An article gets searched 10,000 times, 400 people click in, 120 read it through, and only 2 actually leave their details. Every layer is leaking, but only the layer leaking worst deserves your attention first.
Break the content funnel into four levels — impressions, clicks, reading, conversion — work out the conversion rate at each level, then compare it against your own historical average. The level that’s dropping steepest is the only one you should fix this period; leave the other three alone. That gives you the best return on effort.
Build a four-level funnel you can align
Data comes from two systems: Search Console provides impressions and clicks, GA4 provides reading depth and conversion. The key move is aligning both sides of the data to the same URL — don’t roll up by section or site, or the leak points get averaged away and everything looks fine.
| Level | Data source | Key metric | Where to pull it |
|---|---|---|---|
| Impressions | Search Console | Impressions | Performance → Pages → filter by URL |
| Clicks | Search Console | Clicks, CTR | Same, enable the CTR metric |
| Reading | GA4 | Engaged sessions rate, avg engagement time | Reports → Engagement → Pages and screens |
| Conversion | GA4 | Key event completions | Admin → Events → mark as key event |
If you haven’t configured key events in GA4 yet, the last level of this funnel is empty. First follow the GA4 key events setup guide to mark form submissions, downloads, add-to-cart, and similar actions, then come back to build the funnel.
Pulling data weekly is enough — daily granularity is too noisy. Build a long-running sheet, one row per URL and one column per level; after 8 weeks, the trend line explains far more than any single data point.
Work out the drop rate at each level
Conversion rate per level = this level’s volume ÷ the previous level’s volume. For a content site, start by positioning with the reference lines below, then gradually switch to your own historical averages:
- Impressions → clicks (CTR): 3%–8%; below 3% is usually a title or SERP-format problem
- Clicks → reading (engaged sessions rate): 40%–60%; below 40% means content and intent are misaligned
- Reading → conversion: 1%–3%; below 1% is a CTA or landing-page problem
Reference lines are just a starting point. News-type and tool-type sections can naturally differ in CTR by more than double; applying one site-wide standard hurts pages that were already fine. The safe approach is to use the same section’s 90-day median as the baseline, and only judge a leak when it deviates by more than 30%.
There’s also one scenario worth identifying separately: impressions rising, clicks flat, ranking holding. That’s usually not the page getting worse — it’s the SERP gaining a snippet box or video slot that squeezes out clicks. In these ranking-traffic decoupling cases, rewriting the title helps little; you need a different approach to fight for the snippet slot.
How to fix a leaking click layer
This layer costs the least — you don’t touch the body copy, and changing two lines of metadata may be enough, so it deserves to always go first. The concrete approach is preparing two versions of the title for the same post: one with numbers and results (“5 steps to find why traffic dropped”), one with a pain-point question (“Traffic halved — where do you check first?”), launching them two weeks apart and comparing CTR for the same period.
- Change only the title and description, keep the URL unchanged, to avoid creating duplicate content
- Let each version run at least 500 impressions before drawing conclusions — tiny samples are too noisy
- Write in the description what the reader gets after reading, don’t repeat the title
- Check whether it gets truncated in the SERP; keep Chinese descriptions under 78 characters
How to fix a leaking reading layer
When users click in and leave within seconds, the problem is almost always in the first two screens. Move the conclusion, target audience, and prerequisites all into the first 200 characters of the body; add a table-of-contents anchor for long articles so readers instantly see “the section I need is the third one”.
- Give the answer in the first paragraph, no background lead-in
- Write H2 titles as conclusion sentences, not noun phrases
- Add a TL;DR box for articles over 1,500 words
- Don’t let a big image fill the mobile first screen; place images after the first paragraph
There’s another leak that’s easy to miss: the body doesn’t deliver what the title promised. The title says “template download” but the body only talks methodology — readers scan two screens, see no template, and leave. The check is simple: search for each noun in the title within the body; anything that doesn’t turn up is a promise left unfulfilled.
How to fix a leaking conversion layer
If reading volume is fine but leads are few, the problem is in the CTA’s position and how concrete the incentive is. Move the form up from the end, placing it after the section where readers have absorbed the core method and feel the strongest agreement; change the incentive from “free download” to a measurable benefit description.
| Change | How it works | Observation period |
|---|---|---|
| CTA moved up to the middle of the body | Catches intent at the peak of agreement | 2 weeks |
| Incentive changed to concrete benefit | Lowers the cost of judging “is it worth filling in” | 2 weeks |
| Form fields cut to 3 or fewer | Reduces filling friction | 1 week |
| Instant feedback text after submission | Prevents users misjudging failure and resubmitting | 1 week |
Change only one thing each time, log the date, and two weeks later recheck key event counts on the same batch of pages. If you change three things at once, you’ll never know which one worked — and you can’t replicate it to other pages.
Once the funnel runs smoothly, feed each level’s data into the content ROI calculator and use the same yardstick to decide which topics keep getting investment and which get cut outright.
When it actually comes to execution, the flow is short: first pick your 3 highest-traffic articles and pull each level’s data into one sheet; work out the conversion rate per level, compare with the section’s 90-day median, and circle the level deviating most; this week, fix only that level, noting ideas for other layers in your backlog; if the leak is in the click layer, prepare two title versions for a two-week comparison; after two weeks, review the funnel sheet, confirm that level is back at baseline, then move to the next layer. One point at a time per round — when a fix works you can replicate it, when it doesn’t you know which step to blame, so you never waste effort in the dark.
FAQ
How many levels does a content funnel have?
Four: impressions, clicks, reading, and conversion. Search Console provides impressions and clicks, GA4 provides reading depth and conversion; align by URL, work out conversion at each level, and fix the steepest-drop level first.
What CTR counts as a leak?
3%–8% from impressions to clicks is the normal range; below 3% is usually a title or SERP-format problem. More robust: compare with the same section’s 90-day median, and only judge a leak when it deviates by more than 30%.
Users click in and leave within seconds — which layer is leaking?
The reading layer. Move the conclusion, target audience, and prerequisites into the first 200 characters, write H2s as conclusion sentences; also check whether the body delivers the title’s promise — “template download” in the title with only methodology in the body is an unfulfilled promise.
Reading volume is fine but leads are few — how to fix?
The conversion layer. Move the CTA from the end to the middle where agreement peaks, change the incentive from “free download” to a measurable benefit description, cut form fields to 3 or fewer, and change one thing at a time with two weeks of observation.
