How to Find Patterns Across Outlets, Journalists, and Topics

Key takeaway: the crossing of the three lenses is the pattern. A single lens is only a cut.

Media coverage analysis, and how it differs from clip counting.

Media coverage analysis is the practice of reading what coverage says over time, rather than counting how much of it ran. Clip counting answers how many articles appeared. Sentiment-only reading answers whether the tone was positive. Neither answers whether the coverage is being written by the same few people, in the same frame, about the themes your strategy actually funded. AMEC's Barcelona Principles 4.0 warn against the substitution error, that media clip volumes and tonality of coverage do not equate to the effect on the stakeholder audience, which is why counting articles is the smaller question.

The difference matters most when volume is stable. A quarter can hold its article count exactly while the authorship concentrates, the theme mix narrows, and the messages you set out to establish stop appearing. Nothing in a clip count or an average sentiment score moves when that happens.

Framing itself, how a topic gets characterized and how that characterization shifts, is covered in depth in narrative tracking. This page is the layer next to it: the crossing of outlet, journalist, and topic, and what you do once you can see it.

Key takeaway: media coverage analysis reads what the clips say over time. Counting how many ran is a different, smaller question.

Step by step: how to find the pattern

1. Define the question. Write down the decision the read has to support before you open anything. "Are we being framed the way we intended since the launch?" is a question. "Let's look at last quarter" is not. Without one, every spike looks like a pattern.

What to look for: a question specific enough that a result could disappoint you.

2. Pull the coverage. Set the date range, the brands, and the outlet set. Keep the window long enough to hold several comparison periods, because a pattern is a repetition and one period cannot show one.

What to look for: a window that includes at least one period before whatever you are measuring the effect of.

3. Tag themes. Apply the same theme set and the same message markers to every article in the window. Consistency beats granularity: a time series only holds if the coding at the start matches the coding at the end, and retroactive re-tagging is the expensive way to learn that. This is the same setup decision narrative tracking treats as the foundation of a running program. In Delve, themes and message pull-through sit on each tracked article alongside the outlet and date, which is what makes step 4 possible without rebuilding the set by hand.

What to look for: any theme whose definition drifted mid-window. That one will fake a trend.

4. Compare across periods. Put the same cut side by side: this quarter against last, pre-launch against post-launch. Compare share by theme, outlet mix, and message pull-through rather than raw article counts, because counts move with the news cycle and shares move with the story.

What to look for: a metric that stayed flat while its composition changed underneath.

5. Find repeat authors and angles. Group by author, then by theme within author. Flag anyone who has covered the same theme more than once, and record the angle each time rather than just the count. Angle is your own note on what the piece argued, not a field any tool returns, so write it in your own words as you read.

What to look for: concentration. If a handful of bylines account for most of a theme, that theme has a small number of gatekeepers.

6. Note what is missing. List the themes, outlets, and authors you expected to see and did not. A message that never appears in the coverage that matters to you is a finding, and it is the one that volume reporting can never surface, because absence has no clip.

What to look for: the message you briefed hardest on, and whether it shows up anywhere.

Key takeaway: define the question first. The three lenses only answer what you asked them.

A worked example, with invented figures: one pattern and the action it triggers

The figures in this section are invented. They are here to show the shape of a read-out and the reasoning it supports, not to report findings about any real brand, outlet, or journalist. The point of the section is the sequence, observation to evidence to interpretation to action. Run the method on your own coverage to get numbers that mean something.

Observation. Across two consecutive quarters, coverage of one pricing theme held almost exactly steady: 58 articles in the first quarter, 61 in the second. On a clip count, nothing happened.

Evidence. The authorship moved sharply, and these are directional readings inside periods that cleared the floor set for this example rather than findings in their own right. In the first quarter the three most frequent authors on that theme wrote 14 of the 58 articles, about a quarter of it. In the second they wrote 37 of 61, about three fifths. Of those 37, 29 carried the same angle. Over the same two quarters, the intended message that pricing reflects usage appeared in 22 of 58 articles in the first quarter and 9 of 61 in the second, falling from roughly 38 percent of the theme to about 15 percent. These counts assume every article in the set carries a byline. On a real pull, check the unbylined share before reading the concentration figure.

Interpretation. Volume is flat, so a volume report shows a stable quarter. Underneath it, one theme has concentrated into three bylines running a consistent angle, and the message the team briefed on has been displaced inside the same coverage. That is a narrowing, not a decline, and the three lenses are what make it visible: the topic lens shows the theme steady, the journalist lens shows the concentration, and the crossing shows which specific coverage the message dropped out of.

Action. The three authors go on the target list with the reason recorded against each one: theme, angle, and the article that put them there. The next briefing leads with the usage data rather than restating the message, because the message is not missing from the market, it is missing from their coverage. Re-run the cut next quarter and check whether pull-through recovers.

Key takeaway: a pattern without an action is a chart. The read-out ends in who or what changes next.

Turn the findings into a target list

The list is an output of the read, not a search. You are not looking for every journalist who covers your category, but at the people already writing about you, your competitors, and your themes, and deciding which of them matters next quarter.

1. Start with repeat authors on your priority themes. The ones from step 5, in order of how much of the theme they account for.
2. Add authors covering those themes for competitors or the category. They have demonstrated interest in the subject without having written about you.
3. Sort by recency, theme fit, and whether they carried your message. Someone who covered your theme last month and did not carry the message is a different conversation from someone who did.
4. Record the reason each person is on the list. The article, the theme, the angle. That reason is the opening line of any outreach, and a list without reasons decays into a spreadsheet nobody trusts.
5. Revisit after each comparison period. The list is a snapshot of who is writing now, and it goes stale at the same rate the coverage does.

Because authors sit on the tracked articles themselves, the list comes out of the same set you analyzed rather than a separate exercise, for the articles that carry a byline. The reason attached to each name is the part that survives a handover.

Key takeaway: the list is a byproduct of the pattern read. It is derived from your own coverage rather than looked up.

Where this breaks

Small samples. A handful of articles per period produces patterns that vanish next month. Set a floor before you start and state it in the read-out: something like 50 articles per comparison period before calling a shift a pattern, with anything thinner labeled directional in the prose. Apply the floor to the period total. Cuts inside a period, one author's share or one message's count, will usually be smaller than the floor by definition, and those are directional readings of a set that cleared it rather than findings in their own right.

One-off spikes. A single news event, a wire story, or syndicated copy can make one theme or outlet look dominant. Twelve outlets running the same frame is one story repeated twelve times if eleven syndicated it. Re-run the view without the spike and see what survives.

Inconsistent tagging. If themes were coded differently across periods, the comparison does not hold, and the failure is invisible because the chart still renders. This is a common way a time series lies, and it is why consistent coding is a setup decision rather than a reporting one, as step 3 above says.

Missing bylines. The author field is filled when a byline exists and empty when it does not, so wire copy, aggregated items, and some captures arrive without one. Those articles stay in the theme and outlet counts while dropping out of the author lens, which makes concentration look higher than it is. Check how many are unbylined before reading concentration as a finding.

Tone read as pattern. Steady sentiment can sit on top of a framing shift. An aggregate score does not move when the composition of coverage changes, which is exactly the case the example read-out above describes. Read themes and messages, not only tone.

Key takeaway: thin samples, syndication, inconsistent tags, and unbylined articles all manufacture patterns. Themes and messages beat tone on its own.