The media's coverage of the world runs on negativity and conflict. And it creates attention.
But it leaves readers with a sense of powerlessness and distrust, both of those in power and of the media themselves.
Constructive journalism is a tool for seeing the world with both eyes. We help journalists and editors cover solutions, capture the nuances, and facilitate the democratic conversation.
We built Constructive News Mirror to support the journalist or editor who sees the point of all this but struggles to find room for it in a day of deadlines and ingrained practice.
How do we define whether a news article is constructive?
Each pillar is a set of questions about the text. CINARA answers them one at a time and shows what in the article each answer rests on.
Paste an article and see how it reads against these criteria.
Two tools, one feedback loop
Batch report
An editor uploads the newsroom's published articles as a CSV. The report shows scores by section, topic and time period.
Individual article feedback and newsroom-level patterns both use the same three criteria, making it possible to connect individual improvements to newsroom-wide trends.
A specialist model, trained on journalists' judgement
A batch report can cover thousands of published articles, which would be slow and costly to read with a general AI model. So we trained our own. Fellows of the Constructive Institute rated more than 3,000 news articles from 1 to 5 for how constructive they are, and a language model was trained to predict that rating.
Articles rated by Fellows
Journalists on a fellowship at the Constructive Institute scored real news articles from 1 to 5. Their judgement is what the model learns from.
One trained model
A language model built for long texts learned to predict the score the Fellows gave. It does one thing, and it runs in the EU.
A score for every article
Every article in the CSV gets a score in the same run. Averages by section, topic and month become numbers you can follow over time.
The model points to patterns worth looking into. It does not decide what is constructive; your editors do. That is why every report shows the articles behind each number.
Here is an example of how you can work with CINARA
What CINARA's reading rests on
Three ideas shape every analysis CINARA makes.
Grounded in the text. When CINARA says your draft does something, it quotes the passage, so you can check it against the article rather than take its word for it. When something is missing, it says what.
Fits the article. Expectations scale with the article's length, so a short item is not held to what a long feature can do. The suggestions are about this article, not a checklist.
Research-based. The criteria come from research into, and practical use of, constructive journalism.
Three more tools in Constructive News Mirror
From a suggestion to a specific source
Each of CINARA's suggestions comes with a Find sources button. The search looks for the kinds of source the suggestion asks for, such as experts, practitioners or people with lived experience, and shows the people and pages it found, grouped by kind. Each name appears the way the page describes them, with the passage it came from and a record of how the search ran. Who to call is your decision.
How Find sources works01 · Web search
General search in the article's own language, expanded into English for stories with an international angle.
02 · Academic sources
Researchers and studies via OpenAlex, when the search is for experts.
03 · Your own archive
A search of your newsroom's own past reports, so a lead can point back to coverage and sources you already reached.
Watchlist: how your coverage of a topic adds up
CINARA reads one article at a time. Watchlist looks at many. An editor adds a keyword the newsroom returns to regularly, for example a hospital's waiting times or a housing policy, and the system matches new and existing reports against it automatically, at no cost, showing a live count of matching articles. An analysis then groups the articles by angle and names the gap they share.
Open WatchlistEmail your draft, get CINARA's reading back as a reply
Not every draft starts in a browser tab. CINARA also works over email: paste the draft into the email body, or attach it as a Word document, and send it to cinara@constructivenewsmirror.com from the address you use to sign in. A reply arrives in a couple of minutes with the same three-pillar reading CINARA gives on the web, written in the article's own language, along with a link back to the full result in the app.
Three roles for your newsroom
Every workspace has three roles: owner, editor, and journalist. Roles decide who manages the account and who generates batch reports. Everyone, on any role, can use CINARA and Find sources on their own drafts, and can browse the shared watchlist for story ideas. CINARA results are private to the person who ran them, so no one else on the team, not even the owner, can open a colleague's analysis.
Owner
Runs the account. Invites and removes people, sets roles, manages billing, and edits the outlet profile. Can do everything editors and journalists can.
Editor
Generates and deletes batch reports from CSV files, and browses the shared report library alongside the watchlist. Uses CINARA and Find sources on their own drafts.
Journalist
Uses CINARA and Find sources on their own drafts and browses the shared watchlist for ideas. Does not generate batch reports or see the report library, and cannot change account settings.
Paid newsrooms share one weekly allowance for CINARA and Find sources across the whole team, shown as a counter next to each button. The allowance is set per newsroom and resets at the start of each week.
Developed by the Constructive Institute
The Constructive Institute is a Danish non-profit that trains journalists and newsrooms in constructive journalism. It works with media organisations in Scandinavia and internationally.
Constructive News Mirror is a tool developed in connection with that work.
Visit constructiveinstitute.org
Your text is analysed in Europe
Drafts and articles are analysed within the EEA. Your text is not used to train any model.
Microsoft may store text its systems flag and have it reviewed by staff in the EEA.
Find sources sends search queries written from your article to search services in the United States, with people's names taken out first. Every service that handles your data is on our sub-processor page.
Create an account to see how CINARA works. Access is opened newsroom by newsroom while CINARA is closed for maintenance. Write to cle@constructiveinstitute.org and we will set you up.