Time on page is the wrong metric to optimize for when evaluating interactive content. Content teams embed polls and charts to increase dwell time, then justify the decision by citing a 2016 Demand Gen Report or an Infogram A/B study with no published methodology. Those numbers are a decade old. Nobody can reproduce them.
The evidence chain behind "interactive content increases engagement" is circular: pages citing pages citing a single frozen benchmark never verified in public.
Embedding a poll or chart on a blog post can increase time on page, but the metric itself is incomplete. A controlled A/B experiment on this URL measures per-variant engagement rate and session duration with live data. The more durable signal is what the interactive element generates: first-party data that compounds after the session ends.
The Evidence That Interactive Content Increases Time on Page Is a Decade Old
Every content marketer has heard the claim. Most can name the stat. Few can name the study.
The most-cited number comes from a 2016 Demand Gen Report. Ten years of content strategy built on a number nobody can reproduce. The Infogram study that occasionally surfaces alongside it tested 1,000 users. No published methodology. No control description. No raw data.
Content Marketing Institute surveys from 2019 asked marketers whether they believed interactive content worked, then reported the belief as evidence.
These are zombie statistics: data points that propagate across the web without provenance, long after the conditions that produced them have changed.
The claim might be right. The evidence trail is folklore dressed as research. Every ranking page for "does interactive content increase engagement" cites other pages. The chain circles back to the same frozen benchmarks. Nobody has isolated a specific interactive element, on a specific page, with a transparent methodology and live data.
One URL, Two Variants, Zero Borrowed Data
When you load this page, a first-party cookie assigns you to one of two groups. Variant A renders the poll and chart below. Variant B renders the same page with minimal placeholders where the interactive elements would be. Same URL, same content, same traffic sources.
One variable.
Most "interactive content experiments" compare different blog posts and attribute differences to interactive elements. Different topics attract different audiences. Different publish dates pull from different traffic sources. Word count alone changes scroll behavior. Comparing two posts and calling the difference "interactivity" is comparing five variables and naming one.
GA4 collects session duration and engagement rate for each variant independently through a custom dimension. LiquiChart's experiment infrastructure freezes those numbers in weekly snapshots so the comparison accumulates over six weeks rather than resetting with each analytics window. The experiment uses a poll-backed chart (the same method used to embed a live chart in any blog) as one of the interactive elements under test.
What We Measure
Primary metrics: engagement rate and average session duration per variant.
Secondary metrics: poll participation rate, return visits within 30 days, and first-party data points generated per session. These measure what time on page cannot.
What We Do Not Measure
Bounce rate is deprecated in GA4. Scroll depth is confounded by layout differences between variants. Conversion events require a sample size this experiment cannot produce. What we exclude tells you as much as what we include.
Isolating a single variable on a single URL requires variant gating, per-variant analytics collection, weekly snapshots, and a living content system that updates the interpretation automatically. That infrastructure is why nobody has published this experiment before.
Every Vote Feeds the Dataset That Powers the Experiment
How many of your published posts include something a reader can interact with?
Every response adds to the distribution the chart renders live.
The distribution shifts as more content teams weigh in. LiquiChart's content measurement infrastructure captures the state in weekly snapshots and updates the interpretation as the shape changes. Every vote feeds a living data source that persists beyond the session.
As more content teams respond above, the distribution will reveal whether interactive elements are standard publishing practice or still an experiment most teams have not started. That baseline shapes what the A/B data on this page can actually prove.
Time on Page Answered the Wrong Question
Say Variant A produces a higher average session duration. The interactive elements increased time on page. Hypothesis confirmed. The content team declares victory and embeds polls in every post.
The number answered the wrong question.
A page with five interactive elements and a higher session duration confirms that the page performed differently, while the individual contribution of each element stays invisible. The poll might have added 45 seconds while the chart added zero. Or the reverse. The page-level metric collapses five signals into one number and obscures the components that produced it.
This is container bias: the page absorbs credit that belongs to a specific asset. Measuring the container and calling it a verdict on the asset inside is a category error that most engagement reporting commits by default.
Asset-Level Signals the Experiment Surfaces
The experiment collects data that page-level metrics miss. Poll participation rate: did the reader interact with the poll, or scroll past it? The embed view-to-interaction ratio: did they see it, or use it? First-party data points per session capture what the page produced.
These signals persist after the session ends. Session duration expires the moment the visitor leaves. Five hundred poll responses create a distribution that will still be citable next quarter, next year, and by anyone who embeds the chart. That distribution can be updated as new responses accumulate without anyone revisiting the page.
I have watched content teams celebrate a 15-second dwell time increase while ignoring the 500 data points their poll collected in the same period. The dwell time number goes into a slide deck and is forgotten by the next quarter.
The dataset compounds.
The Null-Result Scenario
If the experiment shows no statistically significant difference in session duration between variants, that finding is more useful than a positive result borrowed from a 2016 report, because it was produced on your content with your audience.
A null result means the interactive elements did not move the page-level metric. The poll still collected responses. The chart still rendered live data. Both generated first-party signals a static page could not. A null result forces a more precise question: if time on page is not the benefit, what is?
The Asset That Generated the Data Deserves the Measurement
Five hundred poll votes. That is not an engagement metric. That is a first-party dataset no competitor has.
The gap between first-party evidence and recycled claims is measurable. LiquiChart's Originality Score measures the percentage of claims backed by original data: your polls, your charts, your experiments. A post that embeds a 500-vote poll scores differently from a post that cites a decade-old Demand Gen Report number. The difference is provenance.
Teams chase content freshness by updating timestamps. A new publish date on a page with the same borrowed statistics changes nothing about the data the page contains. The date changed. The claims inside stayed frozen.
A poll that generates a dataset where none existed before is information gain: data the searcher cannot find on any other page in the SERP.
Running Your Own Experiments Instead of Citing Someone Else's
Every content team publishing "studies show that interactive content increases engagement" is borrowing a conclusion from an experiment they did not run, cannot verify, and cannot reproduce. The alternative is to run the experiment. On your own content, with your own audience, using a methodology you can describe in public.
The evidence on this page is specific to this page. Your audience, your content, your traffic patterns will produce different numbers. A benchmark broad enough to apply to everyone controls for nothing.
The Numbers on This Page Will Be Different Next Month
The results will change as data accumulates. The living content block will update its interpretation when statistical thresholds are crossed. Return in three weeks and the numbers will be different.
You can create a poll on the free tier, visualize the results as a live chart, and watch the data accumulate without touching the page again. The A/B experiment infrastructure is available on the Visionary plan.
The question worth asking goes one level deeper: can your measurement infrastructure tell you which element on the page deserved the credit?