What Is Social Listening?

In communication research, Margaret C. Stewart and Christa L. Arnold described social listening in the International Journal of Listening in 2017 as a type of listening and a means of attaining interpersonal information and social intelligence. In marketing, the term refers to a brand listening to what is said about it.

The practical difference from monitoring lies in what is done with the data. Monitoring logs mentions one by one: how many there are, where they appear, which ones need a reply. Listening looks at the whole: which topics recur, which complaints are growing, how the tone shifts after a launch. The same mention serves both. For monitoring, a complaint about a late delivery is a case someone has to answer today. For listening, it is the third complaint this month about the same carrier, and that points to a logistics decision.

Social listening also finds what the brand does not control: user-generated content that does not tag it, an unlinked brand mention in a forum, or a conversation about a competitor. That makes it part of the measurement side of social media marketing.

The work has three steps: getting access to the data, building the query and classifying what it finds.

Access depends on each platform, and the terms change. As of 27 October 2026, according to each platform's documentation: - X: the API is billed per use, with no subscription; reading a post costs 0.005 US dollars, with a cap of 3 million post reads per monthly billing cycle. Above that, an Enterprise plan is needed. - Meta: CrowdTangle has been unavailable since 14 August 2024. On the same page, Meta points to its Meta Content Library, which only admits academic institutions or not-for-profit organisations with a research purpose (page updated 30 April 2026). - Reddit: commercial use of its Data API requires a separate agreement with Reddit (terms revised 20 July 2026). - TikTok: the Research API is not open to creators, advertisers or commercial users (FAQ updated 1 September 2026). - YouTube: the Data API grants a default 10,000 units per day, with a separate allocation of 100 search calls a day.

The EU Digital Services Act requires very large platforms to give access to publicly accessible data, but only to researchers studying systemic risks in the Union (Article 40(12)). It is not a route for marketing. What a listening tool can see therefore depends on its agreements with each platform, and it is worth asking which sources it actually covers.

The query decides what gets in. It includes the brand name, its variants and misspellings, products, profiles and the brand's own hashtags. Exclusions matter just as much. A car brand called Jaguar has to exclude the animal: ("Jaguar" OR #jaguar) -zoo -wildlife -bigcat. X's API documentation sets some rules that work as an example of the syntax: - A leading dash excludes posts containing that term. - Parentheses group operators. - A whole group should not be negated: instead of skiing -(snow OR day), write skiing -snow -day. - A negated operator cannot be used alone. - Queries have a length limit, between 512 and 4,096 characters depending on the type of search and access.

Before a query is accepted, a sample of results is checked by hand to see what slips in and what is missing.

Listening provides data that no survey asks for: customers' own vocabulary, recurring problems and what is said about competitors. But it has one limit worth knowing before reading a sentiment chart.

Automatic sentiment analysis struggles with irony. Riloff and colleagues described it in 2013: sentiment analysis can be easily misled by words with a strong polarity that are used sarcastically. "Great, the battery's dead again" contains a positive word and a complaint. In SemEval-2018 Task 3, an academic competition on irony detection in English tweets with 3,834 training and 784 test tweets, the best system reached an F1 of 0.71 when deciding whether a tweet was ironic and 0.51 when telling types of irony apart. F1 combines precision and recall; 1 would be perfect. A share of positive mentions without a manual check of a sample says little.

Data protection counts too. A post with a username is personal data even when it is public, and analysing it needs a legal basis under the GDPR (Article 6). If that basis is legitimate interest, the interests and fundamental rights of the people concerned must not override it (Article 6(1)(f)). If the analysis infers special categories of data, such as health or political opinions, the conditions of Article 9 apply.

• Keep monitoring, with daily replies, apart from listening analysis, with periodic reports.
• Write the exclusions before launching the query and check them against a sample.
• Find out which platforms the tool covers and with what access.
• Check a sample of the sentiment by hand in every period.
• Store only the personal data the analysis needs.

• Calling a count of mentions listening.
• Launching a query for an ambiguous name without exclusions.
• Taking automatic sentiment as an exact figure.
• Assuming the tool sees every network equally well.
• Treating public posts as data outside the GDPR.