Media Monitoring: A Working Guide for Communications Teams

Media monitoring is the continuous tracking of press, broadcast, social and other public sources for mentions of a brand, its people and its competitors. In practice it is a stack of separate jobs, collection, analysis, alerting and reporting, that vendors bundle together.

Every team that adopts media monitoring goes through the same conversation eventually. Someone asks what the tool is actually supposed to do, and the honest answer turns out to be less obvious than the sales page suggested.

Monitoring is not a single thing. It is a stack of separate jobs that vendors bundle together, priced together, and demoed together, which makes it easy to buy the wrong shape of tool for what you actually need. This is the guide for figuring out which shape you need before you sign anything.

What media monitoring actually is

At its simplest, media monitoring is the continuous tracking of what gets published about a subject, wherever that publishing happens: newspapers and their websites, broadcast transcripts, blogs, forums, social platforms, YouTube, and now what AI assistants say when someone asks them a question about you.

The word “continuous” is doing real work in that sentence. A search you run once a week is research. A system that watches all day, every day, and tells you the moment something changes, is monitoring. The difference matters because most of the value in this category comes from speed: knowing about a story in the first hour is worth more than knowing about it accurately three days later.

The second thing monitoring has to do, beyond collection, is turn raw mentions into something a human can act on. A thousand unsorted articles a day is not intelligence, it is noise with a subscription fee. What separates a genuinely useful system from an expensive RSS reader is what happens after collection: sentiment, subject attribution, deduplication, and a ranking of what actually needs eyes on it today.

What separates a feed from intelligence

This is the distinction that costs teams the most money when they get it wrong, because feeds and intelligence look identical in a demo and behave completely differently in week three.

A feed shows you everything. It collects mentions, tags them loosely by keyword, and presents them in a list sorted by time. This is useful for a research pass. It is nearly useless as a daily operating tool, because a real brand generates hundreds or thousands of mentions a day once social and community sources are included, and a human cannot triage a list that long every morning.

Intelligence tells you what changed and why it matters. It separates mentions about your own brand from mentions about a competitor that happen to share a keyword. It tracks sentiment against your own historical baseline rather than a generic scale, because a brand that runs at 40% positive normally is not having a bad week at 35%, but a brand that normally runs at 70% positive is. It flags volume spikes only when they exceed what is statistically unusual for that specific subject, not a flat threshold that fires constantly for a large brand and never for a small one.

The practical test: ask a vendor what happens to a mention the moment it is collected. If the answer is “it appears in your feed,” you are buying a feed. If the answer involves sentiment scoring against a subject-specific baseline, deduplication across syndicated copies, and a queue sorted by something other than recency, you are looking at intelligence.

The channels that actually need covering

A monitoring tool’s channel list is the first thing to check, and it is where the gap between marketing copy and reality shows up fastest.

Press. The core of any monitoring system: national and trade publications, wire services, and increasingly the online-only outlets that break stories before print picks them up. The number that matters here is not “how many sources” but how deep the crawl goes into regional and vernacular press, because for most brands outside a handful of English-language markets, the first version of a story runs somewhere the big vendors do not look.

Social. Real listening, not just your own account’s analytics. The distinction is whether the tool can find a public post that mentions your brand from an account that has no relationship to you, versus only showing you comments on posts you made yourself. Most vendors that claim social listening are doing the second thing and calling it the first.

YouTube and video. An increasingly large share of both product reviews and complaint content lives here, and it is the channel most monitoring tools cover worst, because transcript analysis and comment-sentiment scoring are genuinely harder than text search.

Community and forums. Threads where people ask each other what to buy or complain about what they bought. Weaker as a volume signal than press or social, because a lot of what surfaces is discovery rather than confirmed mention volume, but valuable as an early-warning channel precisely because it is where dissatisfaction shows up before it becomes a press story.

AI answers. The newest channel, and the one growing fastest in importance. What ChatGPT, Gemini, Claude and Perplexity say when asked about a brand is now a real part of how people form an opinion before they ever visit the brand’s own site. A monitoring stack that does not track this is missing a channel that did not exist five years ago and now matters as much as broadcast did a decade ago.

Why coverage has holes, and where they actually come from

The most misleading number in this category is the source count, and the reason is not that vendors inflate it, though some do. It is that the hard parts of monitoring have very little to do with how many places a crawler visits. They are a function of what the platforms permit, and over the past three years that has narrowed sharply.

Three changes are worth knowing by name, because between them they explain most of the gap between what a monitoring tool promises and what it returns.

Reddit began charging for API access in 2023. The pricing, reported at the time as $0.24 per 1,000 API calls, was enough to close the major third-party Reddit clients within weeks. Apollo shut down on 30 June 2023 after its developer calculated an annual bill of roughly $20 million. Whatever one thinks of the decision, the effect on monitoring is concrete: comprehensive Reddit coverage moved from free to metered, and any tool relying on a search engine’s index of Reddit threads rather than the API is showing you a partial, delayed, search-shaped view of the platform.

Meta shut down CrowdTangle on 14 August 2024. CrowdTangle was the tool that made public Facebook and Instagram content genuinely observable, and it was the backbone of a great deal of newsroom and academic work. Its replacement, the Meta Content Library, is open to academic and non-profit researchers rather than to news organisations, and a joint investigation by Proof News, the Tow Center for Digital Journalism and the Algorithmic Transparency Institute found the replacement less transparent and less capable than what it replaced, including an inability to export data. The honest statement for a brand context is that broad third-party discovery of public Facebook posts is not something any commercial vendor can offer at the level available before 2024, because the access route itself no longer exists. A vendor still listing Facebook as a monitored channel is usually describing your own page’s comments, which is a different product.

X ended free API access in 2023 and has priced it commercially ever since. Published tiers have started in the low hundreds of dollars a month for volumes that are small by monitoring standards, rising to enterprise agreements negotiated privately. The practical consequence is that X coverage sits on a vendor’s cost line rather than coming from a free crawl, which is why volume caps, refresh intervals and history windows on X data are frequently the thing a vendor is most vague about. Ask for the specific numbers.

None of this is a reason to distrust monitoring as a category. It is the reason to ask a better question than how many sources. The question that actually separates vendors is this: for each channel I care about, is this licensed access, or is it a search engine’s index of that channel?

Those two look identical in a dashboard and behave nothing alike. Licensed access returns real posts with real timestamps, which means a count of them is a real count and a spike in them is a real spike. A search index returns evidence that a conversation exists somewhere, which is genuinely useful for discovery, but a jump in it can just as easily mean a crawler revisited some old pages as that anyone said anything new. Building a crisis alert on the second kind is how teams end up being woken at 3am by Google’s indexing schedule.

Any vendor should be able to answer that question channel by channel without hedging. The ones who cannot are usually not being dishonest. They have simply never been asked to be precise about it.

What nothing can see

The corollary is a list of blind spots that no tool solves, and the value of knowing it is that these are the gaps you build human process around rather than shopping for a product that claims to close them.

Closed channels. WhatsApp groups, private Discord and Slack communities, closed Facebook groups, direct messages. In several markets this is where the majority of real word-of-mouth actually happens, and it is invisible to everyone. A vendor claiming coverage here is either describing something much narrower than it sounds or doing something you would not want to be associated with. The realistic substitute is human: customer service transcripts, sales call notes, community managers who are actually members of the places that matter.

Broadcast. Television and radio monitoring means licensing feeds and running speech recognition over them, which is a fundamentally different business from crawling the web, with different costs and different regional availability. Tools that bundle it are usually reselling a specialist. Ask who actually captures it and for which channels in your market.

Print with no digital edition. Still meaningful in many markets, still largely a manual or scanned-archive process.

Paywalled and subscription-only coverage. Sometimes licensed, often not, and frequently visible only as a headline. Worth establishing which of your most important trade publications fall into this category before you need to know.

Anything deleted. A post that existed for forty minutes and was removed shaped the story anyway. Whether your tool captured it depends entirely on whether its crawl happened to intersect those forty minutes, which is a coin flip dressed up as coverage.

The syndication problem, which quietly distorts everything

One wire story runs in forty outlets. Is that forty mentions or one?

Both answers are defensible and they produce entirely different numbers for volume, share of voice, and whether a crisis threshold has been crossed. What is not defensible is a tool that answers forty in one chart and one in another without saying so, which is more common than it sounds, because deduplication often runs at the article level while the volume timeline counts raw rows.

Worth asking explicitly: how does the system decide two items are the same story? Exact URL matching catches almost nothing, because syndicated copies have different URLs. Canonical URL matching catches more. Title similarity catches more still but starts merging genuinely distinct pieces. Body-text comparison is the most accurate and the most expensive to run at volume.

Then ask what the dashboard actually reports, and make sure the answer is the same in every place a number appears. A share-of-voice figure built on deduplicated stories and a volume chart built on raw rows will disagree, and a client will eventually notice.

What a working system looks like day to day

The best way to evaluate a monitoring tool is not the demo, it is imagining the third Tuesday after you have signed the contract, when the novelty has worn off and the tool has to earn its place in a real workflow.

A team member opens the dashboard once in the morning. It should take them under two minutes to know: is anything currently unusual, is sentiment trending in a direction that needs attention, and are there items that need a response today versus items that can wait for the weekly report. If getting to that answer requires scrolling through a long undifferentiated list, the tool has failed the only test that matters.

Alerts should be rare enough to trust. A system that pings a phone for every mention trains people to ignore it within a week, which means the one alert that mattered gets ignored along with the hundred that did not. Good alerting is conservative by design: it fires when something crosses a threshold calibrated to that specific brand’s normal pattern, not a generic number picked to look responsive in a sales demo.

Reports should answer the question a stakeholder actually asked, not the question the vendor’s template answers. A monthly report full of impression counts and reach estimates that nobody asked for and nobody can verify is worse than no report, because it trains the recipient to skim past the parts that might have mattered.

Questions worth asking before you sign

How is sentiment calculated, and can I see the reasoning behind a specific score? A vendor who cannot explain why a mention scored negative, in plain language, is asking you to trust a black box with a number that will end up in a board deck.

What happens to volume from languages other than English? Ask specifically, because “we cover 40 languages” and “we run 40 languages through the same sentiment model tuned on English” are very different claims, and only one of them produces numbers you can trust.

Is community and social data ever used for a crisis alert? It should not be, unless the vendor can show that specific source can measure real volume rather than search-engine discovery artefacts. A tool that treats a Google-indexed forum thread the same as a verified social post is building crisis alerts on a foundation that cannot tell a real spike from noise.

What is excluded from the “sources” number in the pricing page? Every vendor quotes a large source count. Few explain that regional press, paywalled outlets, or certain social platforms are add-ons, or excluded entirely. Get the actual list for your specific market before comparing prices.

Can I see a client’s dashboard from your category, not a generic demo account? A generic demo is built to look good. A real client dashboard, even anonymised, shows you what the tool looks like on a normal Tuesday with real, messy data in it.

The buying mistake that repeats

The most common mistake is buying breadth before buying reliability. A vendor that promises the most sources, the most languages and the lowest price is usually thin somewhere that only becomes visible once real volume hits the system: sentiment that does not match a native speaker’s read, a crisis alert that fires three hours after the story broke, a report that quietly drops a category of source without saying so.

The teams that get the most value from monitoring are not the ones with the biggest source list. They are the ones whose tool tells the truth about what it can and cannot do, and shows its reasoning instead of asking for trust. That is a harder thing to demo than a source count, and it is the only thing that matters once the contract is signed.

Everything in this guide applies whichever tool a team ends up choosing. What monitoring is for, what to demand from it, and what to be suspicious of does not change based on the vendor’s name on the invoice. What changes is whether the vendor is honest about the gap between what monitoring can tell you and what a team still has to decide for itself.

For teams working in India, see how this applies to media monitoring in India, where regional-language press and video carry stories long before English outlets do.

Common questions

What is media monitoring?

Media monitoring is the ongoing tracking of what is published about a brand, a person, a product or a topic, across press, broadcast, social, video and increasingly AI answers, so a team can see coverage as it happens rather than finding out from a client or a headline.

What is the difference between media monitoring and a press clipping service?

A clipping service collects mentions after the fact, usually once a day. Media monitoring, done properly, watches continuously, attaches sentiment and subject to each item, and surfaces the ones that need a response before the news cycle moves on.

Do I need media monitoring if I already have Google Alerts?

Google Alerts finds a fraction of what a real monitoring engine finds, has no sentiment, no language coverage beyond what Google indexes well, and no way to tell a spike from normal noise. It is a starting point, not a system.

How much does media monitoring cost?

Pricing varies by source breadth, language coverage and how much analysis is layered on top of raw mentions. The honest answer is that a tool priced far below the market for genuine multi-language, multi-channel coverage is usually cutting corners somewhere you will not see until you need it.

What can media monitoring not see?

Closed channels such as WhatsApp groups, private communities and direct messages, which no tool can access. Broadcast without a separate licensing arrangement. Print with no digital edition. Much paywalled coverage. And anything deleted before a crawler reached it. Public Facebook discovery also narrowed sharply after Meta shut down CrowdTangle in August 2024, so a vendor listing Facebook usually means your own page's comments rather than third-party posts.

Can media monitoring replace a PR agency?

No. Monitoring tells you what is being said and how it is moving. An agency, or an in-house comms team, still has to decide what to do about it. The tools that claim to replace judgement are selling something they cannot deliver.