Attribution Modeling for SEM: GA4 Path Analysis

Attribution is the marketing problem that doesn’t have a clean solution. Every model is wrong; some are useful. The job of an SEM-focused marketer isn’t to find the “true” model — it’s to understand the trade-offs of each and make budget decisions that don’t get destroyed by attribution bias.

This guide walks through how attribution works in GA4 in 2026, the practical implications for SEM budget allocation, and the path analysis features most teams don’t use.

The four attribution models that matter

GA4 supports several attribution models. The ones you’ll actually use:

1. Last-click (last-non-direct-click): 100% credit to the last marketing touchpoint before conversion. Direct traffic gets credit only if no other touchpoint exists.

2. First-click: 100% credit to the first marketing touchpoint that brought the user into your funnel.

3. Linear: equal credit across all touchpoints in the path.

4. Time-decay: more credit to touchpoints closer to conversion.

5. Position-based: 40% to first, 40% to last, 20% spread across middle.

6. Data-driven (DDA): GA4’s machine learning attributes credit based on actual contribution to conversion lift. This is GA4’s default since 2024.

For SEM specifically, the most consequential comparison is last-click vs DDA.

Last-click: simple but biased

Last-click attribution is the historical default. It’s cheap (just look at the last source), intuitive, and used by Google Ads natively for default reporting.

Problems:

• Massively overcredits branded keywords and bottom-funnel campaigns
• Massively undercredits awareness channels (display, paid social, YouTube)
• Encourages you to defund anything that doesn’t directly close
• Misses the assist value of mid-funnel touches

A B2B account we audited had spent 18 months scaling branded paid search because last-click showed it had the best ROAS. After moving to DDA, branded was revealed as ~70% air (those users would have arrived organically anyway), and an upper-funnel YouTube campaign that looked terrible by last-click was actually generating most of the new pipeline.

Data-driven attribution (DDA): better but opaque

DDA uses your account’s actual conversion data to model how much each touchpoint contributes to conversion lift. Built on Google’s “Markov chain” approach.

The mechanic: GA4 compares paths that converted vs. paths that didn’t. Touchpoints that consistently appear in converting paths but not in non-converting paths get more credit. Touchpoints that appear in both equally get less.

Advantages:

• Distributes credit more fairly across the funnel
• Reflects actual contribution rather than chronological position
• Automatically adjusts as your channel mix evolves

Trade-offs:

• Requires sufficient conversion volume (~300+ conversions/month for stable model)
• Less transparent — you can’t easily explain why a specific path got specific credit
• Can drift if conversion behavior changes seasonally
• Still client-side based at its core, so it’s susceptible to tracking gaps

For most accounts above 300 conversions/month, DDA is the right default.

Where to find each report in GA4

Attribution → Model comparison: side-by-side comparison of conversions and value across different models. Most useful single report for understanding model impact.

Attribution → Conversion paths: literal paths users took before converting. Shows the actual sequences.

Acquisition → Traffic acquisition (with the conversion model selector): channel-level attribution. Toggle between models to see how channel contribution shifts.

Advertising workspace: aggregate paid performance with attribution baked in.

If you’re not familiar with the Attribution section, spend 30 minutes there before reading further. The conceptual understanding is much easier with the actual UI.

How attribution choice changes SEM budget allocation

Consider a B2B SaaS with three campaigns:

• Brand search (defends brand queries)
• Non-brand search (targets new prospects on intent queries)
• YouTube TrueView for Action (top-funnel awareness)

Spend: $5K, $10K, $5K respectively per month.

Under last-click attribution:

• Brand: 60 conversions at $83 CPA → looks fantastic, fund more
• Non-brand: 30 conversions at $333 CPA → mediocre
• YouTube: 4 conversions at $1,250 CPA → looks awful, cut

Under DDA:

• Brand: 22 conversions of credit (because most branded clicks were from users already exposed elsewhere)
• Non-brand: 38 conversions of credit (often the actual demand creator)
• YouTube: 34 conversions of credit (massive upstream influence on later searches)

Same campaigns, same conversions. Same dollars. Wildly different conclusions. The CMO who defunded YouTube based on last-click is making the wrong call.

Common attribution mistakes

1. Comparing platforms with mismatched models. Google Ads default reporting (last-click) vs Meta Ads default (data-driven within Meta only) vs GA4 (DDA across all). The same conversion can be counted by all three and look different in each. Pick one source of truth (usually GA4 DDA) and compare apples to apples.

2. Trusting Google Ads conversion column blindly. Google Ads counts conversions optimistically — with significant credit for view-through and aided conversions. Cross-check against GA4 DDA.

3. Setting bid targets in Google Ads based on GA4 attribution. Google Ads bids against Google Ads conversion data, not GA4. If you import GA4 conversions as the bid signal, you’re now using DDA at the bid level — usually correct.

4. Ignoring tracking gaps. iOS Safari ITP, ad blockers, and consent declines erase 10-20% of conversion paths. Models built on incomplete data have systematic biases.

5. Not accounting for offline conversions. B2B closed-won happens in your CRM, not in GA4. Without offline conversion upload, GA4’s DDA is optimizing for form fills, not customers.

6. Optimizing for short paths. GA4 by default only tracks 90-day attribution windows. Long B2B sales cycles (180+ days) need explicit window extension.

Path analysis: the underused feature

GA4’s Conversion Paths report shows the actual sequences users took to convert. Most teams glance at it once and forget. It rewards deeper analysis.

What to look for:

1. Average path length. How many touchpoints before conversion? A 1-touch path is “I knew you and came directly.” A 7-touch path is “long-cycle, multi-channel discovery.” Different path lengths require different channel mix.

2. Most common first interactions. Where do users discover you? If most paths start with paid search, you’re attracting demand from existing search intent. If most start with organic, your SEO is driving discovery. If display/video is consistently first, you’re driving net-new awareness.

3. Most common last interactions. Where do conversions actually close? Often branded organic or branded paid — but the upstream channel that put the brand in their head is the real driver.

4. Channels that appear primarily in middle of paths. These are assist channels — the under-credited heroes of multi-touch journeys.

5. Conversion rate by path length. Are 5-touch paths much higher value than 1-touch? Many B2B accounts have this pattern. Implication: invest in nurture sequences to escalate users through more touchpoints.

The honest limits of any attribution model

Any model — last-click, DDA, position-based, custom — has structural limits:

1. Client-side tracking gaps (~10-25% of users invisible) — same gap across all models
2. View-through is uncertain — did the user really see the ad and decide to search later? Can’t be proven
3. Offline interactions (calls, direct mail, word of mouth) are mostly invisible
4. Multi-device journeys — user researches on phone, converts on desktop. GA4 stitches via signed-in Google account, but only when present.

These gaps are why incrementality tests (lift studies) matter for big budget decisions. Attribution tells you correlations; incrementality tests prove causation. Run a holdout test on a campaign you’re uncertain about before defunding it.

A 30-day attribution upgrade plan

If your team is still operating on last-click and seat-of-pants attribution:

Days 1-7: Audit current state. What model is each platform reporting? What conversion windows? Which conversions are flowing where? Document everything.

Days 8-14: Configure GA4 properly. Enable DDA. Set conversion windows to match your sales cycle (90 days for short cycles, 180-365 for B2B). Set up Enhanced Conversions on both web and (for B2B) leads.

Days 15-21: Cross-check. Compare same conversions across Google Ads, Meta, GA4 last-click, GA4 DDA. Where do they diverge? Identify the source of truth.

Days 22-30: Stakeholder education. Re-run last quarter’s reporting under DDA. Walk leadership through how budget allocation conclusions change. Build buy-in for the new model.

By day 30, you have a defensible attribution framework and (often) a meaningfully different view of channel contribution.

Frequently asked questions

Should I report to leadership on DDA or last-click? DDA if your team is sophisticated; last-click if leadership comes from a sales/traditional marketing background where last-click is the mental model. Or report both, side by side, so the audience can see the difference.

Does Google Ads use GA4 attribution when bidding? Google Ads bids on conversion data imported from GA4 — including the attribution model GA4 applies. So if GA4 is on DDA, Google Ads Smart Bidding is effectively bidding on DDA-attributed conversions.

How do I measure brand keyword incrementality? Run a holdout test: pause branded paid in one geography for 4-6 weeks, monitor organic clicks on the same queries. The lift in organic minus the loss in paid equals the incremental cost of branded paid. Most accounts find branded paid is 30-60% incremental, 40-70% cannibalistic of organic.

Can I customize attribution rules in GA4? You can pick from preset models. True custom attribution requires exporting to BigQuery and building your own model — significant engineering work.

Why do my Google Ads and GA4 conversion numbers never match? Different attribution windows (Google Ads typically uses 30-day click; GA4 default is 90), different model assumptions, different deduplication. 10-20% variance is normal; >30% means a config issue.

---

Attribution doesn’t have a “right” answer, but the wrong attribution model can produce a year of bad budget decisions. Most SEM teams should be running on DDA in GA4 by default, supplementing with incrementality tests on high-stakes decisions. If your team is still operating on last-click in 2026, that’s a quick-win audit waiting to happen.