Industries / Finance and business services / Advertising and marketing / Public relations
Monitoring coverage
Monitoring is finding and collecting what newspapers, broadcasters and online sources say about an organization, and passing it to the people who need it.
6
Milestones achieved, from 1881 to 2025.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1881
Clippings for hire
TIME reported in 1932 that Henry Romeike "established the world's first clipping service in London" in 1881 and opened in New York in 1884. Clipping services cut articles that named a client out of newspapers and posted them on. By 1932 Romeike's firm and Luce's Press Clipping Bureau shared about 80% of the US business, reading 1,900 dailies and 5,000 weeklies and charging 7½ cents a clipping for lots of up to 100.
Credit
Henry Romeike
2001
Clipping moves online
Meltwater, which calls itself "one of the first online media monitoring companies," started in 2001 in Oslo with "two people, a coffee machine, and $15,000." It now lists 27,000 companies as customers, by its own count. By 2013 the Reporters Committee for Freedom of the Press was describing it as "a media-monitoring website that sends paying customers customized reports with excerpts."
Credit
Meltwater
2013
Excerpts need a licence
In a 91-page opinion, US District Judge Denise Cote ruled that Meltwater infringed the copyright of the Associated Press (AP) and could not claim fair use when it "sold excerpts" of AP articles to paying customers. She rejected Meltwater's argument that it acted "as a search engine," and found it had "obtained an unfair commercial advantage" by refusing to pay a licence fee.
Credit
Associated Press; Judge Denise Cote
2013
Reading reports on screen
In the UK, members of the Public Relations Consultants Association (PRCA) used Meltwater to monitor the news, and the Newspaper Licensing Agency (NLA) said they needed a licence even to read reports on Meltwater's website. The High Court and Court of Appeal agreed with the NLA. The UK Supreme Court, in a judgment given by Lord Sumption on 17 April 2013, referred the question of whether copies made on screen and in a browser's cache are covered by the EU exception for temporary copies to the Court of Justice of the European Union.
Credit
PRCA; NLA; UK Supreme Court
2018
Broadcast clips need a licence
TVEyes recorded more than 1,400 TV and radio channels and let clients, for $500 a month, search them and watch, archive and share clips of up to ten minutes. In Fox News v. TVEyes, the US Court of Appeals for the Second Circuit held that its "Watch" function was not fair use, because "TVEyes makes available virtually the entirety of the Fox programming that TVEyes users want to see and hear." Fox had not challenged the lower court's finding that searching the recordings was fair use.
LessonMonitors may index broadcasts for search, but handing out the clips without a licence crossed the line.
Credit
Fox News Network; US Court of Appeals for the Second Circuit
2025
Half skip social listening
In Muck Rack's 2025 survey, 53% of PR professionals said they do not use social listening tools, software that tracks mentions on social platforms. 14% used them daily, 8% weekly and 25% case by case.
Credit
Muck Rack
Watch for
Chatbots misreport company results
A study by the firms Reputation Age and Bubbling ran 8,671 conversations with ChatGPT and Gemini about the half-year accounts of 39 French CAC 40 companies between 25 and 28 August 2026, and reported that "40% of the figures returned by the two assistants are incorrect." Accuracy ranged from 67% for net income to 42% for published growth. The questions, model versions and tolerance for rounding were not published, both firms sell related services, and ActuIA notes "Nothing indicates external control of the protocol."
Achieved whenan outside group repeats the test with published questions, model versions and scoring rules.
Claimed by
Reputation Age and Bubbling, October 2026
Open
01
Test what monitors miss
An independent test of media-monitoring services: a hand-built list of every article and broadcast segment that mentioned a set of organizations in a set of outlets over a month, checked against what each service delivered.
The briefCredit · unclaimed02
Corrections and chatbot answers
A dated, repeated test of whether a published correction changes what AI assistants say about an organization, and how long the change takes.
The briefCredit · unclaimed
Open 01 · the brief
Test what monitors miss
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An independent test of media-monitoring services: a hand-built list of every article and broadcast segment that mentioned a set of organizations in a set of outlets over a month, checked against what each service delivered.
Why it is still open
Eric Schwartzman, who wrote a buyer's guide to monitoring services, said in 2020 that "most media monitoring services have access to the same traditional and social media news" and that, with one exception, "none of the services have a public database that you can search and see what they cover." He described no accuracy testing. Searches in November 2026 found vendors' own advice on missed articles but no independent comparison.
Built with what exists
subscriptions or trials of at least three services, used where their terms allow publishing results; outlets' sitemaps and RSS feeds; GDELT and Media Cloud as open baselines; a speech-to-text model such as ElevenLabs Scribe v3 (see the Note "Fewer word errors") for radio and podcasts; a large language model to confirm each mention, with a person checking.
First step
Pick ten organizations and ten outlets, and build the hand list for one week from each outlet's sitemap.
Done looks like
For ten organizations in 50 outlets and three languages over one month, the share of mentions each service caught and the share of its alerts that were wrong, with a second person checking the hand-built list, all published with the method.
Open 02 · the brief
Corrections and chatbot answers
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A dated, repeated test of whether a published correction changes what AI assistants say about an organization, and how long the change takes.
Why it is still open
Muck Rack's December 2025 study found AI models favour fresh sources, with half of citations published in the previous 11 months, which suggests corrections could flow through. Searches found only agencies' advice that changes take "days to a few weeks," with no controlled test behind it.
Built with what exists
the APIs of ChatGPT, Claude, Gemini and Perplexity with web search turned on; a fixed set of prompts; organizations that are correcting a public error anyway and agree to take part; matched organizations as controls; a scheduler and a dated log.
First step
Find five organizations with a known error in AI answers and start logging the answers before they publish their corrections.
Done looks like
For 20 corrections and 20 controls, answers logged daily for eight weeks on four assistants, with the share of answers still wrong in each week and the median days until the answer changed.