About Transparency News
Transparency News follows the news one event at a time. For each event it shows which outlets covered it, in the order they published, and the wording each one chose, with listed loaded words marked and counted.
Automated assessment. Scores measure selected wording patterns in the outlet's own words. They do not assess factual accuracy, completeness, or which stories an outlet covers.
What does the number mean?
It counts listed loaded words and phrases in the outlet's own voice, per 1,000 words, and places the result on a scale from 0 to 100. A report with no counted words scores 100. Words inside quotation marks (except scare quotes, which count), words that are negated, attributed to a speaker, used literally, or discussed as a term are shown but not counted.
85 to 100: few flagged expressions. 70 to 84: moderate density of flagged expressions. Below 70: high density.
Two live examples from the last seven days, with all the arithmetic shown: the news report with the fewest flagged expressions, Read full episode transcripts of "Face the Nation with Margaret Brennan" from 2026 100 CBS News, and the one with the most, HK Pro-Democracy Activist Joshua Wong Sentenced to 10 More Years on 'Foreign Collusion' Charges 15 Breitbart.
What does it measure, and what not?
The site's standard is that unbiased reporting says what happened, who said what, and how we know, in words that match the weight of the event; names its sources; labels opinion; and includes the facts a reader needs. The score covers part of it.
| Part of the standard | How much the site measures |
|---|---|
| Words that match the weight of the event | Partly: listed words in the outlet's own voice, with context rules. Words not on the lists are missed. |
| Who said what | Partly: attribution markers are counted, and quotes are separated from the outlet's own words. |
| Names its sources | Partly: flagged unnamed and vague sourcing phrases, and whether any attribution was found. A panel beside the score counts named and unnamed sources, documents cited and links to primary records. |
| Labels opinion as opinion | Uses the outlet's own label. Does not judge unlabeled pieces. |
| Says what happened, and how we know | Not measured in the score, which does not check accuracy. Vote counts in Senate stories are checked against the Senate's roll-call record and bill numbers are looked up on Congress.gov, shown beside the score. |
| Includes the facts a reader needs | Not measured. Each event page shows which names and figures each report includes, unscored. |
| Which stories an outlet chooses to cover | Not measured in the score. Shown on the Steering page. |
| Slant built from neutral words (who is described in the passive voice, who gets hedged, whose speaker is quoted first) | Not in the score. Counted per party on event and outlet pages once each side has 10 sentences. |
How do I check a score?
Every report has a page with its arithmetic: each word found, its class, the words around it, its points, the rule behind it, and every step to the total. The full method, every word list and every constant are on the Method page, and the code itself is on the source page. The rule book has a fingerprint (method 3.1.0, 85d55227ee29cf17) that changes if any rule changes.
Is it fair to both sides?
Partisan terms come in mirror pairs with equal points. Every report is scored again with party, politician and institution names swapped, and a fair score should not move. Reports that mention the Transparency Party are listed with their scores. Results, including tests that do not pass and cases the method gets wrong, are on the audit page. Outside researchers are invited to try to break it.
Why is an outlet missing, or marked "short"?
Some outlets ask automated readers not to fetch their pages, or refuse them. The site respects that and scores those reports on headline and summary only. The coverage status page shows which outlets are read in full. Anyone can request an outlet.
How do I dispute a result?
Use the corrections page, or the link on any report or event page. Every submission gets a status and a published outcome, and every change is listed on What changed.
What else does the site track?
The Steering page tracks six ways presentation can shape what readers want without saying so: repeated framing, front-page placement, which stories an outlet covers, gain and loss framing of numbers, the first figure a report gives, and sponsor labels. They never change a score. Everything the site gathers is kept in the archive.
Who runs it
- Ownership. Transparency News is run by Ryan McKeen, who also founded the Transparency Party, a political party. The site scores coverage of the Transparency Party with the same rules as everything else, and the audit page lists every such report.
- Who can change the rules. Ryan McKeen approves rule changes. Every change gets a new method version and fingerprint, is listed on What changed, and publishes how outlet rankings moved.
- Disputes involving the Transparency Party. They go through the same public corrections queue, and their outcomes are published like any other.
- Independent review. No outside body has reviewed the method yet. The rules, code and data are public so that anyone can.
- Funding. A statement of who pays for the site will be published here.
- Tools. The site's code and research notes were written with Claude, an AI model made by Anthropic. No AI model produces any score: every score comes from the published rules.