Topic Insights chat assistant

While the Topic Insights dashboards allow you to quickly identify the top readership trends in your country, often editors need to drill deeper to find more subtle connections.

To better contextualize viral readership trends, Topic Insights also offers a generative AI-powered chat interface so editors can identify not just what topics are trending now, but which are new today, why they are trending, where they are trending, relational trends between topics, traffic sources for a topic, and generally how many stories are being tracked for a unique topic.

In all of these cases, the goal is to help newsrooms gain key insights to inform content, search, social and messaging strategy.

To access this chat experience, navigate to the Topic Insights main dashboard and click the light purple “Ask User Agent” button in the top right of the page, as below.

The agentic chat is based on anonymized engagement data from Taboola publisher partners in that region, augmented with article-level data. The agent and its queries operate on aggregated data that has been sampled, pseudonymized and grouped to identify trends and patterns. Questions that attempt to bypass this will be declined.

When you first open the chat, the system will offer suggested questions. After each question, agent will suggest ways to further explore. It also can propose alternative ways to phrase a question to better get actionable insights.

For each session, the agent’s chat is persistent, so conversations can be streamlined. A user also can access their full chat history via the clock icon in the top right of the UI. Each session is labeled with the first question asked. To clear the current chat, select the edit icon from the top right, as below.

Chatting with Topic Insights

There is no limit to the questions you can ask the data. However, there are limitations to the data itself, so some questions are more likely to yield better results than others. In general, there are two types of supported questions — questions centered on topics and questions centered on article data.

For topic questions, approaches like “what’s trending to today that didn’t trend yesterday” or “what’s the top trending search-referred topics” will be supported. Article-level questions can expose named entities mentioned in a stories (like states or officials) or how many articles are associated with a single topic.

For example, on the standard Topic Insights dashboard, the size of the topic bubble corresponds to the relative popularity in pageviews of the topic. Larger bubble; more pageviews. However, in that standard view, it’s impossible to determine how many articles are being tracked in relation to that topic. Does a bubble represent 50 articles or 500? The answer could determine whether a publisher would pursue coverage or not.

Here, you can see that fewer than 200 articles are being tracked under topic “Prank”. The UI also offered up samples of some headlines. Now, an assignment editor can make a data-informed decision.

Topic-related questions

The following are examples of topic-related questions the agentic chat can address. Note, it is advised — but not required — to include the word “topic” as part of the syntax. Punctuation is also advised but not required. Generally, the more refined the input the stronger the output.

Samples:

• Why is topic X trending?
• Is topic X trending more today than it did yesterday?
• What topics are trending today that didn’t trend yesterday?
• What are the top 20 trending topics today?
• What were the top 10 trending topics yesterday?
• What topics trended yesterday that didn’t previously trend in the last three days?
• What topics from the top 20 trending topics yesterday weren’t in the top 20 one day earlier?
• What are the top 10 trending topics from search today?
• List 10 search-referred topics that are trending today that didn’t trend yesterday
• What are 5 topics trending on social that are not trending in the top 20 topics generally?
• What other trending topics are associated with topic X?
• List the top 5 search-referred topics that are not in the top 20 trending topics overall?
• What are the top current trending topics unrelated to the topic X?
• Summarize the articles trending on topic X.
• Give me 5 new story ideas based on topic X

Article-related questions

The following are examples of article-related questions the agentic chat can address.

Samples:

• Do any articles on topic X mention location Y?
• What headlines are associated with both topic X and topic Y?
• Give me three examples of articles associated with topic X
• List 10 headlines for the top articles related to the topic X
• Summarize the top 25 articles on topic X. tell me all the details you have
• Analyze the articles associated with topic X. what new articles should i write associated with topic X that would engage readers?
• Ahat locations are mentioned in articles on topic X?
• Analyze the top 20 articles on topic X. What individual is most mentioned by name?
• List all team names related to articles on X?
• What are the top locations referenced in articles on topic X?
• What are the top officials or politicians mentioned in articles on topic X?
• How many articles are you tracking related to topic X?
• Is the traffic to topic X spread out across many articles or relatively few?
• In what states are articles on topic X trending?
• What are the primary referral sources for articles about topic X?
• What are the main referral sources for articles about topic X?
• What are the main traffic sources for articles about X?

Again, all of the examples above are just some of the approaches you can use to question the corpus of the Topic Insights data. The chat is not limited to these alone. These are just some ideas to get started!

Final note on privacy

There is no potential for obtaining personally identifiable information (PII). Taboola is dedicated to providing users the highest level of transparency and control over the use of their data.