AI social listening platforms are sold on the promise that they will show you what the market is saying about you. Most of them will not, because the platforms doing the talking have closed the door on unlicensed access. The pattern is consistent: the tools that look cheapest show you the least, and the tools that show you the most cannot be priced from a website.
Three facts frame everything that follows. X’s terms of service, effective 9 October 2026, prohibit scraping the service “in any form, for any purpose without our prior written consent” and attach liquidated damages of $15,000 for every million posts accessed in a 24-hour period. Reddit’s Data API Terms require a separate negotiated contract before any commercial use. And TikTok’s Research API is restricted to academic and nonprofit researchers working on a non-commercial basis. There is no free tier that gives a brand the conversation.
Which AI social listening platforms are still available in 2026?
Nine platforms remain commercially purchasable, and three names in older comparison guides no longer mean what they used to. Brandwatch (owned by Cision), Meltwater, Sprout Social, Brand24, BuzzSumo, Emplifi by Quadient, Sprinklr and Quid are all active. Hootsuite is active and now includes Lumen, the rebranded Talkwalker engine. Metryka is dead, its domain is parked and listed for sale at $9,995. Qualtrics Social Intelligence no longer has a product page and has been absorbed into Qualtrics Omnichannel Experience Management. NetBase Quid redirects to quid.com and now trades simply as Quid.
| Platform | Ownership and current status (checked 17 Oct 2026) | Publicly verifiable price |
|---|---|---|
| Brandwatch Consumer Intelligence | Cision. Active. X official partner, full firehose claimed | Quote only. No list price published |
| Hootsuite with Lumen | Active. Talkwalker acquired April 2024; its engine is now Lumen | $99 Standard, $199 Professional, $399 Advanced per user/month annual; Enterprise quote |
| Meltwater | Active. Mira AI assistant across all tiers | Indicative only: Starter from $10,000/yr, Pro from $25,000/yr, Enterprise from $130,000/yr |
| Sprout Social | Active. Listening is a paid add-on on Standard and above | $79 Essentials, $199 Standard, $299 Professional, $399 Advanced per seat/month annual |
| Brand24 | Active. Independent | $199 Individual to $1,499+ Enterprise per month on annual billing |
| BuzzSumo | Active. Weighted toward content and PR discovery | $199 Content Creation, $299 PR & Comms, $499 Suite, $999 Enterprise per month annual |
| Emplifi by Quadient | Active. 2026 Gartner MQ Leader for SMM and Listening | Social Marketing from $1,249/mo Essential, $2,499/mo Advanced, Enterprise quote |
| Sprinklr Insights | Active. 33 products across four suites | Quote only |
| Quid (ex-NetBase Quid) | Active. netbasequid.com redirects to quid.com | Quote only |
| Metryka | Dead. Domain parked, listed for sale | Not available |
| Qualtrics Social Intelligence | Retired as a standalone product. Folded into Omnichannel Experience Management | Not separately priced |
What the coverage and pricing figures mean
Coverage comes from vendor documentation and the platforms’ own access terms; prices reflect pages checked on 17 October 2026. Enterprise amounts are published indicative figures, without negotiated quotes. No platform trials were run for this comparison, and sentiment-accuracy claims remain vendor-reported unless an independent study is named.
What data these platforms can actually see
This is the question vendor pages avoid, so I went to the source documentation. The short version: full coverage is rare, partial coverage is usually industry-filtered, and the most valuable networks are the most heavily gated.
X (Twitter) is a metered, pay-per-use resource. X’s API documentation states plainly that it uses pay-per-usage pricing with no subscriptions. Reading a post costs $0.005 per resource, reading a user $0.010, and reads deduplicate within a 24-hour UTC window. Ten thousand posts a day is roughly $1,500 a month. X’s developer agreement, last updated 27 April 2026, states that the Pay-Per-Use, Basic and Pro plans are “designed for hobbyists, commercial prototyping, initial development, early-stage X product integrations,” and that anything beyond that scope requires an Enterprise plan. Brandwatch describes itself as an X official partner offering the full firehose. Sprout Social’s own documentation says something quite different: under its limitations list, it states “X provides sampled data.”
Reddit is contract-only for commercial use. Reddit’s Data API Terms, dated 14 October 2026, state in section 3.1 that Reddit reserves the right to charge fees and that anyone wanting to use the Data APIs “for commercial purposes … will need to enter into a separate agreement with Reddit.” Reddit publishes no price list. The roughly $12,000-per-month figure repeated by third-party blogs is unconfirmed against an authoritative source. Reddit’s terms also prohibit using user content to train a machine learning or AI model without rightsholder permission, and prohibit deriving revenue from the API without express written approval.
TikTok research data is closed to brands. TikTok’s Research API eligibility page requires applicants to be academic institutions or eligible nonprofit research bodies, to be “independent of commercial interests and able to conduct research on a not-for-profit or non-commercial basis,” to disclose funding, and to provide evidence of ethical review. A brand cannot apply. Commercial TikTok coverage therefore exists only through vendors’ own commercial licences.
Instagram is coverage you have to qualify for. Brandwatch’s Listen documentation is unusually candid: it indexes posts from 720,000 non-owned Instagram accounts across a fixed list of about twenty industries including fashion, banking and finance, food and beverage, retail, pharma and healthcare, and technology. Non-owned posts are captured but their comments are not. Public Instagram data dates back to February 2025 for any new query. Sprout Social’s public Instagram source is narrower still: posts containing specific hashtags in the caption, comments excluded, no historical backfill at all.
YouTube depends on who you ask. Google’s YouTube Data API documentation, last updated 15 October 2026, sets a default allocation of 100 search.list calls per day, 100 videos.insert calls, and 10,000 units per day for everything else combined. A brand running monitoring searches gets 100 searches a day on a default project. Sprout Social states its YouTube source covers video descriptions and excludes comments. Brandwatch states it collects YouTube videos and comments via third-party providers.
The open networks are the ones nobody can charge you for. Brandwatch states it ingests the entire Bluesky firehose, roughly eight million mentions per day. That is the only genuinely complete public feed in this category, and also the one with the least commercial value to most brands.
X’s terms of service, effective 9 October 2026, set liquidated damages of $15,000 USD per 1,000,000 posts for requesting, viewing or accessing more than one million posts in any 24-hour period, on top of an explicit prohibition on crawling or scraping “in any form, for any purpose without our prior written consent.”
Source coverage at a glance
| Source | X | TikTok | YouTube | Notes on the limits | ||
|---|---|---|---|---|---|---|
| Brandwatch | Full firehose (official partner) | 400M posts/month from 28 Aug 2026 | Owned channels + brand handle mentions | 720,000 non-owned accounts, posts only, from Feb 2025 | Videos and comments via third parties | Bluesky full firehose (~8M/day); 13 months Reddit history; legacy Reddit data deleted 30 Sep 2026 |
| Sprout Social | Sampled (vendor-stated) | Realtime; manual backfill only 30 days | Business-page posts with your hashtag only; top 1,000 posts | Hashtag posts in caption only, no backfill | Video descriptions only, no comments | Reddit and TikTok both capped; queries limited to 2,037 characters |
| Hootsuite (all plans) | Yes | Enterprise only | Enterprise only | Enterprise only | Yes | Enterprise adds Threads, LinkedIn, Mixcloud, Vimeo, Dailymotion, Pinterest, Twitch, Disqus, Line and 13 months history |
| Meltwater | Claimed, not quantified | Claimed, not quantified | Claimed, not quantified | Claimed, not quantified | Claimed, not quantified | Vendor states 6M+ unique sources, 240+ languages, 1.3B+ documents/day, 15 months social history |
| Brand24 | Listed, share undisclosed | Listed, share undisclosed | Listed | Listed | Listed | No coverage percentage published; keywords and mentions capped by plan |
Those coverage claims for Meltwater and Brand24 are supplier statements on supplier websites. No independent audit of a vendor’s ingestion percentage is cited here. Ask every vendor directly what proportion of each platform’s public volume they ingest and whether that proportion changes by tier.
How honest is AI sentiment analysis on sarcasm, slang and mixed languages?
No independent, peer-reviewed evaluation of a commercial social listening sentiment engine from the last twelve months is cited here. Every accuracy number attached to these products is the vendor’s own, usually with no methodology, no test set and no confidence interval. When a comparison article tells you Brandwatch sentiment is 88 percent accurate, ask who measured it. Usually nobody did.
What documentation does reveal is the honest shape of the problem. Sprout Social’s support documentation explains that its sentiment model is a deep neural network assigning probability scores across positive, negative and neutral, and that messages can land in an “Unclassified” bucket when the language is not supported, when the message is media-only, or when the text is too short or ambiguous. That fourth bucket is the tell. A vendor that routes unknown inputs to “unclassified” is telling you the truth about its coverage. A vendor that scores everything is telling you its model has never met a language it cannot classify, which is not credible.
Language detection is the first filter and it is vendor-dependent. Brandwatch states that for all sources except X it uses the Compact Language Detector, the system also used by Google Chrome and Google Translate, which supports 83 languages; for X posts it defers to X’s own language identification. A sentiment model for a language is not the same thing as a language detector for it, and the difference is where mixed-language posts get lost.
One independent study is Rahul and Adebayo’s 2026 paper, “Beyond Benchmark Accuracy: Robustness Evaluation of Hinglish Sentiment Models,” published in the Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages at ACL 2026. The authors fine-tuned XLM-RoBERTa and mBERT on a deduplicated 25,543-tweet Hinglish sentiment dataset and got 99.7 percent in-distribution accuracy. On a 400-example human-validated adversarial set spanning negation, sarcasm, contrast, subtle sentiment and true neutrality, XLM-R fell to 42.5 percent, a drop of more than 57 percentage points. Zero-shot transfer to English TweetEval gave 50.8 percent accuracy and 40.8 percent macro F1. This is one workshop paper on one research model, not a verdict on any vendor product. It is also the clearest available demonstration that benchmark accuracy is not real-world accuracy, which is precisely the gap between a vendor demo and your brand’s mentions.
For method taxonomy and known weaknesses, Esmaeilzadeh and colleagues’ review of sentiment analysis from social media platforms in Expert Systems with Applications is still the standard map, but it was published in 2023, so it is nearly four years old.
The pricing cliff: what a small team can actually get
The pricing landscape splits cleanly in two. Four vendors publish numbers you can check. Everything aimed at actual listening is quote-only, and the one vendor that discloses a floor discloses it as tens of thousands of dollars per year.
Meltwater is the outlier that helps everyone else. Its own pricing page states that it does not publish a fixed price list but gives indicative ranges: Starter from $10,000 per year, Pro from $25,000 per year, and Enterprise from $130,000 per year. It also states that standard contracts are annual, that a one-time implementation or onboarding charge may apply to enterprise configurations, and that influencer marketing is only available at Enterprise tier and on Agency plans. That is a $10,000 floor for the smallest credible enterprise configuration: still four times Brand24 Individual’s annual cost at $199 per month, before you add any implementation fee.
The published low end is real but narrow. Brand24 Individual at $199 per month on annual billing, or $249 month to month, includes unlimited users, real-time updates and AI sentiment, topic and event detection. Its Pro plan at $399 annually, or $499 monthly, adds 12 keywords and 40,000 mentions per month. The Business plan at $599 annually, or $699 monthly, raises that to 25 keywords and 100,000 mentions. Enterprise starts at $1,499. API access is a separate $149 fee on the lower plans. Brand24 publishes no ingestion percentages, so its source list tells you which networks it reads but not how much of them.
BuzzSumo publishes prices too, at $199, $299, $499 and $999 per month on annual billing, but it is a content and PR discovery tool that has migrated away from general social listening. Its PR & Comms tier monitors 2,000 articles per month and its Enterprise tier 10,000, with journalist lists capped at 20, 100 or 600 names. Useful, not a replacement for listening.
Sprout Social and Hootsuite are per-seat publishing platforms where listening is an upgrade. Sprout Social Essentials at $79 per seat per month annually caps you at five social profiles and no listening; Listening is an add-on on Standard ($199) and above. Hootsuite Standard at $99 per user per month includes basic listening and benchmarking against up to five competitors; Professional at $199 adds 30-day Quick Search, up to 20 competitors and 90-day trend forecasting. Emplifi’s Social Marketing package starts at $1,249 per month, above every self-serve tool while still not including enterprise listening depth.
The hidden line items are where budgets break. Sprout Social charges customers to buy a 12-month or 24-month historical backfill through their customer success representative, so the deep history that makes competitive benchmarking meaningful is a purchased extra rather than an included feature. Brand24 charges $149 for API access on top of the subscription. Meltwater may add implementation fees.
How to run a social listening programme that survives contact with reality
- Write down the decision, not the platform. A weekly competitor share-of-voice review and a 2 a.m. recall response need different data, and only one justifies enterprise pricing. Vendors pitch platforms; you should pitch yourself a question.
- Map the conversation before you buy. Take the last ten days of real customer questions from your support inbox and sales calls. Which network do those people actually use? If your category lives in Discord, TikTok comments and a subreddit, no tool will find it for you and no budget will fix it.
- Verify each source in writing before signing. Ask each vendor for the documented ingestion rate per network, the historical depth in months, the latency profile, and whether those figures change by tier. “Comprehensive coverage” is not an answer.
- Treat X as a sampled signal, not a census. Unless your contract says full firehose, design your alerts around relative change rather than absolute volume. A share-of-voice number built on sampled X data is directional.
- Build the baseline before you need it. Twelve months of history is the minimum that distinguishes a genuine sentiment shift from a seasonal pattern. If the tool only backfills 30 days automatically, budget for the paid backfill or start exporting on day one.
- Audit the sentiment model on your own mentions. Export 200 to 300 mentions, hand-label them, and compare. Half a day of work, and the only accuracy measurement that actually matters to you.
- Set a human review gate on high-stakes classifications. Anything triggering a crisis workflow, a legal escalation or a public response should be read by a person first. Treat model output as a ranked queue, not a verdict.
- Record your lawful basis in writing. Document the interest relied on, the safeguards applied, the retention period and how you handle deletion requests, before you switch the tool on. Cheapest risk reduction available.
- Review coverage every six months. Brandwatch rebuilt its entire Reddit feed on 28 August 2026 and deleted all legacy Reddit history on 30 September 2026. A contract is not a guarantee of permanence.
What breaks: failure modes and gotchas
The silent coverage gap. The most common failure is not a bad dashboard, it is a missing source nobody told you about. Sprout Social’s documentation states that Facebook listening is limited to Facebook Pages, that articles, blogs and forum posts are capped at 10 KB, that YouTube and Tumblr limit incoming posts because of rate limits and API limitations, and that gender, age and location data are not available at message level in Listening. Brand24’s price list says nothing about what proportion of any source it captures. Ask, in writing, and save the answer.
Query construction limits you did not know existed. Sprout Social caps listening queries at 2,037 characters across all plan levels, including the top tier, and does not support filtering to a specific geographic area within a query. Long boolean keyword lists and location-scoped monitoring are simply not available there.
Latency destroys crisis response. Brandwatch’s August 2026 Reddit notice states that roughly 89 percent of content arrives within three hours, 81 percent within two hours and 55 percent within one hour, with some content taking up to 30 hours. If your crisis playbook assumes same-hour detection on Reddit, that assumption is wrong, for reasons the vendor discloses in advance.
Backfill changes silently destroy baselines. When Brandwatch moved to its new Reddit feed on 28 August 2026, it reset history to 13 months and made all legacy Reddit history unavailable from 30 September 2026. Any year-over-year comparison you had built on the old feed is gone. Snapshot your dashboards and export your data before any vendor migration.
Sentiment gets worse, not better, at scale. The adversarial research cited earlier shows what happens when models meet negation, sarcasm and code-mixing. Large mention volumes do not average out those errors; they multiply them.
Competitor comparison is asymmetric. Share of voice assumes you can see the competitor’s conversation. If a competitor is strong on a network you do not cover, your dashboard will understate them and the gap looks like your own momentum. Always state your covered networks alongside any share-of-voice number.
Dark social is invisible to everyone. Private groups, end-to-end encrypted messaging and email threads are outside the reach of every platform on this list. Meltwater’s own coverage guidance acknowledges that channels such as private messages or encrypted conversations remain largely unmeasurable. Treat share of voice as share of voice in public conversation, and say so aloud whenever you present it.
The legal and ethical line: scraping, GDPR and platform terms
This section is not legal advice. Talk to a qualified lawyer in your jurisdiction before running a monitoring programme that touches identifiable individuals.
Platform terms are the sharpest risk. X’s terms of service, effective 9 October 2026, prohibit accessing or searching the service by any automated means other than published interfaces unless specifically allowed in a separate agreement, and state that crawling or scraping “in any form, for any purpose without our prior written consent is expressly prohibited.” The same terms set liquidated damages of $15,000 per million posts accessed in a 24-hour period. That converts the old legal risk of scraping into a priced one, large enough to end most side projects. X’s developer agreement adds restrictions: no deriving non-public information about individual users, no monitoring sensitive events, and no targeting or profiling individuals on sensitive personal attributes including health status such as pregnancy, negative financial status, political affiliation, racial or ethnic origin, religious belief, sex life or sexual orientation, trade union membership, and alleged criminal conduct. Building a watchlist of individual journalists from public posts is exactly what that clause targets.
A handle is personal data. The EDPB’s Guidelines 8/2020 on the targeting of social media users, adopted 2 September 2020 and published in a version for public consultation, state that social media users may be considered data subjects “insofar as the individual is directly or indirectly identified or identifiable,” and that using a nickname or pseudonym rather than a real name does not change that. That guidance is nearly six years old, so treat it as the regulator’s stated position rather than current binding law, but its core reasoning has not been overturned. A social listening dashboard that stores handles, follower counts and posting histories is processing personal data.
Legitimate interest is the usual basis, and it is conditional. The EDPB’s Guidelines 1/2024 on processing personal data based on Article 6(1)(f) GDPR, adopted 8 October 2024, set out three cumulative conditions: a legitimate interest must be pursued, processing must be necessary for that interest, and the data subject’s interests or fundamental rights must not override it. Reasonable expectations of the individual are an explicit factor in the balancing. The French authority CNIL published a focus sheet on 5 January 2026 stating that scraping publicly accessible data is generally based on legitimate interest but must be assessed case by case, warning about the volume of data collected, the difficulty of exercising erasure rights, and the risk of a chilling effect on freedom of expression. CNIL also notes that scraping can be prohibited by rules other than the GDPR, including terms of use grounded in database producer rights or copyright.
Enforcement is real. On 5 December 2024 the CNIL fined KASPR 240,000 euros for building a database of roughly 160 million professional contacts from LinkedIn and other sources. The restricted committee found breaches of Article 6 for lacking a legal basis where users had restricted contact visibility, Article 5(1)(e) for disproportionate five-year auto-renewing retention, Articles 12 and 14 for late and incomprehensible transparency notices, and Article 15 for failing to explain data sources. The order included deleting unlawfully collected data. If you are building the same kind of profile database from listening data, this is the shape of the problem you are walking into.
The defensive posture is simple. Buy licensed access rather than scraping. Document your lawful interest assessment. Set and enforce a retention period that matches your purpose. Build in a route to honour access and erasure requests, and check your vendor supports it, since listening tools store handles and author metadata alongside mention text. Restrict profiling to aggregate brand and topic analysis rather than individual targeting. Verify deletion on exit: X’s developer agreement requires deletion of licensed material within 24 hours of a written request and evidence of that deletion within ten business days.
The bottom line
AI social listening is a data-access business wearing an AI label. The modelling is commoditised; the rights are not. Nine vendors remain, four publish prices, and the cheapest credible enterprise floor is Meltwater’s $10,000 per year. Everything below that gets you a real product with real limits, and those limits are documented by the vendors themselves rather than hidden from you, which is to their credit.
If you are a small brand, start at Brand24 Pro at $399 per month annually, or BuzzSumo Suite at $499 if your need is PR and content rather than broad listening, and treat the output as directional. If you are a mid-market team, Sprout Social Standard or Professional with the listening add-on gives you publishing, benchmarking and analysis in one place, and its support documentation is unusually candid about what you will not see. If you have genuine crisis-detection requirements across Reddit and X, budget for enterprise and read the coverage documentation before you sign, because latency and historical depth are contractual realities.
What I would not do is scrape, or buy a tool on the strength of a sentiment accuracy percentage the vendor measured about itself.
FAQ
What is the cheapest AI social listening platform for a small business?
Brand24 Individual at $199 per month on annual billing is the lowest-cost credible option, with unlimited users, a 14-day free trial and no card required. Sprout Social Essentials is cheaper at $79 per seat per month but caps you at five social profiles and excludes listening entirely, because listening is a paid add-on on Standard and above. BuzzSumo starts at $199 but is weighted to content and PR discovery. Realistic budget for genuine cross-platform listening: $199 to $499 per month.
Which AI social listening platform has the widest data coverage?
It depends on the definition, and every vendor’s number is a supplier claim. Meltwater states more than 350,000 news sources, blogs and social platforms, 240+ languages, 6M+ unique sources and 1.3B+ documents ingested daily. Talkwalker’s coverage page lists 30+ social networks, 300+ review sites and 150+ million websites. Brandwatch states half a billion posts from over 100 million sources pass through it daily. None of these disclose what percentage of each platform’s public volume they ingest. Ask before you buy.
Can a small brand legally monitor X, Reddit and TikTok for free?
No. X’s terms of service effective 9 October 2026 prohibit scraping without prior written consent and set $15,000 in liquidated damages per million posts. Reddit’s Data API Terms require a separate agreement for commercial use, and Reddit publishes no price list. TikTok’s Research API is restricted to academic and nonprofit researchers working on a non-commercial basis, so a brand cannot apply. The free-access networks, X and Bluesky, are exactly the ones whose commercial access is priced.
How accurate is AI sentiment analysis on social media?
There is no independent published benchmark for commercial listening engines that I could find in the last 12 months; every accuracy figure in circulation is vendor-reported. The best independent evidence is Rahul and Adebayo (2026, ACL DravidianLangTech workshop, DOI 10.18653/v1/2026.dravidianlangtech-1.2), which reports XLM-R accuracy falling from 99.7 percent in-distribution to 42.5 percent on an adversarial set covering negation, sarcasm, contrast and true neutrality. The practical answer is to audit the tool against your own hand-labelled mentions.
What is the difference between social monitoring and social listening?
Monitoring tells you what is happening now and gives you messages to action individually. Listening analyses the whole body of conversation to explain why. Commercially this matters because monitoring is usually bundled into publishing plans while listening is gated: Sprout Social states Premium Analytics and Listening are each available as add-ons to Standard and above, and Hootsuite places Instagram and Reddit listening behind Enterprise.
Is GDPR risk a real concern for social listening?
Yes. A public username is personal data where the person is directly or indirectly identifiable, per EDPB Guidelines 8/2020. CNIL’s 5 January 2026 focus sheet says scraping of public data is not prohibited per se but must be assessed case by case and carries a chilling-effect risk for freedom of expression. CNIL fined KASPR 240,000 euros on 5 December 2024 for collecting LinkedIn contact details without a valid legal basis. Take professional legal advice.
Do I need an enterprise contract to monitor Instagram and Reddit?
On Hootsuite, yes. Its documentation states Instagram, Threads, LinkedIn and Reddit sit behind Lumen advanced listening on Enterprise only, while all plans cover Facebook, X, Bluesky, YouTube and Quora plus podcasts, news, newsletters, blogs and forums. Elsewhere, Brandwatch covers about 720,000 non-owned Instagram accounts, posts only, from February 2025. Sprout Social’s public Instagram source covers hashtag posts in captions only, excludes comments, and offers no backfill.
Which platforms have been acquired or shut down?
Hootsuite acquired Talkwalker in April 2024, and Talkwalker technology now powers Lumen inside Hootsuite. Metryka is dead, its domain parked and listed for sale at $9,995. Qualtrics Social Intelligence has no product page and has been folded into Qualtrics Omnichannel Experience Management. NetBase Quid now redirects to quid.com and trades simply as Quid. Always verify a vendor is live before you build a report on it.
What does X API data actually cost?
X sells API access as prepaid credits with no subscription. Reading a post costs $0.005 per resource and reading a user $0.010, deduplicated within a 24-hour UTC window, so 10,000 posts a day is about $1,500 a month and a million posts is $5,000. Owned reads of your own posts and followers are $0.001 per resource. New pay-per-use accounts can earn up to $70 in free credits, which expire three months after they are granted.
What is the cheapest way to monitor a small brand’s competitors?
Brand24 Pro at $399 per month on annual billing covers 12 keywords and 40,000 mentions per month with unlimited users. Hootsuite Professional at $199 per user per month includes listening with up to 20 competitors benchmarked, which is cheaper if competitive benchmarking is the whole job. Neither gives you full-feed X, Reddit or Instagram coverage. For competitor-only work, start there before committing to an enterprise contract.
Sources
- X API pay-per-usage pricing and credits: per-resource credit costs, deduplication window, Owned Reads pricing
- X API changelog and release notes: launch of pay-per-use pricing, Basic and Pro remaining available, removal of follows/likes/quote-posts from self-serve tiers, free credit incentives
- X Developers on X: launch announcement: dated 6 February 2026 launch of X API pay-per-use
- X Terms of Service: effective 9 October 2026; scraping prohibition and $15,000 per million posts liquidated damages
- X Developer Agreement: last updated 27 April 2026; self-serve tier scope limits, surveillance and sensitive-attribute restrictions
- Reddit Data API Terms: commercial-use agreement requirement, AI training prohibition, retention limits
- Reddit Data API Wiki: commercial use definitions, rate limits, LLM training prohibition
- Key Facts to Understanding Reddit’s Recent API Updates: free access for non-commercial researchers; last updated 30 June 2023, so over three years old
- TikTok Research API: eligibility restricted to academic and nonprofit non-commercial research
- YouTube Data API overview and quota usage: default 10,000 units/day, 100 search.list calls/day; page last updated 15 October 2026
- Brandwatch: Overview of Consumer Intelligence: half a billion posts daily from 100M+ sources; source coverage
- Brandwatch: Sources for Listen Mentions: 720,000 non-owned Instagram accounts, 200,000 Facebook pages, Bluesky firehose volume, review site list
- Brandwatch: Changes to Reddit coverage, August 2026: 400M posts/month, latency distribution, 13-month history, legacy deletion date
- Brandwatch: Language support: Compact Language Detector, 83 languages
- Brandwatch pricing and plans: quote-only structure across Consumer Intelligence, SMM and Influencer Marketing
- Sprout Social pricing: per-seat tiers, listening as paid add-on, brand keyword limits
- Sprout Social: Social Listening data availability and limitations: per-source backfill and update schedules, sampled X data, 2,037-character query cap, 10 KB article cap, paid historical backfill
- Sprout Social: How does Sprout determine sentiment?: DNN method, unsupported-language and unclassified bucket behaviour
- Sprout Social: Social listening: listening source list, Trellis AI agent, NewsWhip integration
- Hootsuite plans and pricing: Standard, Professional, Advanced and Enterprise tiers; benchmark and search limits by plan
- Hootsuite: Social media monitoring and social listening: per-plan source coverage, Lumen enterprise sources, Talkwalker acquisition April 2024
- Lumen by Hootsuite: rebrand of Hootsuite Listening and Talkwalker technology, MCP access
- Talkwalker data coverage: 30+ social networks, 300+ review sites, 150+ million websites
- Meltwater pricing and costs: indicative annual ranges, annual contract terms, influencer marketing Enterprise-only
- Meltwater: The hidden cost of incomplete social listening data: coverage figures and historical depth (vendor claims)
- Brand24 pricing and plans: plan prices, mention and keyword limits, API fee, trial and refund terms
- BuzzSumo pricing: plan prices, alert and topic limits, media mention allowances
- Emplifi pricing: Social Marketing package tiers; page published 1 April 2026
- Emplifi: social listening: product scope and AI features
- Sprinklr product suites: active product portfolio including Social Listening and LLM Insights
- Quid social listening: rebranded from NetBase Quid; licensed-source positioning
- Qualtrics: New CX capabilities at X4 2026: social listening folded into Omnichannel Experience Management
- Qualtrics homepage product navigation: absence of a standalone Social Intelligence product
- CNIL: Legal basis of legitimate interest, web scraping focus sheet: published 5 January 2026
- CNIL: KASPR fined 240,000 euros: 5 December 2024 decision and the breaches found
- EDPB Guidelines 1/2024 on Article 6(1)(f) GDPR: three cumulative conditions; adopted 8 October 2024
- EDPB Guidelines 8/2020 on the targeting of social media users: social media users as data subjects; adopted 2 September 2020
- Rahul & Adebayo (2026), Beyond Benchmark Accuracy: Robustness Evaluation of Hinglish Sentiment Models, ACL DravidianLangTech 2026: 99.7% in-distribution versus 42.5% adversarial accuracy
- A review on sentiment analysis from social media platforms, Expert Systems with Applications: method taxonomy and known weaknesses; published 2023, so nearly four years old
