AI Sentiment Analysis

We're excited to introduce AI Sentiment Analysis, available in NewsWhip by Sprout Social. It reads every article in your search as it publishes and tells you how that coverage lands for your brand, scored automatically and measured your way.
For most teams, that means a clear, reliable read on sentiment without anyone reading the coverage first. For teams tracking a specific issue or strategy, it goes further: you define the question, and every article is judged against your own brief.

What are the benefits?

NewsWhip has always shown you what's being published about your brand, and how fast it's spreading. What it couldn't tell you was how that coverage lands. Until now, answering that took a manual read: an analyst going through articles one by one to decide whether each was good or bad news. That approach doesn't scale with volume, and it rarely keeps pace with a story that's still developing.

Most sentiment tools try to automate this by scoring the tone of the language. The trouble is that tone and reputation are not the same thing. Take a story about a data breach at three companies in your industry that names you as the one that wasn't affected. It's full of negative language, but for you it's good news. A tone-based tool marks it negative. AI Sentiment Analysis reads it the way your own team would.

With AI Sentiment Analysis, you can:

• Know how coverage lands without reading it first. Every article in your search is classified as it publishes, so sentiment is something you can see while a story is live, not something you reconstruct at the end of the month.
• Measure against your own definitions. Tell the Analyzer what positive and negative mean for your organisation, or define your own categories entirely.
• Answer the questions leadership actually asks. Show whether your key messages are landing, which narratives are gaining ground, and who the coverage is holding responsible during an issue.
• Trust what you report. Every point on a chart opens the articles behind it, so every number can be checked.

How does AI Sentiment Analysis work?

AI Sentiment Analysis adds a new widget type to your NewsWhip dashboards, listed as AI Analysis. Each one you create is called an Analyzer. An Analyzer runs on a single NewsWhip search and works continuously in the background. You can run several Analyzers on different searches, each answering a different question.

Three things make it work:

• A reader for every article: the Analyzer reads each article in full as it publishes, headline and body, not just the first few lines. That's how it catches the implication in the fifth paragraph, the passing mention that isn't really about you, or the criticism that's aimed at someone else.
• Your worldview: the Analyzer judges each article against your question and your categories, with any rules you've set. It asks whether a reasonable reader would think differently about you after reading it, not whether the words are positive or negative.
• Results you can trace: classifications build into a trend over time. Select any point on the chart to see the articles behind it and a short summary of what's driving the change. If you disagree with a classification, you can correct it.

Getting started with sentiment

Sentiment is where most teams begin, and for many it's all they need. Setting up your first Analyzer takes less than a minute.

Set up a sentiment Analyzer

1. Open a search on your dashboard and select Add Widget, then choose AI Analysis from the widgets panel.
2. Choose the sentiment question from the suggested list, for example "What is the sentiment toward [your brand]?"
3. Confirm the target, then select Start Analysis.

The Analyzer starts classifying new articles straight away and backfills the last month of coverage, so you have a trend line from day one. You'll usually see results within an hour. The backfill can take up to 24 hours, so your charts will look sparse until it completes.

Get the target right

Sentiment is measured towards your target, not towards the article as a whole, which makes it the most important setting. Be specific: Ford Motor Company rather than Ford, Shell PLC rather than Shell. To measure sentiment towards a topic, make the topic itself the target, for example Ultra-processed foods or Autonomous vehicles.

Tell it what counts (optional)

Classification rules let you define what positive, negative and neutral mean for your organisation. They're worth adding wherever your judgement differs from the obvious reading. For example: "Coverage of an industry-wide problem counts as negative for us, even when we aren't named", or "Routine financial reporting doesn't count as positive." Keep each rule to a short, concrete sentence.

Explore your results

Your sentiment widget shows how coverage of your target is trending, by day or by hour. Switch between Media Interest (how many articles were published) and Public Interest (how much engagement those articles received), and use the Articles published in total bar to see the overall split.

When something moves, select that point on the chart. You'll see the articles classified at that moment and a short summary of the drivers: what's pushing sentiment up or down. The summary saves you reading time, and the articles behind it are always one click away.

If you disagree with a classification, open the article, select ⋮, choose the correct category and select Update classification. Your chart uses your correction from then on. The Analyzer doesn't learn from individual corrections, so if you find yourself correcting the same thing repeatedly, update your classification rules instead.

Going further with narrative intelligence

Sentiment tells you how coverage feels about you. But the questions communications leaders are asked are often more specific. Is the story about our product, or about the regulator? Are the messages we briefed actually being carried? Is that false claim still spreading?

 

These questions can't be answered with a keyword search. "Makes it difficult to access care" is an argument, not a phrase. Answering it takes judgement, which until now meant reading every article yourself. AI Sentiment Analysis applies that judgement at scale: you ask the question and define the categories, and every article is judged against them, the way someone on your own team would judge it.

Choose or write a question
When you add an AI Analysis widget, NewsWhip reads your search and suggests questions a head of communications would ask. For an ongoing search, like brand mentions, the questions focus on your brand. For a developing story, they're drawn from the coverage itself. Pick one, or write your own under Ask your own question. NewsWhip then drafts the categories and their descriptions for you, ready to edit.

What you can ask

Writing good categories
The category description is what the Analyzer judges against, so a sentence or two naming the kind of coverage you have in mind works best. Write for someone outside your company, because internal shorthand won't be understood. There's no need to add a catch-all: one is provided automatically.

Choose the single best category per article when your categories are mutually exclusive, such as who is to blame. Allow multiple categories when an article can carry several narratives at once.

Integrating AI Sentiment Analysis into your workflows

AI Sentiment Analysis works best when it's set up deliberately, on the searches that matter most. We recommend:

• Starting with sentiment on your most important search, usually your brand mentions. Compare the results with how your team would have read the same coverage, and with any sentiment reporting you use today.
• Refining early. Adjust your target, rules and categories in the first week, while there's little history to change.
• Sharing rather than duplicating. Have one person set up Analyzers on a shared dashboard and share it with the team. Everyone can explore the charts, open articles and correct classifications, results stay consistent, and your team's allowance goes further.
• Pausing rather than deleting. Paused Analyzers keep their history and don't count towards your allowance. Deleting an Analyzer removes its data, and a new one starts from scratch.
• Getting results in front of decision-makers. Add AI Analysis widgets to Dashboard Reports, or export any widget as an image for a deck or a morning briefing, or as a CSV.

Your team shares an allowance of Analyzers. To see what's in use and what's left, go to More → AI Analysis.

Good to know:

• Coverage: AI Sentiment Analysis reads news and web articles on searches that include web coverage. It doesn't analyse social media posts.
• History: each Analyzer backfills the last 30 days. Its value builds from the day you switch it on, so start Analyzers on your priority searches early.
• High-volume searches: for searches with more than 1,000 articles a day, results are based on the top 1,000 articles per day by engagement. The widget shows the share of coverage it's based on.
• Duplicates: near-identical articles are recognised, and copies take the original's classification, which keeps your figures consistent.
• Languages: articles in all languages are analysed. Categories, descriptions and driver summaries are shown in English.
• Editing: you can edit a search or an Analyzer at any time. Changes apply from that point forward and are marked on the timeline. Past coverage isn't re-analysed.
• Permissions: dashboard owners can add, edit, pause and delete Analyzers.