unreliable results. The brand in this case study launched at $22,000 per month, which accelerated data accumulation significantly — but smaller budgets can work if expectations around timeline are adjusted accordingly.
Can you run Google AI Mode ads separately from Performance Max?
Yes — and for testing purposes, it is advisable to do so. Performance Max campaigns can include AI Mode inventory as part of their broader placement mix, but they do not expose placement-level breakdowns that would allow you to isolate AI Mode performance specifically. Running a dedicated campaign structure gives you the measurement clarity needed to make informed budget decisions.
How do you measure conversions accurately in Google AI Mode?
Accurate measurement requires enhanced conversions with hashed first-party data, a correctly configured GA4 purchase event, and a multi-session attribution report to capture conversions that begin with an AI Mode click and complete in a later session. Standard last-click attribution materially undercounts AI Mode's contribution — in this case study by approximately 31%. A full breakdown of the technical setup is covered in the guide to Google AI Mode ad conversion tracking.
Is a 3.2x ROAS realistic for all e-commerce categories in Google AI Mode?
Not universally — ROAS benchmarks vary significantly by product category, average order value, competitive density, and audience data quality. The home goods category benefits from high-intent, research-heavy purchase journeys that align naturally with AI Mode's conversational format. Categories with impulse purchase dynamics or very low AOVs may see different results. The methodology and measurement framework in this case study are more universally applicable than the specific ROAS figure.
