X Tweet Viral Score Analyzer with AI | $0.0003/Tweet

English · Español · Türkçe · 简体中文 · 日本語 · 한국어 · Deutsch · Français · Italiano

Watch how Framer uses Xquik scrapers with Claude Code, Codex, Cursor, and more, from 6:07.

Xquik is the world's fastest & cheapest X (Twitter) scraper service with the most complete X data. X Tweet Viral Score Analyzer adds a Viral Score estimate & a verdict to every tweet. Every other Apify Actor charges before filtering or deduplicating. Xquik charges only for delivered, unique, filter-matching results. AI costs are included in the per-tweet price. You pay no AI provider, buy no tokens & bring no key.

Learn why tweets spread or flop & keep the original tweet data. X Tweet Viral Score Analyzer with AI collects matching tweets. The AI rates 8 traits of each post. The Actor turns those answers into a Viral Score estimate from 0 to 100 & a verdict. Every row keeps real likes, reposts, replies & quotes, so you can compare each estimate with what happened.

• Viral Score per post from fixed, published weights you can audit.
• 8 trait answers show why a post scored high or low.
• Hard stops cap posts that read as spam, ragebait or generic machine copy.
• Complete source records with every field the tweet exposes.

The Viral Score is an estimate of how well the wording works. It does not predict likes or views. It does not reproduce how X ranks posts.

How to check a tweet's viral score

1. Add search terms, profile handles, tweet URLs or tweet IDs.
2. Set maxItems & the extraction filters your task needs.
3. Describe your audience in analysis.context, or leave the default.
4. Run the Actor & open the Viral Score dataset view.

{
  "searchTerms": ["from:NASA -filter:replies -filter:retweets"],
  "maxItems": 150,
  "analysis": { "context": "Space fans & general readers." }
}

What the Actor answers

Question	Answer
Hook	0 no hook, 1 clear opening, 2 sharp opening
Clarity	0 confusing, 1 takes effort, 2 clear on first read
Informative	0 nothing new, 1 familiar point, 2 useful takeaway
Funny	0 not funny, 1 mildly amusing, 2 funny enough to share
Ragebait	Probability that the post mainly provokes outrage
AI written	Probability that the text reads like generic machine copy
Spam	Probability of spam, scam, giveaway or engagement farming
Reaction	Share, reply, like, argue or ignore

The AI-written answer judges style only. It does not establish who wrote the post.

How the Actor computes the Viral Score

The Actor scales each 0-2 score to a share from 0 to 1. Then it adds points:

Part	Points
Hook	up to 30
Clarity	up to 20
Payoff, the higher of informative & funny	up to 30
Reaction	up to 20
AI-written probability	minus up to 15

Reaction earns a share of its 20 points: share 1, reply 0.8, like 0.6, argue 0.4 & ignore 0. Hard stops then cap the score. Spam probability from 0.7 caps it at 20. Ragebait probability from 0.7 caps it at 35. AI-written probability from 0.8 caps it at 60. The Actor rounds the result to a whole number.

Verdict	Score
send_it	70 to 100
edit_first	40 to 69
sleep_on_it	0 to 39

viral.weights names the version of these rules, such as viral_lite:1. We bump it whenever a weight, stop or threshold changes. The score is null when the analysis failed, the Actor skipped it, or a default trait answer is missing. The Actor never fills a missing score with a guess.

Algorithm Score estimate

X published its ranking weights in the repository xai-org/x-algorithm, file home-mixer/params/param.rs. The Actor applies 4 of them to the public counts of each post:

Count	Weight
Like	0.5
Reply	5
Repost	1
Quote	5

viral.algorithmWeightedSum is the sum of each count times its weight. viral.algorithmScore divides that sum by views & multiplies by 1,000. A post without a view count uses followers instead. viral.algorithmBasis names the divisor, views or followers. Compare scores with the same basis only. viral.weightsVersion names the weights, such as x_algorithm_params:2026-09-18.

Limits:

• X multiplies each weight by a probability it predicts for one viewer. The Actor multiplies by observed counts. The result is an estimate, not the score X computes.
• X publishes no weight for bookmarks or views. The sum leaves both out.
• X uses more signals than these 4, such as dwell time & shares. Public data does not show them.
• The score is null when a post has no views & no follower count.
• The AI never sees these counts. It reads the text & context only.

Predicted against actual

The Actor compares each Viral Score with what happened. viral.actualEngagementRate is log10(1 + weighted sum per 1,000 followers). The log limits the effect of one very large post. The rate is null when the follower count is missing or 0.

The run summary's viral.calibration block reports:

• comparedPosts: posts with a Viral Score & an actual rate.
• rankCorrelation: a Spearman rank correlation from -1 to 1. It asks whether higher scores went with higher rates.
• calibrationScore: 100 times the correlation, floored at 0.
• overperformers & underperformers: up to 5 posts each, with tweet ID, URL, Viral Score, actual rate & gap.

gap is the standardized actual rate minus the standardized Viral Score. A post enters a list when its gap reaches 1 standard deviation.

Limits:

• Fewer than 10 compared posts give a null calibration with the reason too_few_posts. Identical scores or rates give no_variation.
• The Actor groups rates into buckets 0.1 wide to keep memory flat. Posts in one bucket count as tied, so the correlation is approximate.
• The calibration describes one run. A low score can mean the posts differ in timing, topic or audience, not that the wording estimate failed.
• Young posts have not finished collecting engagement. Compare posts of similar age.

Account report

The run summary's viral.accounts block reports each author handle:

• Post count, average Viral Score & average actual engagement rate.
• The best & worst post by Viral Score, with tweet ID & URL.
• Average Viral Score per bucket: posting hour in UTC, text length band, has media, has link & self-thread.

Text length bands are short to 80 characters, medium to 200, long to 280 & extended above. A self-thread post replies to its own author.

Limits:

• The report lists the 50 handles with the most scored posts.
• The Actor follows the first 1,000 handles of a run. untrackedPosts counts scored posts from later handles & posts without a handle.
• A bucket with few posts says little. Check posts before you compare averages.
• Buckets show what went together in this run. They do not show cause.

Leaderboard

The run summary's viral.leaderboard block ranks the handles of the account report. byViralScore ranks by average Viral Score. byActualEngagementRate ranks by average actual rate. Each list holds up to 20 handles with rank, posts & average.

Limits:

• A handle needs at least 3 scored posts to rank.
• The rate list skips handles without a follower count.
• More posts, then the handle name, break ties.
• The leaderboard covers the posts of one run, not an account's whole history.

Score a draft before you post

Paste your own text in texts. The Actor scores it & fetches nothing from X.

{
  "texts": [
    "We shipped dark mode today. Try it and tell us what breaks.",
    "5 things we learned from 1,000 support tickets."
  ],
  "analysis": { "context": "Developers who use our app." }
}

• Each text becomes 1 row with viralScore, viralVerdict & viral.stops.
• tweet.id is text:1, text:2 & so on, & tweet.type is text.
• A draft has no likes or views yet, so viral.algorithmScore stays null.
• Each analyzed text costs the same $0.0003 as an analyzed tweet.
• With texts set, the run analyzes only those texts. Run X targets separately.

Pricing

AI costs are included in the per-tweet price. You pay no AI provider, buy no tokens & bring no key.

From $0.0003 per successfully analyzed tweet, with no start fee. The price includes collection & the Viral Score. The analysis allowance is 8 questions, 8,000 bytes per question definition & 12,000 bytes of context per tweet. Extraction filters & deduplication run before analysis, so filtered-out & duplicate rows are never analyzed or charged. Failed & skipped analyses & diagnostic rows have no result charge. Apify bills platform usage separately. The Pricing tab shows it.

Input & output examples

The input above is copy-ready. Output rows look like this (abbreviated):

{
  "tweet": { "id": "2100493544842494265", "text": "...", "likeCount": 12 },
  "viral": {
    "score": 74,
    "verdict": "send_it",
    "weights": "viral_lite:1",
    "stops": [],
    "algorithmScore": 8.5,
    "algorithmBasis": "views",
    "algorithmWeightedSum": 17,
    "actualEngagementRate": 0.7202,
    "weightsVersion": "x_algorithm_params:2026-09-18"
  },
  "viralScore": 74,
  "viralVerdict": "send_it",
  "viralAlgorithmScore": 8.5,
  "viralActualEngagementRate": 0.7202,
  "analysis": {
    "status": "succeeded",
    "answers": [
      { "questionId": "hook", "type": "score", "value": 2, "confidence": 0.84 },
      { "questionId": "spam", "type": "probability", "probability": 0.03 },
      {
        "questionId": "reaction",
        "type": "choice",
        "value": "share",
        "confidence": 0.7
      }
    ]
  }
}

Each result contains tweet, analysis & viral. Answers include types, question versions & available probabilities. viral.stops lists the hard stops that capped the score. A failed or skipped analysis keeps the collected tweet with an empty answer list, a reason & a null score. Free diagnostics in the key-value store explain invalid inputs, missing results & interrupted collection. The run report separates collected rows, charged analyses & pending charges.

Run summary & flat answers

Each run writes an analysis-summary record to its key-value store & repeats it under results.analysisSummary in the run report. It counts analyzed, failed & skipped rows, sums engagement, and summarizes every question. Its viral block reports averageScore, the count of each verdict, & how many rows the Actor scored or left unscored. The same block holds calibration, accounts & leaderboard, described above. Score questions report a mean & an engagement-weighted mean. The reaction split shows how many tweets fall into each reaction, & top lists the three most engaged tweets per reaction. An empty run reports zero counts & a null average. Every row lists sourceDomains, the hostnames it links to, & cashtags such as $NVDA found in its text. With monitor.baselineDatasetId set, the summary's monitor block counts comparison statuses & lists up to 50 changed rows.

Every result row also carries viralScore, viralVerdict, viralAlgorithmScore, viralActualEngagementRate & answers, a flat map from question ID to the chosen category, score, or probability. The Viral Score dataset view & CSV or Excel exports show these columns beside the tweet, so spreadsheets need no JSON parsing. Failed & skipped rows carry an empty map.

Compare with an earlier run

Pass monitor.baselineDatasetId, the dataset ID of a completed earlier run with the same analysis settings. Every row then gains a monitor object. Its status is first_run without a baseline, new_to_baseline for tweets the earlier run did not have, & unchanged or changed for tweets it had. changes lists each trait decision that moved from previous to current. Decisions compare by category, rounded score level, or yes/no at 0.5. A decision counts as changed in three cases. The earlier category falls below 0.4 probability. A score moves at least 0.6 levels. A yes/no probability lands at least 0.1 from the threshold. Near-tie jitter between runs stays unchanged. Baselines above maxBaselineRows (default 100,000) or from different settings stop the run before collection with a diagnostic row.

Task examples

Choose from 50 public tasks. Each starts from a real English search with a bounded maxItems & the Viral Score dataset view. Some add audience context. Edit the search or context before running.

• Viral score of AI startup launch tweets
• Viral score of SaaS founder build in public posts
• Viral score of Product Hunt launch posts
• Viral score of Developer tool announcements
• Viral score of Open source release posts
• Viral score of Crypto project announcements
• Viral score of Parenting humor posts
• Viral score of Office humor posts
• Viral score of Pet photo captions
• Viral score audit of NASA posts
• Viral score audit of Duolingo posts
• Viral score audit of Wendy's posts

The remaining tasks cover more topics & brand accounts on the Actor page.

FAQ & support

Does a high score mean a tweet will go viral?

No. The score estimates how well the wording works for a general reader. Timing, audience size, media & luck also decide reach. Compare scores with the real engagement counts on each row before you rely on them.

Can I use my own questions?

Yes. Custom analysis.questions replace the defaults: 1-8 choice, score or probability questions with 2-255 categories or at least 2 ordered levels. The Viral Score needs all 8 default questions, so custom questions leave it null.

Why did a row come back with analysis.status of failed or skipped?

The Actor collected & delivered the tweet, but the AI analysis did not complete. analysis.reason names the cause, such as context_limit when the tweet & its context exceed maxContextBytes, or service_unavailable after retries. These rows carry no result charge & no score. Raise maxContextBytes (up to 12,000) or rerun the affected IDs.

Does the analysis verify facts?

No. Answers describe what the post expresses & how the post frames it. Probabilities express model confidence, not truth. Review important classifications against the original tweet, which every row keeps.

Which languages work?

Extraction supports every language X serves. We validate analysis on English customer scenarios first. Other supported languages return answers with the same structure.

How do I limit cost?

Filters, deduplication & maxItems run before analysis, so the Actor analyzes & charges only unique, filter-matching tweets. Use precise search operators, date bounds & engagement floors, & start with a small maxItems to check answer quality before a large run.

Where do I get help?

Open an issue on the Actor page or contact support@xquik.com with the run ID. Free diagnostics in the key-value store explain empty, partial or interrupted runs.

Xquik is an independent third-party service. Not affiliated with X Corp. "Twitter" and "X" are trademarks of X Corp.

Related Xquik Actors

Every Xquik Actor shares the same extraction engine, filter-first billing & diagnostics. Pick the one that matches the data you need.

• X Tweet Scraper: Scrapes tweets from searches, profile timelines, Lists & tweet IDs with 50+ filters & flat exports. Use it when you need tweet data without analysis. From $0.00015 per row.
• X Profile Scraper: Scrapes profiles plus their posts, replies, media & followers from handles, IDs or URLs. Use it when you start from accounts rather than searches. From $0.00015 per row.
• X Reply Scraper: Scrapes replies, comments & whole conversations under posts with 25+ filters. Use it when you need the discussion beneath tweets. From $0.00015 per row.
• X Engagement Scraper: Scrapes replies, quotes, retweeters & threads for post URLs or IDs in bulk. Use it when you measure who engaged with posts. From $0.00015 per row.
• X Follower Scraper: Scrapes followers, following, List members, subscribers & Community members as profile rows. Use it when you need audience or member lists. From $0.00015 per profile.
• X User Search Scraper: Searches users by handle, bio & location with follower, verification, age & location filters. Use it when you build account lists from search. From $0.00015 per profile.
• X List Scraper: Scrapes List posts, members & followers from List URLs or IDs. Use it when a curated List defines your sources. From $0.00015 per row.
• X Community Scraper: Scrapes Community info, posts, searches, members & moderators. Use it when your sources are X Communities. From $0.00015 per row.
• X Trends Scraper: Scrapes real-time trends by location with rank, volume, query & WOEID. Use it when you track what is trending where. From $0.00015 per trend.
• X Article Scraper: Scrapes long-form X Articles as Markdown & text with covers, authors, dates & metrics. Use it when you need article bodies, not tweets. From $0.00015 per article.
• X Media Downloader: Extracts or stores photos, videos & GIFs from posts or profiles with MP4 & metadata options. Use it when you need the media files themselves. From $0.00015 per media row.
• X (Twitter) Brand Monitoring with AI Analysis: Tracks brand mentions with AI relevance, sentiment & customer-experience answers & compares runs. Use it when you watch a brand over time. From $0.0003 per analyzed tweet.
• X Tweet Sentiment Analysis with AI: Labels attitude, intensity & sarcasm probability for every tweet with AI. Use it when you need general sentiment on any topic. From $0.0003 per analyzed tweet.
• X (Twitter) Stock & Crypto AI Trading Signals: Labels bullish, bearish, neutral or mixed stance, content type, conviction & asset relevance with AI. Use it when you follow stocks, crypto or trading talk. From $0.0003 per analyzed tweet.
• X (Twitter) News Monitor with AI Analysis: Labels news posts by format, source attribution & topic relevance with AI. Use it when you separate reporting from commentary. From $0.0003 per analyzed tweet.
• X Tweet Classifier with AI Analysis: Answers your own category, score & yes/no questions for every tweet with AI. Use it when the preset analyses do not fit your labels. From $0.0003 per analyzed tweet.