Alert fatigue kills monitoring programmes. When a comms team receives 80 alerts per day and only 3 are genuinely actionable, they stop reading them. Within weeks, the team reverts to manual scanning, and the monitoring platform becomes expensive shelfware. This guide covers how to tune alerts in Meltwater, Cision, Signal AI, and other platforms so that what fires is actually worth reading.
Define Urgent in Business Terms
Before touching alert settings, agree with your team on what counts as urgent. Urgent means "requires a response within hours" and nothing else. If a story can wait until the daily brief, it is not urgent.
Typical urgent triggers for UK comms teams:
- FCA, CMA, ICO, Ofcom, ASA, or HSE names your organisation in a public statement or enforcement action
- Coverage in a Tier 1 outlet (BBC, Sky News, FT, Guardian, Times, Telegraph) with negative framing about your CEO, brand, or a safety issue
- Any mention of a data breach, product recall, fatality, or serious injury connected to your operations
- Social media volume spike exceeding 3x baseline within a two-hour window with negative sentiment
Write these triggers down as rules. If the trigger does not appear on the list, it does not get a real-time alert.
Tier Your Alerts
One alert feed for everyone is the main cause of alert fatigue. Different audiences need different cadences:
| Alert type | Recipients | Cadence | Examples | |---|---|---|---| | Crisis / real-time | Duty comms officer, comms director | Immediate (push notification + email) | Regulator action, Tier 1 negative, safety incident | | Priority daily | Comms team, CEO office | Daily digest by 08:00 | All Tier 1 coverage, competitor Tier 1 coverage, executive mentions | | Standard weekly | Broader stakeholders, analyst | Weekly summary | Tier 2 and Tier 3 coverage, sector trends, trade press | | Data feed | Reporting analyst | API or scheduled export | Full dataset for dashboard and measurement |
In Meltwater, set these up as separate saved searches with different alert schedules. In Signal AI, use the alert routing rules to direct different triggers to different recipients. In Cision, configure alert groups by topic and recipient.
Build an Exclusion List
Exclusions are where most of the noise reduction happens. A well-maintained exclusion list can cut false positives by 40-70%.
Categories of exclusions:
Brand Name Collisions
If your brand name is a common English word, this is your biggest challenge. Examples:
- "Shell" (the company vs the object)
- "Next" (the retailer vs the word)
- "Sage" (the software company vs the herb)
- "Compass Group" (the catering company vs compass as a navigation tool)
Use Boolean operators to require additional context: `"Shell" AND ("oil" OR "energy" OR "refinery" OR "FTSE" OR "Shell plc")` rather than just `"Shell"`.
Unrelated Entities
Companies with similar names in different sectors or geographies. If you are monitoring "Aviva" the UK insurer, exclude "Aviva" the Italian company or "Aviva" as a first name in unrelated contexts.
Syndicated Wire Duplicates
A single PA Media wire story can appear on 30+ websites. If your alert fires for every duplicate, one story creates 30 alerts. Most platforms offer deduplication settings -- turn them on. In Meltwater, use the "similar articles" grouping. In Cision, enable the deduplication filter in your alert settings.
Old Product Names and Legacy Terms
After a rebrand or product discontinuation, remove old names from your query. If they are left in, you will receive alerts for historical content that resurfaces or for unrelated entities that have taken the old name.
Measure Alert Quality Monthly
Keep a simple log for one month:
| Week | Alerts received | Actionable | False positives | Missed stories | |---|---|---|---|---| | Week 1 | | | | | | Week 2 | | | | | | Week 3 | | | | | | Week 4 | | | | |
Target: At least 60% of real-time alerts should be actionable. If you are below 40%, the query needs significant tuning.
False positive rate: Count how many alerts were irrelevant. Each false positive erodes trust. Investigate the cause (brand name collision, syndication duplicates, over-broad keywords) and add an exclusion.
False negative rate: Log stories you saw through other channels (direct journalist calls, social media, colleague tip-offs) that the monitoring system missed. Each false negative means a query gap. Common causes: paywalled content not indexed, new outlet not in the source list, or a story that used language your query did not capture.
Common Mistake: The Never-Tuned Query
A UK energy company set up monitoring queries when they onboarded Meltwater in 2022 and never adjusted them. Two years later, the team was receiving 95 alerts per day, with a false positive rate of 73%. A product line had been discontinued (but the old name was still in the query), a competitor had rebranded (but the old name was still excluded), and three new relevant trade publications had launched (but were not in the source list). A two-hour tuning session reduced daily alerts to 22, with a false positive rate of 14%.
Schedule a mandatory alert tuning review every month for the first quarter, then quarterly once the system stabilises.
Platform-Specific Tips
Meltwater
- Use the "Explore" feature to test queries before setting live alerts -- check the results for relevance before activating
- Set up deduplication in alert settings to collapse wire stories
- Use the "prominence" filter if you only want alerts where your brand is a major focus of the article, not a passing mention
Cision
- Use the "relevancy" slider to filter out low-relevance hits
- Set up "smart tags" to auto-classify incoming coverage by topic, which improves routing
- Use negative keywords in Boolean queries aggressively
Signal AI
- Signal AI's AI classification is stronger than keyword matching for reducing false positives -- use topic-level alerts rather than keyword-only queries where possible
- Set up entity-level monitoring for organisations and people, which handles name variations automatically
- Use the risk alert feature for regulatory and compliance monitoring
Brandwatch
- Primarily for social listening alerts -- set volume spike thresholds relative to your baseline (3x is a standard starting point)
- Use the "signals" anomaly detection rather than static threshold alerts where available
- Exclude bot accounts and spam sources from your panels
When to Ask Your Vendor for Help
If after two rounds of tuning your false positive rate is still above 40%, schedule a call with your account manager. Vendors like Meltwater and Signal AI have professional services teams that can review your queries and suggest improvements at no additional cost. Take advantage of this -- it is part of what you are paying for.