How to Write a Guide People Actually Finish
Completion rate is the metric nobody reports. Here's what moved mine from 22% to 61% without shortening a single article.
Completion rate on my long articles went from 22% to 61% over about a year. I didn't shorten a single one — the longest piece on this site is 4,000 words and it has the highest completion rate of anything I've published.
Here's what I measured, where readers were actually leaving, and the five changes that moved the number.
The metric nobody reports
Everyone reports pageviews and time on page. Almost nobody reports what proportion of readers reach the end, which is strange, because it's the closest available proxy for whether the thing was worth reading.
Time on page is actively misleading here. A confused reader re-reading a paragraph three times generates excellent time-on-page and a terrible experience. Completion rate has the opposite bias, which makes the two useful together and either one dangerous alone.
How to measure it without a surveillance apparatus
A single scroll-depth marker near the end of the article body — not the footer, which people hit by accident. Fire one event when it's been visible for a few seconds. That's the entire implementation, it's a handful of lines, and it needs no personal data at all.
Add markers at roughly 25%, 50% and 75% and you get a drop-off curve, which is where the actual information is. The single completion number tells you that there's a problem; the curve tells you where.
Where people actually left
My curve had three cliffs, and none of them were where I expected:
- The first screen, before any content. About 30% of the total loss. Header image, byline, tags, share buttons, a newsletter box — the reader had to scroll past furniture to find out whether the article answered their question.
- The first long paragraph. A visible wall of text around the 20% mark reliably shed readers. Same words, split into three paragraphs: substantially better retention.
- Immediately after the first subheading that didn't match their question. People were scanning headings to check the article was about their specific problem, not reading linearly. A vague heading read as "this isn't for me".
The five changes that moved it
1. Answer the question in the first two paragraphs. Not a teaser — the actual answer. The fear is that people will leave once they have it. In practice they stay, because you've proved the rest is worth reading. This was the single biggest change: roughly half the total improvement.
2. Cut everything above the first sentence. Title, then text. Byline and tags moved below the first paragraph, share buttons to the end, newsletter box to the end. Nothing between the reader and the content.
3. Headings that state a claim, not a topic. "Where people actually left" beats "Analysis". A reader scanning headings should be able to reconstruct the argument from them alone — and if they can't, the piece probably doesn't have one.
4. A visible progress indicator. Marginal on short pieces, meaningful past about 2,000 words. Knowing there are four minutes left is very different from an unknown amount of scrolling ahead.
5. Paragraph discipline. Maximum four sentences, hard rule. No exceptions, including where it hurts the prose. The visual rhythm matters more to completion than the sentence-level craft does, which was an unwelcome thing to learn.
Answer the question up front. The fear is that people leave once they have the answer; what actually happens is that they trust you enough to keep reading.
What didn't work
Shortening. I cut a 3,000-word piece to 1,400 and completion went down. My best guess is that the cuts removed the specifics, and specifics are the reason people stay.
More images. No measurable effect either way. I'd assumed this was a reliable win; it wasn't, at least for my material.
Summary boxes at the top. Slight negative. People took the summary and left — which is the failure mode everyone predicts for change #1, and it turns out the difference is whether the answer is given as prose you're already reading or as a box you can extract.
That last finding is the one I'd want someone to take away. The mechanism isn't "give people less". It's "prove quickly that you're worth the time", and a summary box proves the opposite — that the article is padding around a paragraph.
Reading a drop-off curve
The single completion number tells you there's a problem. The curve tells you where. Here's my own, before and after the five changes, at four scroll markers.
| Marker | Before | After | Change |
|---|---|---|---|
| Reached 25% | 58% | 84% | +26pt |
| Reached 50% | 41% | 76% | +35pt |
| Reached 75% | 29% | 68% | +39pt |
| Reached end | 22% | 61% | +39pt |
The shape of the improvement is the interesting part. Almost all of it happens by the 25% marker, and after that the two curves decline at a similar rate. In other words: I didn't make the middle of my articles more compelling. I stopped losing people at the start, and the ones who got past the opening were always going to finish.
That's a useful diagnostic for anyone measuring this. A cliff in the first quarter is a promise problem — the reader can't tell quickly enough that this is for them. A steady decline throughout is a pacing problem. A cliff in a specific section is a section problem, and you can go and read that section knowing something is wrong with it.
Long isn't the problem; unearned is
The finding I keep having to defend is that shortening made things worse. Cutting a 3,000-word piece to 1,400 dropped completion, and my longest article has my highest completion rate.
My best explanation: the cuts removed specifics, and specifics are the reason people stay. What got cut was the exact numbers, the worked example, the failure case — because those felt like elaboration. They weren't elaboration; they were the evidence that the writer had actually done the thing. Remove them and you're left with a shorter piece that reads like every other piece on the topic, and readers leave generic writing quickly regardless of length.
So the rule isn't short or long. It's whether each section earns its place:
- Does this section contain something only I can say? A number I measured, a failure I had, a decision I made. If not, it's a candidate for cutting regardless of length.
- Could a reader predict this paragraph from the heading? If yes, it's adding words rather than information.
- Am I explaining, or establishing that I'm qualified to explain? The second one is worth exactly one paragraph, near the top, and no more.
Applied honestly, this usually makes a piece longer in the places that matter and much shorter in the throat-clearing. Which is roughly what happened here.
Questions I get about this
Found an error in this piece? Tell me — corrections get made in the open with a dated note, never a silent edit. The log is public.
If you want the long version: there's a course opening later this year — The Research Method, waitlist only for now. If this post already solved your problem, you don't need it.