Organic traffic rose 30% in three months, lead form submissions didn’t budge. That’s where conversion funnel analysis comes in: break down the path from search results to lead capture, see clearly at which layer people drop off, rather than vaguely saying “conversion rate is low.”
Cut the path into five layers: impressions, clicks, reading, intent, lead capture — each layer measured by only one metric. When finding bottlenecks, look at absolute lost user count, not percentage — because a drop from 80% to 60% at a high-base layer loses far more than a 5% to 3% drop at the tail. First fix the layer with the largest loss volume, change only one thing at a time, look at the next layer after four weeks.
Five-layer breakdown: each layer answers only one question
More layers isn’t better. After cutting to 7-8 layers, each layer has insufficient sample, data jitter will overwhelm real differences. Five layers is the common choice balancing granularity and statistical stability.
- Impression layer: did content appear in front of users, look at search impressions
- Click layer: did title and snippet make people want to click, look at CTR
- Reading layer: did the landing page retain people, look at effective reading ratio, don’t use bounce rate
- Intent layer: did they trigger a next-step action, look at key event trigger rate
- Lead capture layer: did the form submit successfully, look at submission completion rate
The reading layer metric is most easily chosen wrong. Bounce rate counts people who only viewed one page but read the full article as failures — “effective reading” combined from scroll depth plus dwell time is closer to reality. Content-side细分口径 can reference content funnel analysis, that article breaks it down more finely by content type.
Pair each layer with one metric, don’t糊弄 with session count
| Layer | Main metric | Healthy range | Data source |
|---|---|---|---|
| Impressions | Search impressions | Year-over-year by topic cluster not declining | Search Console |
| Clicks | Search CTR | First page above 4% | Search Console |
| Reading | Ratio scrolling past 50% and staying 30s | 35%–55% | GA4 custom events |
| Intent | Key event trigger rate | 8%–15% | GA4 key events |
| Lead capture | Form submission completion rate | 40%–65% | Form tool or events |
Healthy ranges need to be reset for your own business. B2B solution-type content’s intent layer trigger rate can reach 20%, tool-type traffic content may only be 3% — applying someone else’s benchmark to yourself leads to wrong conclusions. After running a full quarter, replace with your own median.
Find bottlenecks by absolute loss volume, not conversion rate
A real example. One topic cluster’s one-month funnel data is below — by intuition most people would go fix the last layer, because that layer’s percentage looks the worst.
| Layer | Entering users | Flowing to next layer | Layer conversion rate | Absolute loss |
|---|---|---|---|---|
| Impression to click | 120,000 | 3,600 | 3.0% | 116,400 |
| Click to reading | 3,600 | 1,300 | 36.1% | 2,300 |
| Reading to intent | 1,300 | 104 | 8.0% | 1,196 |
| Intent to lead capture | 104 | 47 | 45.2% | 57 |
Last layer conversion rate 45.2%, looks okay; first layer 3.0%, looks bad. But raising CTR from 3.0% to 4.2% gives 18 extra leads at the funnel end; raising the last layer from 45% to 55% only gives 10 extra. Which layer to fix — the table already gives the answer.
Three types of high-frequency leaks and corresponding fixes
- Click layer leak: ranking on first two pages but nobody clicks — change the first 30 characters of the title, add numbers and specific benefits, feedback visible within two weeks
- Reading layer leak: come in and leave immediately — check if first screen starts with three paragraphs of background, move conclusion to the beginning, change long tables to horizontally scrollable components
- Intent layer leak: finished reading but no action — mostly because the page simply doesn’t have a suitable next step, insert strongly related resources in the middle of body text rather than a site-wide unified signup button
- Lead capture layer leak: form fields over 5, requires phone number, doesn’t clearly state what happens after submission — cut them one by one
Intent layer action definitions directly determine whether data is usable. Which behaviors count as key events, how to pass parameters — it’s recommended to first uniformly name according to GA4 key event setup standards, otherwise after three months nobody can understand the tracking points they buried themselves.
Build the funnel into reports, not manually pulling every time
The value of funnel analysis comes from continuous observation — one-time data pulling only gives a snapshot. When landing, fix the five-layer metrics into one weekly report, presented grouped by topic cluster, so each content lead can see the leak points in their own part.
- Impression and click layers pulled from Search Console API, aggregated by landing page then mapped to topic clusters
- Reading to lead capture three layers pulled from GA4, use page path as association key, paths with parameters first cleaned
- Each layer simultaneously records current period value, previous period value, four-week average — only looking at single period gets fooled by weekend fluctuations
- Annotate current bottleneck layer for each topic cluster, change records written in the same table, convenient for post-hoc attribution
Sites with multiple channels coexisting should also note one thing: lead capture at the funnel end often isn’t solely driven by organic search — users may have first read content then clicked in from ads. In this case, last-click attribution underestimates content value — the attribution model selection article gives several fairer allocation methods.
Next-step action checklist
- Pick the topic cluster with the largest traffic, fill the last 30 days of data into the table by five-layer structure
- Calculate absolute lost user count for each layer, sort descending by loss volume,圈定 the first-priority layer to fix
- Make only one change targeting that layer, simultaneously record change content and date in shared document
- Four weeks later pull the funnel again, compare loss volume changes, confirm effective then推广 to other topic clusters
- Fix the five-layer metrics into weekly report template, annotate current bottleneck layer for each topic cluster
FAQ
How many layers should a conversion funnel be cut into?
Five layers is enough: impressions, clicks, reading, intent, lead capture — each layer only uses one metric. Cutting to 7-8 layers leaves insufficient sample per layer, data jitter overwhelms real differences.
How to find the layer most worth fixing?
Look at absolute lost user count, not percentage. A one-point drop at a high-base layer loses far more than at the tail — first fix the layer with largest loss volume, change only one thing at a time.
Can the reading layer be measured by bounce rate?
No. Bounce rate counts people who read the full article as failures — “effective reading” ratio combined from scroll depth plus dwell time is closer to reality.
What are common reasons nobody submits the form?
Fields over 5, mandatory phone number, doesn’t clearly state what happens after submission. Putting strongly related content entry in the middle of body text is more effective than a site-wide unified signup button.
Can healthy ranges directly apply someone else’s benchmark?
No. B2B content intent layer trigger rate can reach 20%, tool traffic may only be 3% — after running a full quarter, replace benchmarks with your own data median.
