Many vendors have promised news publishers fully autonomous systems, so what does such a system really decide? A dynamic paywall decides, reader by reader, when a wall goes up and what appears on it. It reads signals about the reader and the visit, then sorts or scores that reader with rules or a model. The reply is a paywall, a registration wall or no wall.

Behind that promise, vendors often meant different levels of automation and of flexibility in setting rules. The mechanics and results below come from accounts by single publishers and firms, and some are several years old. We give the year beside a fact wherever the source states one.

What is a dynamic paywall?

A dynamic paywall is a system that adjusts both when to show a paywall and what to show on it. The first dimension is the moment, meaning the point in a visit at which the wall appears. The second is the screen, meaning what the reader finds on the wall once it is there.

Such a system has three parts. Signals describe the reader and the visit. The decision logic turns them into a choice for that one reader. The outcome is what the reader then meets.

How does a dynamic paywall differ from a classic paywall?

A classic paywall is triggered by one fixed rule applied to all readers. A dynamic paywall is triggered at different times or with different subscription offers, based on the reader's behaviour and characteristics. One example of a fixed rule is a paywall that appears after three articles within 30 days.

Hybrid paywalls are a separate case, because they run two fixed models side by side. The Globe and Mail launched one in 2012, with a simple meter for most articles and about 10% of articles behind a hard paywall. Editors decided which articles went behind it, so the reader's behaviour played no part in that choice.

Setting the limit of a metered paywall and comparing all the paywall models are separate subjects that this article does not cover.

Analytics charts on a laptop screen
Photo: Luke Chesser / Unsplash

Which signals does a dynamic paywall read?

The signals a dynamic paywall reads depend on the data a publisher has, and a typical minimum set includes two groups:

  • Real-time behavioural data, for example time on site, content consumed, number of articles read, traffic sources and device used
  • Engagement data, for example registration, newsletter sign-up and usage of product features

At The Wall Street Journal (the Journal), each non-subscribed visitor to the Web site received a propensity score. That score is a number for the chance that the visitor will subscribe. It rested on more than 60 signals, such as a first-time visit, operating system, device, what the reader clicked on and location.

Paywalls of this kind used dozens of signals to measure each visitor's likelihood of subscribing, and Piano's system started with 76 metrics. One count is in signals and the other in metrics, so the two do not add up.

How does a dynamic paywall decide who sees an offer?

Each reader is sorted or scored, and the dynamic paywall then acts on the result. At The Globe and Mail, a model studied readers' sequential online behaviour. It then predicted when, and on which article, each reader should see a paywall, a registration wall or no wall at all.

The logic behind that decision sits on a spectrum. At one end are rules-based systems, which rely on pre-set rules and reader segmentation. At the other are what Jonathan Harris from Zuora calls "self-learning" paywalls, which use multi-agent reinforcement learning.

Publishers often build or refine a machine learning model for the job, typically a classification model. It predicts how likely a user is to subscribe from behavioural and contextual data. The model then triggers a tailored response, such as locking content or adjusting a call to action.

What can a rules-based dynamic paywall change?

On a rules-based dynamic paywall, the rules can change what the reader sees on the pay prompt. NZZ used 100 to 150 rules that altered the messaging, text, placement and colour of that prompt. The rules took account of data like reading history, device and time of day.

One rule covered the morning commute. Between 5am and 9am, people were commuting, reading on their smartphones and unlikely to type in payment details. NZZ therefore showed no payment messages in those hours, and readers could access the site freely.

How does a propensity score become a paywall decision?

A propensity score becomes a paywall decision once it is set against a threshold. How the model behind the score is built and trained is a topic of its own.

At the Journal, non-subscribed visitors fell into three groups, roughly defined as hot, warm or cold. Visitors who scored above a certain threshold hit a hard paywall. Lower scorers might browse stories free for one session and then hit the paywall. The source gives no value for the threshold.

From 2018 NZZ scored registered readers on some 30 attributes, like time spent on articles, frequency and number of newsletters received. Those who reached the top 20% of the propensity score were served the payment prompt that suited their needs.

How does a model choose between paywall variants?

At Svenska Dagbladet (SvD), a model serves the paywall variant most likely to convert an individual reader, based on real-time behavioural data. The publisher says the model "could be described as" a multi-armed bandit. That is a decision system which selects the best-performing paywall in different user contexts. A context is a mix of features such as referrer, login status and device type.

In its first iteration the model simply optimised for conversions, and SvD sees potential in steering it towards overall revenue. It also planned to integrate its meter with the dynamic paywall logic. That would allow it to optimise revenue along the entire user journey. The publisher's account of how Svenska Dagbladet uses a dynamic paywall model sets out the model and the plan in full.

Why does a dynamic paywall depend on registration?

Registration matters to a dynamic paywall because it was the strongest signal in the models. Being a registered reader was the most predictive feature in all five propensity-to-buy models that Deep.BI deployed at mid-sized news publishers. Yet at the median news brand in INMA Benchmarks, only 8% of monthly users were registered in the third quarter of 2024.

Segmenting users by propensity to buy was proven to lead to higher conversion rates than crude methods. The low share of logged-in readers and cookie depreciation still limited the overall impact on sales.

Since 2019 the first article a reader opens on the Times' Web site has been free. The second sits behind a registration wall. That lets its in-house dynamic paywall assign further free content individually. The registration wall itself deserves a separate article.

A woman reads on a tablet at a table
Photo: Brooke Cagle / Unsplash

What have publishers measured after switching to a dynamic paywall?

Publishers that switched to a dynamic paywall have reported two classes of evidence, an A/B test result and results without a named baseline:

  • The Globe and Mail counted 51% more subscriptions than with its old paywall, in an A/B test run at the same time
  • The Financial Times reported increases of 92% in conversion rate, 118% in progression through the subscription funnel and 78% in subscriber lifetime value about a year after launch, with no baseline or method given
  • NZZ said it had increased its conversion rate fivefold in three years, with 2.5% of people who viewed a payment message prompted to subscribe, a statement backed by no control group

These three results measure different things at different publishers, so they cannot be compared.

Does a dynamic paywall cost you advertising revenue?

Yes, a dynamic paywall can cost you advertising revenue, because a broader paywall application often reduces ad views. How many views are lost, the source does not say.

Publishers that targeted more than 10% of pageviews with paid offers had 39% more active subscriptions than those below 10%. That figure compares two groups of publishers and comes from no test. The comparison stands in an earlier post on audience segmentation and nuanced revenue strategy. That post also sets out how to balance the two revenue streams.

The Globe and Mail reported that subscription revenue gained was 10 times the advertising revenue given up.

What should you know before you start with a dynamic paywall?

Start a dynamic paywall with what your data allows, and expect limits. One adviser calls rules-based approaches "often a practical starting point", particularly when data is limited or internal capabilities are nascent.

The Globe and Mail first limited its A/B tests to audiences in a few small cities. After encouraging results it moved to small segments of its entire audience and gradually increased them.

Heavy readers kick-started subscription programmes but could not sustain the growth. As brands matured, they increasingly converted light and new readers. Such readers are harder to segment by engagement. Some INMA members reported their disappointment over the years, as the post on how dynamic paywalls gain momentum describes.

Questions publishers ask about dynamic paywalls

Is a smart paywall the same as a dynamic paywall?

Yes, the sources use both labels for one kind of paywall. One names "dynamic" and "smart" side by side, and another lists "propensity", "dynamic" and "intelligent" together.

Is a dynamic paywall the same as dynamic pricing?

No. Dynamic pricing changes the price a customer sees, while a dynamic paywall changes when, where and how the subscription ask appears. Publishers sometimes conflate the two.

How many news publishers use a dynamic paywall?

The figure is 22% for hybrid and dynamic paywalls together, in one online survey of event attendees. A separate 2020 study of news sites in 33 countries put the two types at only 5%. Neither source counts dynamic paywalls alone.

Can your newsroom overrule a dynamic paywall?

Yes, at The Globe and Mail it could. The newsroom retained the power to manually override any algorithmic paywall decision. It kept that power for content it believed must be free as a matter of public interest.

Does a dynamic paywall work on readers you cannot identify?

Yes, by one publisher's account. At Times Internet in India, predictions about converting content supplemented a more traditional behavioural model. That allowed targeting 70% of monthly users who were new to the site or anonymous. The account gives no sales result.

What should you settle before a dynamic paywall decides for you?

Settle four questions in your own team before a dynamic paywall decides for you:

  • What share of your monthly users is registered today?
  • Should the paywall optimise conversions or overall revenue?
  • Who in your newsroom can override the algorithm, and in which cases?
  • Will an A/B test against your current paywall measure the result, or a before-and-after comparison?

Signals, thresholds and outcomes have to be tuned on your own readers. The public accounts come from single deployments. Work through these questions with other publishers at the Subscriber Acquisition Master Class, November 2026.

Sources