Three messages, forty articles, one tired executive
A client's campaign had three agreed key messages: the product is made in the UK, it is cheaper to run than the market leader, and the company is hiring in the North East. The client's communications director wants to know, for every piece of coverage, which of the three appeared, whether the spokesperson was quoted, and whether the piece was positive, neutral or negative.
An executive reads each article with the message sheet open, highlights sentences and fills in columns. By article twenty she is scoring more generously than she was at article three. A colleague who helps with the last ten scores tone differently. The client, reading the final report, questions why a piece that only mentioned the hiring plan in its last line counts as carrying a key message.
Why message tracking is so uneven
- Messages are paraphrased by journalists, so a keyword search misses most of them.
- Two people reading the same piece reach different judgements on partial mentions.
- Tone is judged on feel, with no written rule for what counts as negative.
- Spokesperson quotes are sometimes attributed to the company rather than the person.
- The analysis happens at month end, long after the pieces ran, in a single tiring batch.
The analysis is valuable because it is what separates coverage that helped the client from coverage that simply mentioned them. That is also why the inconsistency matters.
What inconsistent scoring costs
When clients doubt the scoring, they doubt the report, and the conversation shifts from what to do next to whether the numbers are fair. Teams defend old judgements instead of planning. It also hides useful signals: if one message never makes it into coverage, that should change the next pitch, but nobody sees the pattern when scoring is uneven and late.
How we build the analysis step
- Each campaign has its key messages written in plain language, with a few example phrasings the account team agrees.
- When a piece is confirmed in your coverage record, its text is read by a language model such as OpenAI or Anthropic Claude against those messages.
- For each message the step returns one of three answers, fully present, partly present or absent, with the sentence it relied on quoted beside it.
- Spokesperson quotes are found and attributed, using the spokesperson names on the campaign.
- Tone is suggested using a short written rubric your team agrees, again with the evidence sentences shown.
- The executive reviews each suggestion in a side-by-side view and accepts or changes it. Changes are kept, so you can see where the tool and people disagree.
| Output | Shown with | Final say |
|---|---|---|
| Message present, partial or absent | The sentence it relied on | Account team |
| Spokesperson quoted | The quote and attribution | Account team |
| Tone | Evidence sentences and your rubric | Account team |
| Prominence, such as headline or last paragraph | Position in the piece | Account team |
The text used is only what your monitoring and clipping steps already hold under your licences. Paywalled pieces without stored text are marked for manual scoring.
Scoring as you go
Pieces are analysed the day they are confirmed. The executive spends a few minutes each morning confirming suggestions for yesterday's coverage, with the evidence in front of her. At month end, the analysis is already done and consistent, and the account director can see that the hiring message appears in almost nothing, which becomes the focus of next month's regional sell-in.
When the client challenges a score, the report shows the sentence it was based on.
Is this your month end?
- Key message scoring is done in a spreadsheet at the end of the month.
- Different people score tone differently.
- Clients question why a piece counted as carrying a message.
- Nobody notices which messages never land until the quarterly review.
- Scoring evidence is not recorded, only the score.