# Topic-First: Monitoring by Topic and Keyword

> Topic-First starts with a topic: keyword monitoring, media monitoring, social listening. What it is good for, where it stops and how it relates to Source-First.

- Author: Dr. Karsten Richter
- Publisher: Picasi GmbH
- Topic: Approaches compared
- Last updated: 2026-09-24
- Canonical URL: https://source-first-intelligence.com/en/topic-first/
- Language: en


The term brings several common practices together under one name. What they share is the starting point: you decide what you want to learn about and leave open who the results come from. Topic-First asks “what?”. On this site it is one of three logics of monitoring, alongside Source-First (“who?”) and [Platform-First](https://source-first-intelligence.com/en/platform-first/) (“where?”). All three are compared in [Market Monitoring: Topic-First, Source-First, Platform-First](https://source-first-intelligence.com/en/three-approaches-to-market-monitoring/).

## Three practices with the same starting point

**Keyword monitoring** reports new results for a search term. Google describes its Google Alerts service like this: you can get emails [“when new results for a topic show up in Google Search”](https://support.google.com/websearch/answer/4815696?hl=en). In the [interface](https://www.google.com/alerts?hl=en), “Sources” lists types of channel such as News, Blogs, Web or Video. There is no setting for the sender.

**Media monitoring** tracks what the press, broadcasters and online media report about a company or a topic. The measurement association AMEC calls this third-party coverage [“earned media”](https://amecorg.com/glossary/): “third-party media coverage secured through a relationship or news worthy event, rather than paid-for advertising”.

**Social listening** analyses what people say on social networks and across the web about a brand, a product or a topic, often with sentiment analysis. AMEC describes the related field of “social analytics” as listening to [“the distributed conversations about a particular brand, product or issue”](https://amecorg.com/glossary/). Margaret C. Stewart and Christa L. Arnold describe social listening as [“an emerging type of listening”](https://scholars.unf.edu/en/publications/defining-social-listening-recognizing-an-emerging-dimension-of-li/).

## What Topic-First is good for

Topic-First is fixed on the topic and open on the sources. That is its strength when you do not know who will speak about a topic.

**Discovering the unknown.** A list of senders contains only the ones you know. George Day and Paul Schoemaker describe in Harvard Business Review how Mattel, between 2001 and 2004, [lost 20% of its share of the worldwide fashion-doll segment to smaller rivals](https://hbr.org/2005/11/scanning-the-periphery) and “didn’t see it coming”. New rivals, the authors write, often seem to come out of left field. Watching a topic is one way to see such players early. Social media research treats [discovery as a step of its own before data collection](https://www.research.ed.ac.uk/en/publications/social-media-analytics-challenges-in-topic-discovery-data-collect/).

**Seeing the breadth of a topic.** For a new technical term, a standard or a draft law, the question really is “where does this show up?”. If completeness matters, an open search space is the right choice. Manning, Raghavan and Schütze’s textbook notes that [paralegals and intelligence analysts](https://nlp.stanford.edu/IR-book/html/htmledition/evaluation-of-unranked-retrieval-sets-1.html) aim for as high recall as possible and tolerate fairly low precision to get it.

**Knowing what third parties say.** In crisis communication, what matters is the conversation about a company, not its own statement. A systematic review by Mats Eriksson (2018) lists [“use social media monitoring”](https://instituteforpr.org/lessons-for-crisis-communication-on-social-media-a-systematic-review-of-what-research-tells-the-practice/) among the five lessons research offers for effective crisis communication on social media. For PR evaluation, coverage analysis is a common method.

## Where the limits are

**Language.** A word can have several meanings, and then the search returns irrelevant hits; the textbook calls this [polysemy](https://nlp.stanford.edu/IR-book/html/htmledition/latent-semantic-indexing-1.html). The same thing can be said in different words, and then hits are missing: synonymy [affects the recall of most information retrieval systems](https://nlp.stanford.edu/IR-book/html/htmledition/relevance-feedback-and-query-expansion-1.html). For Boolean search the authors add that [“it is difficult or impossible to find a satisfactory middle ground”](https://nlp.stanford.edu/IR-book/html/htmledition/the-extended-boolean-model-versus-ranked-retrieval-1.html) between precision and recall.

**The sender.** Because the sources are open, the sender plays no part in whether you are notified. That is intended when you are looking for something new. It becomes a limit when you want to know what a specific market player says itself: its announcement appears only if it contains the keyword.

**Machine-generated content.** In [Findings of ACL 2024](https://aclanthology.org/2024.findings-acl.103/), Thompson and colleagues show that low-quality, machine-translated content makes up a large fraction of all web content in those lower-resource languages. Shumailov and colleagues write in [Nature](https://www.pure.ed.ac.uk/ws/portalfiles/portal/460496122/ShumailovEtalNature2024AIModelsCollapseWhen.pdf) that “it is unclear how content generated by LLMs can be tracked at scale.” Neither study measures its share of the web as a whole, so this site gives no figure. In our assessment, a search without control over the sources can push a lot of machine-generated content to the top, as long as it contains the keyword. Checking the origin is then left to the reader.

## How it relates to Source-First

Topic-First and Source-First answer different questions. One is “where does this show up?”, the other “what is this market player saying?”. For market and competitor analysis this site argues for Source-First, because there the second question usually matters more. That does not make Topic-First redundant.

The two work together as a division of labour: **keywords to find, sources to follow.** Topic-First finds new names at the edge of your field of view. For every hit, ask who said it and whether that sender belongs on your list of sources. For the question of what specific market players say, Source-First then takes over ongoing monitoring. For brand perception and crisis communication, Topic-First remains the right tool. Questions such as “Who is hiring right now?” are answered by [Platform-First](https://source-first-intelligence.com/en/platform-first/); how it relates to Source-First is shown in [Source or Platform?](https://source-first-intelligence.com/en/source-or-platform/). The comparison [Sources, not keywords](https://source-first-intelligence.com/en/sources-vs-keywords/) shows how Topic-First and Source-First work together in detail, and the [Source-First Manifesto](https://source-first-intelligence.com/en/manifest/) sets out the underlying idea.

## References

1. Google: Create an alert. Google Search Help, undated. [https://support.google.com/websearch/answer/4815696?hl=en](https://support.google.com/websearch/answer/4815696?hl=en) (accessed 23 September 2026)
2. Google: Google Alerts, interface with options (how often, sources, language, region, how many). Undated. [https://www.google.com/alerts?hl=en](https://www.google.com/alerts?hl=en) (accessed 23 September 2026)
3. AMEC (International Association for Measurement and Evaluation of Communication): Glossary. Undated. [https://amecorg.com/glossary/](https://amecorg.com/glossary/) (accessed 23 September 2026)
4. Stewart, Margaret C.; Arnold, Christa L.: Defining Social Listening: Recognizing an Emerging Dimension of Listening. International Journal of Listening 32 (2), pp. 85–100, 4 May 2018, DOI 10.1080/10904018.2017.1330656. [https://scholars.unf.edu/en/publications/defining-social-listening-recognizing-an-emerging-dimension-of-li/](https://scholars.unf.edu/en/publications/defining-social-listening-recognizing-an-emerging-dimension-of-li/) (accessed 23 September 2026)
5. Day, George S.; Schoemaker, Paul J. H.: Scanning the Periphery. Harvard Business Review, November 2005. [https://hbr.org/2005/11/scanning-the-periphery](https://hbr.org/2005/11/scanning-the-periphery) (accessed 23 September 2026)
6. Stieglitz, Stefan; Mirbabaie, Milad; Ross, Björn; Neuberger, Christoph: Social media analytics – Challenges in topic discovery, data collection, and data preparation. International Journal of Information Management 39, pp. 156–168, 30 April 2018, DOI 10.1016/j.ijinfomgt.2017.12.002. [https://www.research.ed.ac.uk/en/publications/social-media-analytics-challenges-in-topic-discovery-data-collect/](https://www.research.ed.ac.uk/en/publications/social-media-analytics-challenges-in-topic-discovery-data-collect/) (accessed 23 September 2026)
7. Manning, Christopher D.; Raghavan, Prabhakar; Schütze, Hinrich: Introduction to Information Retrieval, section 8.3 “Evaluation of unranked retrieval sets”. Cambridge University Press, 2008. [https://nlp.stanford.edu/IR-book/html/htmledition/evaluation-of-unranked-retrieval-sets-1.html](https://nlp.stanford.edu/IR-book/html/htmledition/evaluation-of-unranked-retrieval-sets-1.html) (accessed 23 September 2026)
8. Eriksson, Mats: Lessons for Crisis Communication on Social Media: A Systematic Review of What Research Tells the Practice. Author’s summary at the Institute for Public Relations, 23 October 2018 (original article: International Journal of Strategic Communication 12 (5), pp. 526–551, 2018). [https://instituteforpr.org/lessons-for-crisis-communication-on-social-media-a-systematic-review-of-what-research-tells-the-practice/](https://instituteforpr.org/lessons-for-crisis-communication-on-social-media-a-systematic-review-of-what-research-tells-the-practice/) (accessed 23 September 2026)
9. Manning; Raghavan; Schütze: Introduction to Information Retrieval, section 18.4 “Latent semantic indexing”. Cambridge University Press, 2008. [https://nlp.stanford.edu/IR-book/html/htmledition/latent-semantic-indexing-1.html](https://nlp.stanford.edu/IR-book/html/htmledition/latent-semantic-indexing-1.html) (accessed 23 September 2026)
10. Manning; Raghavan; Schütze: Introduction to Information Retrieval, chapter 9 “Relevance feedback and query expansion”. Cambridge University Press, 2008. [https://nlp.stanford.edu/IR-book/html/htmledition/relevance-feedback-and-query-expansion-1.html](https://nlp.stanford.edu/IR-book/html/htmledition/relevance-feedback-and-query-expansion-1.html) (accessed 23 September 2026)
11. Manning; Raghavan; Schütze: Introduction to Information Retrieval, section 1.4 “The extended Boolean model versus ranked retrieval”. Cambridge University Press, 2008. [https://nlp.stanford.edu/IR-book/html/htmledition/the-extended-boolean-model-versus-ranked-retrieval-1.html](https://nlp.stanford.edu/IR-book/html/htmledition/the-extended-boolean-model-versus-ranked-retrieval-1.html) (accessed 23 September 2026)
12. Thompson, Brian; Dhaliwal, Mehak; Frisch, Peter; Domhan, Tobias; Federico, Marcello: A Shocking Amount of the Web is Machine Translated: Insights from Multi-Way Parallelism. Findings of the Association for Computational Linguistics: ACL 2024, pp. 1763–1775, August 2024. [https://aclanthology.org/2024.findings-acl.103/](https://aclanthology.org/2024.findings-acl.103/) (accessed 23 September 2026)
13. Shumailov, Ilia; Shumaylov, Zakhar; Zhao, Yiren; Papernot, Nicolas; Anderson, Ross; Gal, Yarin: AI models collapse when trained on recursively generated data. Nature 631, pp. 755–759, 24 July 2024, DOI 10.1038/s41586-024-07566-y, version of record in the University of Edinburgh repository. [https://www.pure.ed.ac.uk/ws/portalfiles/portal/460496122/ShumailovEtalNature2024AIModelsCollapseWhen.pdf](https://www.pure.ed.ac.uk/ws/portalfiles/portal/460496122/ShumailovEtalNature2024AIModelsCollapseWhen.pdf) (accessed 23 September 2026)