Sentiment Analysis Tools: Social Suites, APIs and Limits
Sentiment analysis tools label text as positive, negative or neutral, and some add a score for how strong the feeling is. For social media they come in three forms: listening suites that run sentiment over the mentions they collect, cloud APIs that developers call with their own text, and free open-source models. Their weak spots, which several vendors admit on their own pages, are sarcasm, mixed feelings and text unlike what the model was trained on.
Short answer. If you already pay for a listening tool, use its sentiment and check a sample by hand. If you have a developer, a cloud API or VADER costs little or nothing to try on your own posts. Either way, never report a sentiment percentage you have not spot-checked.
Part of our social media marketing for small business playbook. Tool descriptions come from each vendor’s own pages, dated by their stamps; order is not a ranking.
Table of Contents
- What sentiment analysis tools do
- How sentiment scores work
- Social media sentiment analysis inside listening tools
- Sentiment analysis APIs for developers
- Free and open-source sentiment models
- What sentiment analysis gets wrong
- How to test a sentiment tool on your own posts
- Sentiment analysis tools FAQ
What sentiment analysis tools do
They sort text by the feeling it expresses, so you can follow brand sentiment across thousands of posts instead of reading each one. Brand24, for example, says it sorts the mentions it collects into positive, negative and neutral using machine learning. Checked against Brand24: Sentiment AnalysisVendor's own description
Which kind you need depends on where your text comes from:
| Kind | You give it | Best for | Needs a developer? |
|---|---|---|---|
| Listening suite with sentiment | Keywords; the tool collects public posts it can access | Marketing teams tracking brand sentiment over time | No |
| Cloud NLP API | Your own text, sent by code | Scoring comments, reviews or survey answers you already hold | Yes |
| Open-source model | Your own text, run on your machine | Free experiments and custom pipelines | Yes |
Sentiment is one layer of listening; social listening tools compares the suites on everything else, and social media analytics tools covers measuring your own accounts.
How sentiment scores work
Each engine reports sentiment differently, so a “neutral” from one tool is not the same as a “neutral” from another.
- Google Cloud Natural Language returns a score from -1.0 (negative) to 1.0 (positive) and a magnitude from 0 upward for the overall strength of emotion, for the whole text and each sentence. Checked against Google Cloud: Natural Language API basicsVendor's own description
- Amazon Comprehend labels text Positive, Negative, Mixed or Neutral and returns a likelihood score for each label; a separate targeted feature scores sentiment towards specific entities. Checked against Amazon Comprehend Developer Guide: SentimentVendor's own description
- Azure AI Language labels text positive, neutral or negative with confidence scores at sentence and document level, and its opinion mining (aspect-based sentiment) ties opinions to attributes such as product features. Checked against Microsoft Learn: What is sentiment analysis and opinion mining in Azure Language?Vendor's own description
Aspect-based sentiment matters for brands. As an illustration, a post that praises a product but complains about delivery reads as mixed overall, but as two clear opinions once split by aspect.
Social media sentiment analysis inside listening tools
For most marketing teams, sentiment arrives as a feature of a social listening suite, not a separate purchase. None of the pages cited here gives an accuracy figure; only the pricing model is shown.
| Tool | What the vendor says about sentiment | Pricing model |
|---|---|---|
| Brand24 | Positive, negative and neutral, in more than 90 languages. Checked against Brand24: Sentiment AnalysisVendor's own description | Four published plans plus Enterprise; 14-day trial, no card; no free plan. Checked against Brand24: Pricing and PlansVendor's own description |
| Brandwatch | Listen classifies conversations as positive, negative or neutral and alerts on spikes in mentions. Checked against Brandwatch: Listen (social listening software)Vendor's own description | No public price; demo request. Checked against Brandwatch: Listen (social listening software)Vendor's own description |
| Lumen by Talkwalker (Hootsuite) | Talkwalker is now Lumen by Talkwalker Checked against Talkwalker home pageVendor's own description; Hootsuite says Lumen tracks sentiment over time with real-time alerts. Checked against Hootsuite: Listening (Lumen)Vendor's own description Talkwalker’s page says it offers sentiment in 127+ languages. Checked against Talkwalker: Social media listeningVendor's own description | Quote-based, through a demo. Checked against Talkwalker home pageVendor's own description |
| Meltwater | Alerts on volume spikes and sentiment shifts via Slack or Teams. Checked against Meltwater: Social ListeningVendor's own description | No prices shown; tailored quotes; no self-serve trial, only a guided demo. Checked against Meltwater: Pricing & CostsVendor's own description |
| YouScan | Aspect sentiment analysis across social media, blogs, forums, reviews and news. Checked against YouScan: Pricing PlansVendor's own description | One published entry plan, larger plans on request; trial is a demo account. Checked against YouScan: Pricing PlansVendor's own description |
| Awario | Mention sentiment and share of mentions across social media, news, blogs, videos, forums and reviews. Checked against Awario home pageVendor's own description | Published plans; free trial of Starter, no card. Checked against Awario: PricingVendor's own description |
| Sprout Social Listening | A sentiment tracking and scoring model with spike alerts, sold as an add-on to Standard and up. Checked against Sprout Social: PricingVendor's own description | Add-on; no add-on price shown. Checked against Sprout Social: PricingVendor's own description |
| Emplifi | Customer sentiment tracking inside its Unified Analytics. Checked against Emplifi: Social Media Analytics (Unified Analytics)Vendor's own description | “Starting at” prices for two tiers, custom quote for the top tier. Checked against Emplifi: PricingVendor's own description |
| Keyhole | Sentiment as part of social listening; Muck Rack announced it had acquired Keyhole on 3 September 2024. Checked against Muck Rack blog: We're Bringing Social Media Listening to Muck Rack (3 September 2024)Archived copy, 19 August 2026 View the archived copy | Sold through Muck Rack by demo; pricing not verified. Checked against Keyhole: HomeArchived copy, 23 October 2026 View the archived copy |
A suite only scores what it can collect: public posts its sources can access, not every mention of your brand. Mention, for one, says it does not monitor Facebook public profiles or groups, citing Facebook’s data policy. Checked against Mention Help Center: Mention Sources Explained (updated 25 Aug 2026)Vendor's own description If your text is surveys, reviews and support tickets rather than social posts, customer-experience platforms fit better: Chattermill sends Slack alerts when sentiment shifts or a theme spikes Checked against Chattermill: HomeVendor's own description, Medallia lists text analytics among its features and sells annual tiers with no published prices Checked against Medallia: PricingVendor's own description, and Qualtrics asks you to request pricing. Checked against Qualtrics: Pricing & PlansVendor's own description
Sentiment analysis APIs for developers
The three big clouds each sell sentiment as a pay-as-you-go API with a free monthly allowance (Amazon’s lasts 12 months).
| API | Output | Free allowance and billing | Status |
|---|---|---|---|
| Google Cloud Natural Language | Score and magnitude, document and sentence level | First 5,000 sentiment units a month free; then per 1,000-character unit. Checked against Google Cloud: Natural Language API pricingVendor's own description | Available |
| Amazon Comprehend | Positive, Negative, Mixed or Neutral with likelihood scores | 50,000 units per API per month for 12 months; per 100-character unit, 3-unit minimum. Checked against AWS: Amazon Comprehend pricingVendor's own description | Topic modeling, event detection and prompt safety classification are closed to new customers; AWS says other Comprehend features are not affected. Checked against Amazon Comprehend Developer Guide: feature availability changeVendor's own description |
| Azure AI Language | Positive, neutral or negative with confidence scores; opinion mining | 5,000 free text records a month, shared with other Language features; per 1,000-character record. Checked against Azure: Language service pricingVendor's own description | Retires on 31 March 2029; Microsoft points new projects to Microsoft Foundry. Checked against Microsoft Learn: What is sentiment analysis and opinion mining in Azure Language?Vendor's own description |
Don’t start a new Azure build without that retirement date in your plan.
Free and open-source sentiment models
If you can run a few lines of Python, you can score sentiment for free.
- VADER is an open-source (MIT licence), lexicon- and rule-based tool that its authors say is attuned to social media text, including slang, emoticons such as :) and emojis. Checked against GitHub: cjhutto/vaderSentiment (README)Vendor's own description It was published by C. Hutto and Eric Gilbert of Georgia Tech at ICWSM 2014. Checked against Hutto & Gilbert (2014), VADER: A Parsimonious Rule-Based Model for Sentiment Analysis of Social Media Text, Proceedings of ICWSM 8(1), 216–225 (AAAI)
- NLTK, the free, open-source Natural Language Toolkit, ships VADER in its sentiment module. Checked against NLTK: Natural Language ToolkitVendor's own description
- TextBlob, a Python library, returns a polarity score from -1.0 to 1.0 and a subjectivity score from 0.0 (very objective) to 1.0 (very subjective). Checked against TextBlob documentation: Tutorial: QuickstartVendor's own description
- Cardiff NLP’s twitter-roberta-base-sentiment-latest is an open model trained on about 124 million tweets from January 2018 to December 2021, labelling text Negative, Neutral or Positive and suited to English. Checked against Hugging Face: cardiffnlp/twitter-roberta-base-sentiment-latest (model card)Vendor's own description
Can ChatGPT do sentiment analysis? A 2023 preprint by Zhang and colleagues tested large language models such as ChatGPT on 13 sentiment tasks across 26 datasets: they performed satisfactorily on simpler tasks but lagged behind smaller models trained on domain-specific data on complex tasks or ones needing structured sentiment output. Checked against arXiv 2305.15005: Sentiment Analysis in the Era of Large Language Models: A Reality Check (Zhang et al., May 2023) The models tested are older than today’s, so test on your own posts; ChatGPT’s pricing page lists a Free plan to try it with. Checked against ChatGPT: Plans and PricingVendor's own description
What sentiment analysis gets wrong
The vendors’ own pages list the limits; read them before you put a sentiment chart in a report.
- Sarcasm. Microsoft says its model may not recognise sarcasm because it sees only the text; it adds that the model was trained on product and service reviews and may be less accurate for under-represented dialects and other domains. Checked against Microsoft Learn: Transparency note for Sentiment AnalysisVendor's own description Brand24 says its models understand sarcasm “in some cases at least” and that its sentiment is “not 100% accurate”. Checked against Brand24: Sentiment AnalysisVendor's own description
- Neutral vs mixed. Google warns that a score near 0.0 can mean a low-emotion text or mixed emotions that cancel out, and that magnitude helps tell them apart. Checked against Google Cloud: Natural Language API basicsVendor's own description
- New slang. The Cardiff model’s training tweets end in December 2021. Checked against Hugging Face: cardiffnlp/twitter-roberta-base-sentiment-latest (model card)Vendor's own description For how brands use emojis in posts, see emojis for social media marketing.
- Lab results are not your results. VADER’s authors report an F1 accuracy of 0.96 against 0.84 for individual human raters, but on the tweets they tested in 2014, not on your brand’s posts. Checked against Hutto & Gilbert (2014), VADER: A Parsimonious Rule-Based Model for Sentiment Analysis of Social Media Text, Proceedings of ICWSM 8(1), 216–225 (AAAI)
Microsoft also says there is no universal accuracy figure, because sentiment is subjective. Checked against Microsoft Learn: Transparency note for Sentiment AnalysisVendor's own description
How to test a sentiment tool on your own posts
Hand-label a sample, run the tool on it and compare, before you trust any trend line. Microsoft’s transparency note itself tells users to test on their own data. Checked against Microsoft Learn: Transparency note for Sentiment AnalysisVendor's own description
- Pull a sample of comments or mentions from your own accounts’ exports or from a tool’s trial, large enough to include every kind of post you usually get. Never scrape a platform; use data you are allowed to access.
- Label each by hand as positive, negative, neutral or mixed, with two people if you can, and note sarcasm and emoji-only posts.
- Run the same sample through the tool and count where it agrees with you.
- Set your own thresholds. Google recommends defining what counts as clearly positive or negative, then adjusting after testing. Checked against Google Cloud: Natural Language API basicsVendor's own description
- Look at the misses. If they cluster on sarcasm, slang or one language, review those posts by hand in every report.
- Keep settings fixed month to month so a change in the chart means a change in opinion, not in setup.
Sentiment is one input to brand health tracking; report it next to the social media KPIs you already track, in the social media report template.
Sentiment analysis tools FAQ
How accurate is sentiment analysis?
There is no single figure. Microsoft says sentiment is subjective, so it gives none, and Brand24 says its own sentiment analysis is not 100% accurate. Test any tool on a hand-labelled sample of your own posts before you trust its numbers.
Are there free sentiment analysis tools?
Yes. VADER is open source under the MIT licence and its authors say it is attuned to social media text, and Cardiff NLP’s twitter-roberta-base-sentiment-latest is an open model for English. Google Cloud Natural Language, Amazon Comprehend (for the first 12 months) and Azure AI Language also have free monthly allowances.
Can sentiment analysis detect sarcasm?
Only sometimes. Microsoft says its model may not recognise sarcasm, and Brand24 says its models understand sarcasm in some cases at least. Check sarcastic posts by hand.
What is VADER sentiment analysis?
VADER (Valence Aware Dictionary and sEntiment Reasoner) is a free, open-source, lexicon- and rule-based sentiment tool that its authors say is tuned to social media text, including slang, emoticons and emojis. It was published at ICWSM 2014, and NLTK includes it.