Skip to content

Counting content assisted registrations without pretending the content caused them

Content studioUpdated 2026-08-237 min read

In short

Content assisted registrations counts registrants who viewed at least one editorial page inside a stated window before registering, reported as coverage of the registration file. On 4,200 registrations with 1,100 carrying a prior editorial view in 30 days, the assist rate is 26 per cent, which describes overlap and claims no causation.

The quarterly review has a slide claiming that content drove 1,100 registrations. Somebody asks how that was worked out. The answer is that 1,100 registrants had visited an article at some point, which is a real fact producing a false sentence, because the verb on the slide is doing work the query cannot support.

Content assisted registrations is a measurement worth having, and it survives scrutiny only if you report it as coverage. The number describes how much of your registration file had contact with your editorial. It says nothing about what would have happened without it.

What an assist can honestly claim

Define it as an overlap and the definition writes itself. An assisted registration is one where the registrant viewed at least one editorial page within a stated window before the registration was created.

That sentence contains everything the number claims. There was a person, there was a page view, there was a registration, and the view came first. No mechanism, no counterfactual, no share of credit.

Reporting it that way costs you the exciting version of the slide and buys you a number nobody can knock over. When a finance lead asks whether the content paid for itself, the correct answer is that this measurement cannot tell you, and that the measurement which can is a different exercise with a control group.

Four decisions to make before the query runs

Every one of these moves the answer, and none of them has a right answer, so the discipline is to choose and publish.

  • The window. How many days before registration a view counts. Google's Analytics 4 documentation, current in 2026, sets a default lookback window of 30 days for acquisition key events, switchable to 7, and 90 days for all other key events with 30 and 60 available. Borrowing 30 days is defensible and so is 90. Quoting a figure with no window is not.
  • What counts as editorial. An article, yes. A speaker profile, probably. The agenda page, no, because that is registration flow wearing editorial clothes and including it will lift your assisted share by ten points overnight.
  • The join. Which identifier links a reader to a registrant. This is the hard part and it is a matching problem rather than a content one, so it belongs with the identity work in cluster J. What matters here is that you record which identifier you used.
  • The direction of time. Before the registration timestamp, which is easy, or before the first touch in the registration journey, which is more honest and much harder to establish.

Write all four into a definition block that sits next to the figure wherever it appears. When somebody recomputes it next year and gets a different answer, the block tells you which decision changed.

Working the assist rate on one edition

Take one edition with 4,200 registrations and a 30 day window.

Join the registration file to identified editorial sessions and 1,100 registrants have at least one qualifying view. The assist rate is 1,100 divided by 4,200, which is 26.2 per cent. Round it to 26 per cent and put the window next to it.

That is the whole headline number, and it is a perfectly good one. Roughly a quarter of the people who registered had read something you published in the month before they did. For a show that has never measured this, the figure is usually either much lower than editorial expects or much higher than sales expects, and either way the conversation it starts is worth the afternoon.

Report the count alongside the rate. A 26 per cent assist rate on 4,200 registrations is 1,100 people. A 26 per cent rate on 900 registrations is 234, and the second one should not be driving a headcount decision.

Check the direction of every view before you trust the join, because a naive query catches reads that happened after the registration. Those are usually the larger group, since a registered attendee gets your emails and clicks them, and counting them as assists inflates the rate while measuring something else entirely. Post registration reading is worth its own number as a signal of how engaged your registrants are ahead of the show, and mixing the two produces a figure that rises every time your pre show email programme improves.

Why the same data cannot say the content caused anything

Here is the arithmetic that stops the word drove from reaching the slide.

Suppose your identified audience database holds 180,000 people, and 4,200 of them registered, an overall rate of 2.3 per cent. Split them by whether they viewed editorial in the window. The 38,000 who did produced 1,100 registrations, a rate of 2.9 per cent. The 142,000 who did not produced 3,100 registrations, a rate of 2.2 per cent.

The gap is 0.7 percentage points, and in relative terms the readers registered about a third more often. It looks like an effect and it is mostly a selection.

People read your editorial because they already know the show exists, already work in the market and are already thinking about attending. The reading and the registering share a cause, which is prior interest, and prior interest is exactly what your database cannot observe. A gap of this size is what selection alone produces, and the honest reading is that it neither supports nor refutes the claim that the articles did anything.

Separating the two needs somebody to be withheld from the content, which is a design question rather than an analysis question and the only method that answers it.

Why does the assist rate understate the reading?

Because a growing share of the reading now happens where you cannot see it.

Pew Research Center analysed 68,879 Google searches from 900 US adults in March 2025 and found that 18 per cent of them produced an AI summary. Where a summary appeared, users clicked a traditional search result on 8 per cent of visits, against 15 per cent where no summary appeared. Clicks on the sources cited inside the summary ran at 1 per cent of visits to pages carrying one. Pew also found that people ended their browsing session entirely on 26 per cent of pages with a summary, against 16 per cent without.

Read that against your assist measurement. A buyer who searches your market, reads a summary that draws on your explainer, learns what she needed and never lands on your site has read your work and appears nowhere in the join. Your assist rate counts sessions, and an increasing share of the reading produces no session at all.

That gap has its own measurement and its own metric, because being named in an answer nobody clicks is still reach and needs counting separately. For the assist rate the consequence is one word in the reporting: floor.

How should the number be reported?

One line, four parts, the same every time.

The rate, the count, the window, and the identification rate. Something like: 26 per cent of registrations, 1,100 of 4,200, with a 30 day window, on editorial sessions of which 45 per cent could be identified.

That last figure is the one people leave off and it is the one that determines whether the first is a floor or an estimate. If fewer than half your editorial sessions can be tied to a person, your 26 per cent is a lower bound on the coverage and everybody reading the slide should know it.

Track the identification rate as a series in its own right. A jump in the assist rate that coincides with a new consent banner or a newsletter login wall is a measurement change, and it will be presented as a content result by somebody who did not notice.

Where this stops

The measurement is bounded above by your identification rate and it always will be. Anonymous readers are real readers, and nothing in this method reaches them.

It is also silent on timing, which is where most of the misreading happens. A view three days before registration and a view twenty nine days before both count as one assist inside a 30 day window, and they are almost certainly different behaviours. Whether your editorial works months ahead of the decision or in the last week is a separate distribution worth building and it changes how you would judge a piece.

The window itself carries an assumption nobody states. Choosing 30 days asserts that reading something 31 days before registering had no part in the decision, which is obviously false and unavoidable. Widening the window raises the assisted share mechanically, so a rate that improved from 22 to 31 per cent should always be checked against whether anybody changed the window.

Last, coverage flatters channels that reach people who were going to register anyway. Your newsletter goes to past attendees, past attendees register at a high rate, and an assist rate computed over newsletter driven reads will look excellent while telling you nothing about acquisition. Split the assist rate by whether the registrant attended a previous edition and the two halves usually look very different.

Run the join once on your most recent edition with a 30 day window, and publish four numbers: the rate, the count, the window and the share of editorial sessions you could identify. That single line does more for the credibility of the content operation than any slide with the word drove on it, and it takes one query and one honest sentence about what the query cannot see.

Questions people ask about content assisted registrations

How do you measure whether content drives event registrations?
Coverage and causation need different work. Coverage is a join: count registrants with a prior editorial view inside a stated window and divide by all registrations. Causation needs a withheld group who never saw the content, because everyone who reads your editorial was already more interested than the people who did not.
What lookback window should you use for a content assist?
State one and hold it. Thirty days is a defensible default and matches the acquisition window Google Analytics 4 uses by default. Longer windows raise the assisted share mechanically, so a figure quoted without its window is uninterpretable and two figures on different windows should never appear on the same chart.
Why is an assisted registration not an attributed registration?
An assist records that a person read something before registering. Attribution claims the reading changed the outcome. People who read a show's editorial are self-selected: they already knew about the show and were already interested, so the overlap between readers and registrants would exist even if the articles had no effect at all.

Related reading

All content and media articles