In short
News monitoring with AI follows a defined subject across chosen sources and reports what is new. Keyword alerts miss documents written in other terms and return the same story many times. AI agents can read the sources, group duplicates and flag what changes a subject. Each item needs its source and the exact passage before you use it.
News monitoring runs every day, for weeks or months. This guide sets out a method in six steps for journalists, researchers and analysts. It works with simple alerts, and it shows what AI agents add at each step.
Define the subject and the signals
Write the subject in two sentences: what you follow and why. List the people, organisations, places and documents involved. Then list the signals that would change the subject, such as a new decision, a filing in a register, a court date, a figure that contradicts an earlier one or a person who changes position.
URFIST Rennes, a French public unit that trains university and research staff in information skills, quotes the definition of monitoring by the French standards body AFNOR. It describes a continuous and largely iterative activity of active surveillance, aimed at anticipating changes. The same training warns against too many sources and against the illusion of covering everything. A subject with clear limits produces fewer alerts, and a larger share of them are useful.
Then write the keywords. Include names, abbreviations, former names, spellings in other languages and the terms used in official documents. A local paper writes “the plant”. The inspection report uses the legal name of the site and its registration number. Your keywords need both.
Choose the sources
The URFIST training distinguishes two modes. Targeted monitoring follows known sources, such as an expert’s blog, an institution’s newsletter or specific accounts. Radar monitoring runs broad queries on a theme, to find content outside the sources you know. Most monitoring combines both.
For most subjects, the sources fall into five groups: - the press: national, local and trade titles, radio and television - public registers: company filings, court decisions, public tenders, official gazettes - official decisions: council minutes, regulator decisions, inspection reports - the social media accounts of the people and organisations involved - specialist sites: trade bodies, research centres, NGOs, academic publications
For each source, note how you follow it (RSS feed, newsletter, alert or manual check) and how often. Note also how you found it. GIJN’s guide to digital investigations calls this point essential.
Set up alerts
Alerts bring new items to you, so you do not have to visit each source. According to its help pages on 17 October 2026, Google Alerts sends emails when new results for a topic appear in Google Search. Its options set how often you are notified, the types of sites, the language, the region and the number of results.
Write each alert as a precise query. Put an exact name in quotation marks. Exclude the words that bring irrelevant results. The URFIST training notes that an alert written too broadly produces too many results, which it calls noise.
Use RSS feeds for the sites that publish them, and newsletters for the institutions that send them. For registers and pages with no feed, schedule a manual check. GIJN advises against letting days or weeks pass between checks. Regular checks make patterns, new content and new actors easier to spot.
Sort the results and remove duplicates
The same event often appears in many articles. Dozens of sites republish the same agency story. The local paper, a radio station and the council’s own website all cover the same council decision. Group the items by event and keep one line for each. For each event, keep the original: the decision itself, the first article or the press release.
Then sort the events into three groups: - what changes your subject - what needs an action from you, such as a reply to send or a date to note - background, to read later
Review the plan every week. Keep the keywords and sources that brought useful items, and remove the others. GIJN describes monitoring as an iterative process, in which keywords and sources are added and removed as the work progresses.
Check an item before you use it
An item found by monitoring is a lead. Before you rely on it, check five points: - the date of publication, since old articles resurface - the author or the publisher - the original document behind the article - whether the passage says what the headline or the summary says - whether another independent source confirms it
The URFIST training lists comparable criteria for evaluating information: currency, relevance, authority, accuracy and purpose. When an item concerns a person or an organisation, contact them before you publish.
Keep a record of where each item came from
Pages change and disappear. GIJN’s guide advises documenting what you find as you find it, with screenshots, links, archived copies and notes. It warns against relying on memory.
For each item you keep, record: - the link and the date you saw it - the date of publication - an archived copy or a screenshot - the exact passage that matters - how you found it and why it matters for the subject
With this record, you can show the source of every fact months later. A colleague can also take over the monitoring without starting from scratch.
The limits of keyword alerts and AI summaries
Keyword alerts match words. They miss a document that describes the same fact in other terms, and they return every copy of the same story. Google Alerts reports results that appear in Google Search, so a register or a document that the search engine does not index stays outside it.
AI summaries of the news raise another problem. In October 2025, the European Broadcasting Union and the BBC published a study in which journalists from 22 public service media organisations assessed more than 3,000 answers from four AI assistants. 45% of the answers had at least one significant issue. 31% had serious sourcing problems, such as missing, misleading or incorrect attributions. 20% had major accuracy problems, including invented details and outdated information.
In March 2025, the Tow Center for Digital Journalism at Columbia University tested eight AI search tools on 1,600 queries. Each query gave an excerpt of a news article and asked for its headline, publisher, date and link. Together, the tools gave incorrect answers to more than 60% of the queries. Some cited republished copies instead of the original article, and some gave broken or fabricated links.
An AI summary without its source cannot be checked. Treat it as a lead and find the original passage before you rely on it. Our glossary entry on AI hallucination explains where these errors come from.
News monitoring with Verbatim
Verbatim gives you a team of AI agents that search, read and follow your sources, for hours or for weeks. Its Monitoring agent follows hundreds of sources day and night: the press, radio and television, social media, public registers, court records, public tenders, company filings, official decisions and the sites you add. It runs new searches every day to find sources you do not follow yet. It covers every language relevant to the subject and translates the titles into yours.
For each project, the agent draws up a monitoring plan and starts at once. You change the plan at any time, or write to the agent in plain words. It removes duplicates and groups the articles on the same event into one line. It tells you about anything that changes your subject, with the passage that proves it and the reason it concerns you. Every day, at the time you choose, a short briefing covers all your projects, in Verbatim and by email, and soon on iPhone.
Every finding stays tied to its source. Source verification compares each claim with the exact excerpt it comes from and gives it a status: verified, contradicted, outdated or inconclusive. A quote found only in an AI summary is never counted as verified.
The Monitoring agent works alongside the other agents of your project, which read your documents, prepare your interviews and write in your style. They work in parallel and check each other’s work. Verbatim tests the models of several AI labs on each task and keeps the one that does it best.
Questions
How do you set up news monitoring?
Define the subject and the signals that would change it. Choose sources in five groups: the press, public registers, official decisions, social media and specialist sites. Set alerts and feeds, check the other sources on a schedule, then sort, check and record each item.
Is Google Alerts enough for news monitoring?
Google Alerts sends emails when new results for a query appear in Google Search. It suits a name or an exact phrase. Registers and documents that the search engine does not index fall outside it, so most monitoring plans add feeds and scheduled checks.
Can I trust an AI news summary?
Check it first. In a study published in October 2025 by the European Broadcasting Union and the BBC, 45% of the answers from four AI assistants to news questions had at least one significant issue. Use a summary only when it points to the passage it comes from, and read that passage.
How do I avoid getting too many alerts?
Narrow each query with exact names in quotation marks, excluded words, a language and a region. Group the copies of the same story. Every week, remove the alerts that brought nothing useful.
Sources
- Google Search Help, Create an alert
- GIJN, Investigating Digital Threats: Introduction (monitoring, documentation and archiving)
- URFIST Rennes, Démarche et outils pour développer une veille numérique personnelle, Enssib digital library (in French)
- EBU, Largest study of its kind shows AI assistants misrepresent news content 45% of the time, 22 October 2025
- Columbia Journalism Review, Tow Center, AI Search Has a Citation Problem, 6 March 2025