How to automate PR monitoring: a mention is not a result

Media monitoring is easy to set up and produces a feed of mentions that somebody skims for a fortnight and then stops opening. The volume is not the problem in itself; the problem is that a mention count treats a trade publication's feature and a scraped aggregator's copy of a press release as the same event.

Useful monitoring does three things the default does not: it collapses syndication into single stories, it distinguishes the handful of mentions that are worth a person's attention, and it reaches somebody quickly for the small set of things that actually cannot wait.

What we'll cover

• Why mention counts mislead
• Syndication and deduplication
• What makes a mention worth something
• Building it with CodeWords
• The things that cannot wait
• Competitors and the wider conversation
• Reporting it to people who are not in PR
• What to do when the alert fires
• Making it survive
• Limits worth knowing about
• What to build first
• Frequently asked questions

Why mention counts mislead

One story becomes fifty mentions through syndication, aggregation and scraping.

Quality varies by orders of magnitude. A trade publication read by your buyers and a content farm are not comparable.

A passing reference is not coverage, and lists that mention you alongside twenty competitors count as a mention in every tool.

Negative mentions inflate the total, so a bad week can look like a good one on volume.

Reach figures are estimates, often generous ones, and adding them up produces a number with no relationship to anything.

The count rewards the wrong work. Optimising for mentions produces press-release syndication, which is cheap and persuades nobody.

Syndication and deduplication

Cluster by story, not by URL, which is the single change that makes the feed readable.

Match on content similarity rather than headline, since syndicated copies are retitled routinely.

Identify the original, by earliest timestamp and by which outlet others link to, since that is the one worth responding to.

Count the syndication as a property of the story — "picked up by eleven outlets" — rather than as eleven events.

Distinguish a genuine follow-up from a copy, since a second outlet reporting independently is a different thing from a reprint.

Show one line per story in whatever people read, with the detail available underneath.

What makes a mention worth something

Does your audience read it? A small publication your buyers actually read is worth more than a large general one.

Is it about you or does it mention you? A feature and a name-check are different outcomes.

What does it say? Positive, critical, or neutral — and specifically about what.

Did it link? Which matters for search and is a concrete, checkable outcome.

Who wrote it? A journalist who covers your area regularly is a relationship; a one-off aggregator is not.

Did anything follow? Traffic, enquiries, other outlets picking it up — which is the only measure that connects coverage to the business.

Building it with CodeWords

CodeWords connects to more than 3,000 integrations, so the monitoring sources, the analytics, the CRM and your channels can meet in one automation.

1. Open CodeWords and start a new automation.
2. Describe what should happen in plain language to Cody, the automation builder: which terms, how stories should be clustered, and what warrants an immediate alert.
3. Authorize the connections — the monitoring or news sources, your analytics, and Slack or email.
4. Describe the exceptions: a syndicated copy, a competitor comparison, an unrelated same-name match, a known aggregator.
5. Run it over last month and compare its story list against what the team actually noticed.

You describe the outcome; Cody builds it, connects it, and deploys it. The free plan covers light use, with Pro at $39 per month and Business at $100 per month as usage grows; details are on the pricing page.

The things that cannot wait

A short list, and keeping it short is what makes it work.

A significant negative story in an outlet that matters, where the first hours determine how it develops.

A factual error about your company, product or people, which is easiest to correct immediately.

A security or safety claim, which needs the relevant team rather than the communications one.

A regulatory or legal mention, which has its own process and should route into it.

Sudden volume on any term, which is the general early signal that something is happening.

An executive or employee named in something unexpected, which has a personal dimension and should not reach them through a public feed.

Competitors and the wider conversation

Track competitors on the same terms, which turns your own coverage into a comparison rather than an absolute.

Watch the journalists, not only the outlets. Somebody who has covered your area three times recently is the most useful contact information a monitoring system produces.

Notice the framing shifting in how your category is described, which precedes changes in what buyers ask about.

Catch comparison articles early, since being absent from a roundup of your category is actionable and is invisible in a mention feed.

Watch for your customers being covered, which is both a relationship moment and occasionally a story of your own.

Share of voice needs care. It is a reasonable relative measure and a poor absolute one, and it rewards volume over quality exactly as mention counts do.

Reporting it to people who are not in PR

Lead with the stories, not the numbers — three or four, with one line each on why they matter.

Say what changed rather than restating the same categories monthly.

Connect it to something observable: traffic, branded search, enquiries mentioning the article.

Name the specific opportunity — a journalist worth talking to, a comparison you were left out of — since a report with an action in it gets read.

Keep it short. A monthly PR report is read by people with no professional interest in PR, and length is what stops them.

Be honest about the negative, since a report containing only good coverage is recognised as incomplete and discounted entirely.

What to do when the alert fires

Monitoring that detects a problem and delivers it to a channel has done half a job. The half that matters is what happens in the following hour, and that is decided before anything happens rather than during it.

Name who decides, in advance, including out of hours. The most common failure in a real incident is several people waiting for somebody to say something.

Route by type, not to one channel. A security claim needs the security team; a regulatory mention needs whoever handles that; a factual error needs whoever can contact the outlet.

Distinguish responding from reacting. Most negative coverage warrants no public response, and the decision not to respond should be an explicit one rather than an omission.

Keep the facts to hand. The delay in most responses is assembling what is actually true, which can largely be prepared beforehand.

Correct errors of fact quickly and privately with the journalist, which works far more often than people expect and works less well after a day has passed.

Record what happened — what was said, what was decided, what followed — because the next one benefits and nobody will remember otherwise.

Making it survive

Handle source outages by reporting them, since a quiet feed looks like a quiet week.

Deduplicate on the story across runs, so a re-run does not re-alert.

Filter same-name matches, which for any common company name is most of the volume.

Keep an archive of the full text, since articles change and disappear and a monitoring record is frequently the only copy.

Report the outcome: stories found, clusters formed, alerts raised, and sources contributing.

Limits worth knowing about

Coverage of sources is incomplete in every tool, particularly for print, broadcast, podcasts and closed communities.

Paywalled content may be detected without the text being available.

Sentiment classification on news is weak, since much reporting is deliberately neutral in tone while being unfavourable in substance.

Reach and value estimates are modelled, vary between tools, and should not be summed.

Attribution from coverage to outcome is loose, and a spike in branded search after an article is a correlation rather than a measurement.

What to build first

Story clustering. Collapse syndicated copies into single stories with a pick-up count, and show one line per story. It changes nothing else and it turns a feed nobody opens into a list somebody reads, which is the precondition for everything else being worth building.

Two habits make the difference. Keep the urgent list short and specific, since an alert channel that fires for ordinary coverage teaches people to ignore it before the one real crisis arrives. And track journalists rather than only outlets, because the name of somebody who has covered your area three times this quarter is the most actionable output this whole system produces.

Frequently asked questions

Why are mention counts misleading?

One story becomes dozens of mentions through syndication, quality varies by orders of magnitude, passing references count the same as features, and negative mentions inflate the total — so a bad week can look like a good one.

What is the most important processing step?

Clustering syndicated copies into single stories, matched on content similarity rather than headline, with the pick-up count as a property of the story. It is what makes the feed readable.

What makes a mention actually valuable?

Whether your audience reads that outlet, whether the piece is about you or merely names you, what it says, whether it linked, who wrote it, and whether anything followed in traffic or enquiries.

What should trigger an immediate alert?

A significant negative story in an outlet that matters, a factual error, a security or safety claim, a regulatory mention, sudden volume on any term, and an individual being named unexpectedly. Keeping this list short is what makes it work.

Is share of voice a useful metric?

As a relative measure over time, reasonably. As an absolute one, no — it rewards volume over quality in the same way mention counts do.

What is the most actionable output of media monitoring?

The names of journalists who have covered your area recently. That is a relationship you can act on, and most monitoring reports track outlets instead.

Why is sentiment analysis weak on news coverage?

Because much reporting is deliberately neutral in tone while being unfavourable in substance. The tone classifier sees neutral and the reader does not.

Should a PR report include negative coverage?

Yes. A report containing only favourable coverage is recognised as incomplete and then discounted entirely, including the parts that were accurate.

What should happen after a negative story alert?

Route by type rather than to one channel, with a named decision-maker agreed in advance including out of hours. The usual failure is several people waiting for somebody to say something.

Should every negative story get a response?

No — most warrant none. The point is that not responding should be an explicit decision rather than an omission, and factual errors are the exception worth correcting quickly and privately with the journalist.

Why keep an archive of the full text?

Because articles are edited and taken down, and a monitoring record is frequently the only remaining copy of what was originally published.

Related reading

• Automating newsletter curation
• Automating blog post distribution
• Automating sentiment analysis on reviews