GA4 Explorations for Paid Traffic in 2026: Funnels, Paths, Cohorts

Standard GA4 reports answer “how many,” which is exactly why GA4 Explorations exists. They do not answer “where the money leaks.” You can see that a campaign brought 4,000 sessions and 38 conversions. You cannot see which step lost the other 3,962, what separated the people who converted, or which page systematically ejects paid traffic from the funnel.

That is what Explorations are for — a report builder that sits next to standard reports but works on entirely different logic. This is a practical method: what must be configured before you open the section, four specific explorations built for paid traffic, the limits and interpretation traps that produce wrong conclusions, and how to turn findings into actual changes in Google Ads.

Explorations are not a replacement for Google Ads reporting, and the two are not supposed to match. Their job is to show behavior between the click and the conversion — precisely the part of the funnel the ad platform cannot see.

Prerequisites before you build anything

Explorations do not repair data. They display what was collected, faithfully including its flaws. Minimum checklist:

• Google Ads and GA4 linked, auto-tagging on. Without it your paid traffic collapses into an undifferentiated google / cpc bucket with no campaign detail.
• Meaningful key events. A single generic “conversion” event guarantees useless funnels. You need distinct steps: product view, add to cart, begin checkout, form start, form submit, purchase.
• Parameters attached to events so you can break results down later: product category, form type, checkout step, plan tier.
• Consent Mode configured properly, and a clear idea of what share of your data is modeled. Otherwise some of your conclusions describe a model, not people.
• A known baseline gap versus Google Ads. A steady 10–20% difference over time is normal given different attribution models and windows. A gap that appears overnight is a tracking issue — see our guide to diagnosing conversion tracking.

If your site fires exactly two events — session start and purchase — do not start with Explorations. Start by instrumenting the intermediate steps. The selection logic is covered in our piece on micro-conversions and funnel optimization.

The five GA4 Explorations types, and when each earns its place

Type	Question it answers	Use it when
Free form	How any metric relates to any dimension	Ad hoc slices standard reports do not offer
Funnel exploration	Which step loses users	Your primary tool for paid traffic
Path exploration	Where people actually go	The funnel showed a drop-off and you need to know where they went instead
Segment overlap	What distinguishes one group from another	Finding traits correlated with conversion
Cohort exploration	How groups acquired at different times behave over time	Judging traffic quality beyond the first session

For paid traffic, four of these do real work: funnel, path, segment overlap and cohorts. Free form is a support tool for testing a hunch quickly.

Exploration 1: the paid traffic funnel

Goal: see exactly where ad-driven visitors fall out, and compare that against organic.

How to build it

1. Create a new Funnel exploration.
2. Define steps from broad to specific. Ecommerce: session start → product view → add to cart → begin checkout → purchase. Lead gen: session start → key page view → form interaction → form submit → qualified lead, if you feed status back.
3. Choose funnel type. Open counts anyone who reached a step by any route. Closed enforces strict sequence. Default to open — closed funnels routinely “lose” users who simply arrived in a different order.
4. Add a breakdown by campaign name or ad group.
5. Apply a paid-traffic segment (source/medium containing cpc, or your specific source).
6. Enable elapsed time so you can see how long each transition takes.

What to read

• The largest drop. Losing 60–70% between product view and add to cart is ordinary. Losing 60% between begin checkout and purchase is a checkout problem, not an advertising one.
• Differences between campaigns. If one campaign converts half as well at the same step, the issue is query-to-landing-page fit, not the site as a whole.
• The gap against organic. If paid traffic drops out noticeably earlier than organic at the same step, your ad promises something the landing page does not deliver. Fixes are in our breakdown of landing page conversion for paid traffic.
• Elapsed time between steps. An abnormally long transition usually signals friction rather than deliberation: slow loads, validation errors, re-entering details.

Exploration 2: reverse path from conversion

The funnel tells you where people left. Path tells you where they went instead.

The most valuable configuration is the reverse path: set the ending point to your purchase or form submit event and look backward at the screens and events that preceded it. This reveals the routes people actually take, which are rarely the routes designed into the site.

The second configuration worth running is a forward path from paid landing pages. What to look for:

• Mass exits into non-commercial content. If a third of ad traffic detours into the blog and never returns, your landing page is advertising the blog too effectively.
• Loops back to a prior step. A form → error → form cycle is validation friction, visible in one glance.
• Pages where sessions terminate. Candidates for technical review: speed, mobile layout, embedded payment widgets.
• Site search immediately after landing. People are hunting for what the ad promised and the page did not show. The internal search terms report becomes a ready-made list of ad copy corrections.

Exploration 3: segment overlap

This answers “how do buyers differ from everyone else.” Build three segments and inspect the intersections:

1. Users from paid traffic.
2. Users who reached a key step, such as viewing the pricing page.
3. Users who converted.

Then add a fourth hypothesis segment: mobile devices, a particular metro area, new versus returning, or people who read a specific page. The overlaps show which traits track most closely with conversion.

Typical findings: mobile converts at half the rate on equal traffic, pointing at interface problems; visitors who reached pricing convert several times better, arguing for surfacing pricing directly from the landing page; returning users drive most conversions, meaning remarketing is underfunded relative to its contribution.

One caution that matters: correlation here is not causation. The fact that people who read the reviews page buy more often does not mean forcing everyone onto that page will lift sales. Findings generate hypotheses; experiments confirm them.

Exploration 4: cohorts by acquisition week

Cohort analysis shows what happens after the first visit — the part of traffic quality that cost per acquisition completely hides.

Configuration: cohort inclusion by first touch, return criterion by any engagement or by a specific key event, weekly granularity, metric set to active users or revenue per user. Break down by campaign.

What this surfaces in practice: campaign A produces cheap conversions whose users never come back; campaign B costs more on first purchase but its cohorts stay alive for months. If budget decisions run purely on cost per conversion, money flows steadily toward campaign A — and that is the most common explanation for “our ads look profitable but the business is not growing.” Connecting this view to attribution models is covered in our guide to GA4 attribution for media buyers.

Segments, breakdowns and limits

The practical constraints people trip over (directional — verify in your own property):

• Up to 10 segments per exploration and 10 breakdown values. Beyond three or four, the report stops being readable anyway.
• Row limits apply per tab, and large datasets get sampled, with an indicator shown in the report. Sampled data is fine for spotting anomalies, unacceptable for numbers you present as fact.
• Explorations are personal by default. Share them explicitly if the team needs access.
• Historical depth is bounded by your user and event data retention setting. Left at the shortest option, long cohorts simply cannot be built — check this before planning retention analysis.

Hitting sampling and row limits is the signal to move to warehouse-level analysis. The Google Ads to BigQuery pipeline removes interface constraints and lets you compute anything against raw data.

Six ways to misread an exploration

1. Comparing GA4 conversions to Google Ads conversions head-on. Different attribution, different windows, different definitions. A discrepancy is expected; identical numbers would be more suspicious.
2. Using a closed funnel for a non-linear journey. A user who lands directly on a product page from an ad falls out of a funnel that starts at the homepage. The report shows catastrophe; reality shows a configuration error.
3. Ignoring the modeled share. Under strict consent settings, part of the behavior is estimated. For small segments this makes conclusions unstable.
4. Confusing sessions with users. GA4 funnels count users. Session-based intuition produces systematically wrong percentages.
5. Drawing conclusions from tiny numbers. Four conversions versus seven is noise, not a finding.
6. Stopping at the finding. The report locates a problem; it does not prove a cause. Causes are established by experiment.

Free form: fast hypothesis checks

The fifth exploration type gets ignored because it looks like an ordinary table. Its value is speed: a question with no ready-made view in standard reports takes about two minutes to answer.

Three configurations that repay the effort most often:

• Landing page × campaign × conversion rate. Immediately shows which campaign-to-page pairings work and which merely spend. The common discovery: half the campaigns point at one generic page because that was easier at some point.
• Device × city × key event. Catches localized failures invisible in aggregate — mobile conversion collapsing in one metro, which turns out to be a payment method unavailable there.
• Day of week × hour × conversions and engagement. The evidence base for ad scheduling, provided the pattern holds for several weeks rather than one.

A good habit is keeping one permanent “sandbox” free-form exploration and never deleting its tabs. Within a couple of months it accumulates a personal library of slices that replaces a large share of manual analysis.

Case sketch: how a funnel finds lost leads

A composite but recognizable lead-gen scenario. Campaigns are stable, cost per lead is creeping up, and nothing in the ad account looks wrong.

1. Build the funnel: session → service page view → form interaction → form submit. Break down by campaign, segment to paid traffic.
2. The numbers: 18% reach the form, 11% start filling it in, 3% submit. The collapse sits between starting and submitting.
3. Enable elapsed time: non-submitters spent minutes on that step, so they were genuinely trying.
4. Switch to path exploration: many users re-view the same page right after interacting with the form — the loop signature of a validation error.
5. Reproduce on a phone: the phone field rejects spaces in the number, and the error message renders below the fold.

The conclusion: nothing was wrong with the advertising. Fixing one form field delivers a lift no bid optimization could match. And the inverse is the real lesson — had the team scaled that week, every extra dollar would have flowed through the same broken form.

Turning findings into Google Ads changes

What the exploration shows	What to change in the account
A campaign’s traffic dies at the very first step	Audit query-to-ad-to-landing-page fit, add negatives, point the ad at a more relevant URL
Drop-off concentrated on mobile	Device bid adjustment as a stopgap; fixing the mobile experience as the real solution
One geography converts far worse at equal volume	Check presence-based targeting, then exclude or de-prioritize if confirmed
One campaign’s cohorts retain longer	Reallocate budget on user value rather than first-conversion cost
Most conversions happen after several visits	Strengthen remarketing, revisit conversion window and attribution model
Visitors use site search immediately after landing	Rewrite ad copy and the landing headline around the query people actually have

One class of finding is not about advertising at all: if you see breaks at technical steps, duplicated events, or parameters disappearing, the problem is data collection. The durable fix is server-side tagging with sGTM.

A workable cadence

• Weekly: the paid funnel broken down by campaign — ten minutes, watching only whether the largest drop-off step moved.
• Monthly: cohorts and segment overlap, feeding budget allocation decisions.
• On any performance drop: reverse path from conversion, to determine whether behavior changed or the traffic changed.
• Before every scale-up: confirm the bottleneck is not on-site, or you will simply multiply the leak.

That last one carries the most money. Scaling a campaign whose ceiling is a broken checkout is the most expensive way to fund someone else’s problem. Related reading: invalid clicks and budget protection, and for shopping accounts, Merchant Center product disapprovals. If you want an outside review of the campaigns-to-analytics-to-site chain, that is part of what we do at PPC Rebels.

FAQ

How are Explorations different from standard GA4 reports?

Standard reports are fixed views for quick answers. Explorations are a builder: your own funnel steps, segments, breakdowns and visualization types unavailable elsewhere. The trade-off is setup effort and the fact that they are not automatically shared with your team.

Why don’t GA4 conversions match Google Ads?

Different attribution models, conversion windows, timestamps and treatment of modeled data. A steady 10–20% gap is normal. What matters is not the gap but a sudden change in it.

Open or closed funnel?

Default to open. It counts users who reached a step by any route, which reflects how ad traffic actually behaves. Closed funnels are for genuinely sequential processes such as multi-step checkout.

How many segments can I use?

Directionally up to ten per exploration, with a similar limit on breakdowns. In practice anything past three or four makes the report unreadable.

What do I do about sampling?

Narrow the date range, narrow the segment, or move to raw data in BigQuery. Sampled reports are fine for finding anomalies, not for figures you put in front of leadership.

Can I analyze specific Google Ads campaigns?

Yes, with the accounts linked and auto-tagging enabled — campaign, ad group and keyword dimensions become available. Without the link, paid traffic appears only as a generic source.

How do I judge traffic quality beyond the first session?

Cohort exploration: inclusion by first touch, return by a key event, weekly granularity, broken down by campaign. Campaigns whose cohorts decay fast are buying cheap one-time conversions.

Do Explorations work retroactively?

Yes, within your configured data retention period. If retention is set to the minimum, long cohorts and historical analysis cannot be built — verify this setting before you plan the work.

How can I tell whether it is a site problem or an ad problem?

Compare the paid funnel with the organic funnel at the same steps. Equal drop-off in both means the site. Drop-off only in paid means a mismatch between the ad and the landing page.

Do Explorations replace Google Ads reporting?

No. The ad platform owns auction, bids and cost; GA4 owns post-click behavior. Decisions live at the intersection, and forcing the two systems to agree numerically usually produces worse decisions, not better ones.

Do I need BigQuery if I have Explorations?

Not initially. While you are inside row limits and free of sampling, the interface is enough. BigQuery becomes necessary at scale, when joining CRM data, or when you need reproducible reporting.

Where should a beginner start?

One funnel exploration on paid traffic with four or five steps, broken down by campaign. That single report typically surfaces as many actionable problems as everything else combined.