The tone of mainstream media coverage of Parliament, tracked from a fixed
allowlist of Kenyan outlets — see exactly how it works.
This is a media-coverage tone tracker, not an opinion poll.
It measures how allowlisted Kenyan newsrooms covered Parliament, as read by an
AI classifier with human review — it says nothing about what Kenyans as a whole
think or approve of. No figure is shown from fewer than 5 reviewed items,
and every figure traces back to the listed items below. See the
methodology for the full pipeline and its limits.
What the National Assembly is working on — for each Bill: its current
stage, how media is covering it, and what
citizens have submitted. These are three separate readings,
never blended into one score.
Parliamentary activity as of the last recorded
order paper (sitting): 13 Aug 2026 — this can lag by a
few weeks, and Parliament may be in recess.
Everything below measures the tone of media coverage of Parliament
across all topics — background context for the Bills above. It is not public opinion
and not a poll.
What’s moving this week
Topics ranked by current coverage volume × how far their tone moved against
the trailing four-bucket baseline (weekly buckets where the week has
5+ reviewed items, else monthly). A topic only ranks when both
windows clear 5 items.
Devolution
↑
0.13
n = 34 items this month
· baseline -0.37 (n = 56)
Education
↑
0.22
n = 16 items this month
· baseline -0.30 (n = 33)
Economy & Taxation
↓
-0.67
n = 5 items this week
· baseline 0.11 (n = 10)
This page tracks the tone of media coverage — it is NOT a statistical
opinion poll and never measures public opinion or approval. The pipeline:
articles from a fixed allowlist of Kenyan outlets (Nation, The
Standard, The Star, People Daily, Citizen Digital, NTV, K24, KTN News, KBC and
Business Daily) are ingested every few hours; a classifier first scores each
item’s relevance to the Parliament of Kenya, and items
below the relevance floor are rejected outright.
Surviving items are classified positive / neutral / negative
(with a −1.0 to +1.0 score) by an AI language model; confident
calls publish, borderline ones wait in a human review queue.
Items are then linked to the members and Bills they name, with uncertain links
held for editor review. No average is shown from fewer than
5 reviewed items, every average is printed with its
sample size, and every figure traces to the listed items with source, link and
date. Full details: methodology.
