Measure what the country is saying, properly.
Mention counts move with platform volume, not with public opinion. We measure the rate, the direction, and the specific arguments carrying a subject, on a basis that survives a statistician.
- Refresh
- 15 minutes
- Languages
- 25+
- Baseline
- 7 and 28 day
- Output
- Rate, sentiment, probes
The problem with mention counts
The number most platforms report is the count of posts containing a keyword. That count rises when a platform grows, when a scraper improves, and when an unrelated homonym trends. It falls when an API tightens. None of those movements tell a minister anything about public opinion.
A monitor is built from three inputs that stay independent of each other.
The query controls collection: keywords, combinations, handles, or a handle paired with a term. The subject controls measurement: a relevance gate and the target sentiment is scored against.
The subject does not have to appear in the query, which is how you exclude a homonym or a subsidiary that shares a brand name.
Idea probes are optional and test whether specific associations appear inside the relevant mentions. Each returns present or absent with the passage that triggered it, so "one in three mentions now frames this as foreign interference" arrives with its evidence attached.
Coverage-normalised rates
Every figure we publish is divided by the collection coverage that produced it. If the scraper ran for six hours of a day rather than twenty-four, the day is scaled, and the scaling factor is stored with the observation.
This matters more than it sounds. Keyword collection on most platforms samples rather than enumerates, which means an unadjusted series records your collection schedule as if it were public mood. Analysts who have been burned by this recognise the pattern immediately: the sentiment dip that turns out to be a worker outage.
What arrives on the desk
A daily relevant-mention rate with confidence bounds. A sentiment split scored toward the subject rather than the tone of the post. The share of relevant mentions carrying each idea probe, with representative passages an analyst can quote.
Series plot across a boundary such as a policy announcement or an opposition campaign launch, with before and after tested rather than asserted.
What the system holds
The parts of this capability a technical evaluator will want to interrogate before a procurement decision.
Subject registers
Persistent records for people, institutions, programmes, and brands, with aliases resolved across scripts so Arabic, Cyrillic, and Latin renderings collapse to one entity.
Versioned prompts
Classification instructions are versioned per client and stored beside the results, so a rescoring is reproducible and a definition change is visible in the record.
Instant plus refined
Rule-based scoring returns immediately for the operator, and language model refinement upgrades the same items asynchronously without blocking the view.
Evidence retained
Every score keeps the passage that produced it. A number without its evidence is not usable in a ministerial setting.
Asked in most evaluations
Answers we would give in the room, written down so you can circulate them without a meeting.
How is this different from a social listening tool?
A listening tool counts mentions of a keyword and reports the average tone of the posts containing it. We resolve the subject first, discard mentions that are not about it, score sentiment toward the subject rather than the post, and normalise by collection coverage. The outputs answer different questions.
Can it measure sentiment in dialects and mixed scripts?
Yes. Classification runs on a multilingual model set with per-client prompts, and Arabic dialects, transliterated text, and code-switched posts are handled in the same pipeline. Accuracy is validated against a hand-labelled sample from your own environment during the instrument phase.
How long before a monitor produces a usable baseline?
Seven days for a working baseline and twenty-eight for a stable one. Historical backfill can shorten this where the sources allow it, which we assess per platform during the assessment phase.
Does it take any action on the platforms?
No. The measurement system is read only. It dispatches nothing, posts nothing, and holds no publishing credentials.
Adjacent capability
Each capability runs on the same collection and classification core, so evidence gathered for one is available to the others.
Disinformation detection
We map amplification networks, score account authenticity at the profile level, catch synthetic video and cloned voice, ...
Open-source intelligence
Collection across press, broadcast, social, forums, registries, imagery, maritime and aviation data, fused against a per...
AI representation (AEO/GEO)
Measured control of how AI systems describe you: citation share and consensus depth tracked per engine, facts driven to ...
Bring us the question your last briefing could not answer.
Tell us the jurisdiction and the mandate. We will tell you within a week whether we are the right people for it.