X is among the most permissive social networks

1. Pass Title matches the evidence

   Title says X is among the most permissive social networks, and evidence item 1 shows X and Rumble restrict none of six contested topics while the other major sites restrict more, which matches.

2. Pass No obvious contradictions

   True rating agrees with the conclusion that the claim holds on written rules, and the caveat that X still removes posts is stated consistently in key points, evidence and objections.

3. Pass Plausible justification

   Conclusion gives a clear reason for True at moderate confidence: direct comparison of written rules across nine platforms, with the enforcement gap acknowledged.

4. Pass Sources are relevant

   Platform policy pages, X and Twitter transparency reports, Pew survey data, news on Musk-era policy changes and peer-reviewed research are all on topic for the viewpoint-permissiveness claim.

5. Pass Common questions answered

   Seven common questions (only site, smaller sites, Musk, government pressure, legal orders, bias, slurs) each receive a response that addresses the question.

Coverage: Read the full article payload from the saved tool output (the inline result was truncated before the elements list, but the payload's sources, evidence, objections and conclusion were all read); sources judged by name/title/description/role only, per First Look's rules; no source snapshots read. startedAtUtc equals the submission time because the article was fetched earlier and the start time was not recorded then.

1. Concern (minor) Central claims are accurate

   Fix obj-bias: the cited Nature study reports 19.6% vs 4.5% suspended (Trump vs Biden hashtag users), not 35.6% vs 7.7%; also align Singal petition and Truth Social ban details with sources.

   Main claims: on six contested topics X's written rules restrict none while YouTube, Meta, TikTok and Reddit restrict some; Musk's changes produced this; X still removes content and acts on most government demands. Ratings. (1) X restricts none of six topics: True. x-abusive says it reduces visibility of misgendering only "Where required by local laws" and "we will not take action on such critical commentary"; x-hateful has no misinformation or immigration-viewpoint rule. (2) Meta three, explicitly allows mental-illness allegations based on gender identity: True; meta-hateful: "We do allow allegations of mental illness or abnormality when based on gender or sexual orientation". (3) TikTok six, YouTube two, Reddit one: Mostly True, resting on the dataset; tiktok-sc archive is empty. (4) Pew shares (84/71/50/37/26/21, Threads 8, Bluesky 4, Truth Social 3): True per pew-factsheet. (5) Musk changes (Trump reinstated after poll, COVID policy not enforced from Nov 23 2022, 11,000+ suspensions, amnesty, misgendering rule removed April 2023): True per breakingnews-trump, denver7-covid, tc-amnesty, news4jax-misgender. (6) X action rates 70.82%/81.89%, Twitter 29-54%: True per ds-takedowns (PDF unreadable). Rest of World 83%, no refusals: True. (7) Journalist suspensions, Mastodon links, Brazil: Mostly True; npr-elonjet does not say accounts were reinstated within days. (8) Rumble France, Rumble "sole, unfettered, and arbitrary discretion": True. (9) Meta 7.4M to 3.4M hateful-conduct removals: True per emarketer-meta (Q1 2024 vs Q1 2025). Errors found, all peripheral: obj-bias says Republicans were suspended "35.6 percent vs 7.7 percent"; the Nature text says "only 4.5% of the users who shared Biden hashtags had been suspended as of July 2021, 19.6% of the users who shared Trump hashtags" and neither 35.6 nor 7.7 appears (False as stated; direction correct). ev-user-moderation says the Singal petition had "more than 28,000 signatures" and he was "suspended twice"; the cited archive says "over 18,000 signatures" and that Bluesky banned him, reinstated him, then labeled him (Mixed). ev-alternatives says the June 2023 Truth Social bans "hit users who criticized Trump"; newsweek-truth-social reports bans across "conservative, liberal, it doesn't matter" (Mostly True but selective). ev-musk-changes says Dorsey "acknowledged that blocking the links was wrong"; cnbc-ny-post says he called blocking "without explaining its reasoning" unacceptable (Mostly True). None changes the core message. Confidence: High.

   About conclusion, source:nature-2024, evidence:ev-user-moderation, source:tc-singal, evidence:ev-alternatives, source:newsweek-truth-social, evidence:ev-musk-changes, source:cnbc-ny-post The finding

2. Concern (minor) Sourcing is strong and transparent

   Cite Zuckerberg's and Alphabet's letters and Meta's Q1 2025 integrity report directly, and list Truth Social's Community Guidelines as the source of the quoted viewpoint pledge.

   Central sourcing is strong and primary: platform policy pages quoted verbatim, X's and Twitter's own transparency figures, Pew, SEC filings in the size dataset, and a peer-reviewed Nature study. Advocacy and secondhand sources are labeled (Public Citizen "progressive", PiunikaWeb "secondhand account"). Gaps: ev-shift relies on news reports of letters to Congress (CTV/AP, TechCrunch) and a trade-press summary (eMarketer) where the primary documents (the letters and Meta's integrity report) were readily available; the archived CTV copy contains only the headline, so the claim that Zuckerberg said the pressure "was wrong" cannot be checked against the cited copy. ev-alternatives quotes Truth Social's guidelines ("In no instance will TRUTH Social determine..."), but neither cited source (ts-tos, ts-faq) contains that sentence; it comes from the ts-guidelines page cited only inside the dataset. These affect peripheral claims. Confidence: High.

   About evidence:ev-shift, source:ctv-zuckerberg, source:emarketer-meta, source:tc-youtube, evidence:ev-alternatives, source:ts-faq The finding

3. Concern (minor) Context is complete

   Address post-2022 reports of X restricting speech on its own initiative, such as treating 'cis' and 'cisgender' as slurs in 2023, so readers can weigh practice against written rules.

   The article supplies most needed context: written rules vs enforcement, the trade-offs under Musk (journalist suspensions, Mastodon link block, Brazil, rising compliance with government demands), smaller platforms that match X, user-driven moderation elsewhere, and Meta's and YouTube's loosening. Missing: X restrictions after December 2022 that cut against the comparison on the same topics. The article does not address whether X restricts terms or speech on its own initiative outside its written policies, for example the 2023 statement that "cis" and "cisgender" are treated as slurs on X, or later reported suspensions and link throttling of accounts or sites critical of Musk. These rest partly on my own knowledge and are raised as questions, not established facts. Because the article already states that enforcement "could narrow or widen the gap" and rates on written rules, the omission does not change the core impression. Confidence: Medium.

   About conclusion, evidence:ev-x-incidents, limitation:ev-policies The finding

4. Pass Claims are supported and labeled

   Each central claim is cited: the six-topic counts to ds-policies with verbatim quotes and primary policy pages (x-abusive, x-hateful, meta-hateful, rumble-terms, ts-tos); the Musk-era changes to six dated reports; the removal-demand rates to ds-takedowns and X's own report; size to Pew and ds-size. The article keeps the measured fact (written rules) apart from inference: the answer says "enforcement was not measured", and the conclusion says the government-demand figures show "X is not censorship-free, not that it is narrower than its rivals". The claim that X's lead is "largely Elon Musk's doing" is a reasonable inference from the dated policy changes rather than an assertion. Classification judgments are disclosed ("counting vaccine and climate misinformation rules is especially debatable"). Small source mismatches are reported under claims-accurate. Confidence: High.

5. Pass Framing is fair and neutral

   The original claim is stated as it circulates and the stronger "only site that doesn't censor" version is addressed and rejected (obj-only-one). The article applies the same written-rule test to every platform and reports evidence unflattering to X and Musk (journalist suspensions, 81.89% action rate, Brazil) alongside evidence unflattering to Bluesky, Reddit and Truth Social. Objections from both directions are answered: distrust of Musk (obj-musk) and claims of anti-conservative bias (obj-bias), with the Nature finding that asymmetric sharing could explain asymmetric suspensions. The title "X is among the most permissive social networks" is measured. The selective description of Truth Social's 2023 bans is noted under claims-accurate and does not amount to a framing pattern. Confidence: High.

6. Pass Uncertainty is stated

   Confidence is Moderate and the article says why: "The evidence is written policy... and enforcement in practice was not measured and could narrow or widen the gap", and LinkedIn, Pinterest, Snapchat and WhatsApp were not reviewed. Limitations flag that classifying vaccine and climate rules as viewpoint restrictions "is especially debatable", that Twitter's and X's demand rates "are not like-for-like", and that Meta's lower removals may mean fewer mistakes or more harmful content "is disputed". The rating rests on a measure that is directly checkable, and the uncertain parts are labeled. Confidence: High.

Coverage: Read the article payload, the three cited FactHub datasets (viewpoint policies, X removal demands, platform size; saved analyses and limitations, not the CSV cells), and the snapshot of every cited web source. Read in full: x-abusive, x-hateful, meta-hateful, rumble-terms, ts-tos, ts-faq, rest-of-world, npr-elonjet, nine-to-five-mastodon, jurist-brazil, fortune-x-links, piunika-reddit-bots, tc-singal (Dec 13, 2024 archive), newsweek-bluesky, carnegie-defederation, public-citizen, newsweek-truth-social, fnac-rumble, meta-jan-2025, emarketer-meta, tc-youtube, nature-2024, pew-factsheet, cnbc-ny-post, x-blog-trump, denver7-covid, breakingnews-trump, tc-amnesty, news4jax-misgender. Not readable: x-report-2024h2 (PDF fetch timed out twice; its 2024 figures were checked against the ds-takedowns dataset instead); tiktok-sc (archive returns empty text, so TikTok's six-topic count rests on the dataset only); ctv-zuckerberg (archive returns the headline but not the article body). Policy pages for YouTube, Reddit, Bluesky and mastodon.social are cited only inside the policies dataset and were not read, so per-cell verdicts for those platforms were taken from the dataset's saved analysis.

1. Concern (minor) Rating follows from the evidence

   Consider Mostly true: the evidence shows X's written rules restrict the fewest topics, but the claim is about what X allows and enforcement was not measured.

   Before reading the finding I would have chosen Mostly true, moderate confidence. Taken as stated, ev-policies shows X's written rules restrict 0 of 6 contested topics versus Reddit 1, YouTube 2, Meta 3 and TikTok 6, and ev-musk-changes shows why. That firmly supports X being the most permissive major platform on paper. But the claim says X "allows" a wider range of views, which a reader takes to mean in practice, and the article's own answer says "enforcement was not measured". The record of X restricting outside or beyond its written rules (journalist suspensions after the @ElonJet ban, blocking Mastodon links, reach limits the conclusion mentions) is evidence that the written-rule gap may overstate the practical one, and the article cannot show the gap holds in practice. The article chose True, which is defensible; Mostly true, one step away, better reflects that the evidence covers rules rather than what is actually allowed. Borderline between adjacent labels, so Minor.

   About conclusion, evidence:ev-policies, limitation:ev-policies, evidence:ev-x-incidents The finding

2. Concern (minor) Strongest objections answered

   Add an objection on whether owner-directed and algorithmic reach limits on X (e.g. 'cisgender' treated as a slur, throttled links) narrow the views it allows in practice.

   Strongest objections I would expect: (1) X restricts views in practice through owner-directed decisions and reach limits rather than written rules, so a rules comparison flatters it; (2) the stronger "X doesn't censor" claim; (3) smaller platforms such as Rumble and Truth Social are as permissive; (4) other platforms' restrictions came from government pressure; (5) mainstream platforms suspend conservatives more. Objections 2 to 5 are answered with evidence (obj-only-one, obj-smaller-sites, obj-government-pressure, obj-bias, the last with the Nature study). Objection 1 is acknowledged as a limitation ("enforcement in practice was not measured") and partly illustrated by ev-x-incidents, but it is not posed and answered as an objection. The article does not address whether X's visibility filtering, which its own coding would count as "restricted", or decisions such as treating "cis" or "cisgender" as slurs and throttling links to rival sites, fall unevenly on particular political views. This rests partly on my own knowledge, so it is raised as a question. Because the article flags the limit and lowers confidence for it, a reader's belief is shifted only modestly; Minor.

   About conclusion, evidence:ev-x-incidents, limitation:ev-policies The finding

3. Pass Rates the real claim

   The claim as it circulates has two forms: the strong one ("X is the only site that doesn't censor") and the comparative one ("X allows a wider range of political views"). The article names and rates the comparative form, and explicitly rates the strong form in obj-only-one: "That stronger version goes too far... X still removes posts... Rumble's written rules are also as permissive". That is not a strawman: the comparative version is what most people repeating the claim mean, and the stronger version is not hidden. "Major" is defined transparently (public political forums used by at least 10% of US adults per Pew 2025), smaller sites are treated separately (obj-smaller-sites), and the conclusion stays within US platforms as of October 2026. The narrowing from what X "allows" to what its written rules allow is disclosed in the answer and is assessed under rating-follows-evidence. I would rate the same claim.

4. Pass Confidence matches the evidence

   Confidence is Moderate, which matches evidence that is direct but single-method for the core comparison (document analysis of written rules). The conclusion gives the right reasons: "enforcement in practice was not measured and could narrow or widen the gap" and four widely used platforms were not reviewed. Limitations that matter are stated: classification is a judgment and "counting vaccine and climate misinformation rules is especially debatable"; vague rules are not counted; Twitter and X demand rates "are not like-for-like"; Meta and YouTube loosened rules, so "the gap is also narrowing". The title "X is among the most permissive social networks" and the summary ("only much smaller sites such as Rumble match X") claim no more than the evidence shows; if anything the title is softer than the finding. I would give Moderate confidence too.

5. Pass Evidence is independent

   The direct comparison rests on ev-policies, a FactHub coding of each platform's own policy pages; its ten listed institutions are the platforms being coded, so it is one line of evidence, not ten. The article does not present it as more. The supporting lines are independent of it and of each other: ev-musk-changes (company announcements and news reports on dated policy changes), ev-x-reports (Twitter and X transparency reports plus Rest of World's separate Lumen analysis, flagged as a different database), ev-x-incidents and ev-user-moderation (separate news and policy-research sources), ev-alternatives, ev-shift and ev-size (Pew and company filings). No two items share an underlying dataset or study. The reliance on one coded dataset for the core comparison is reflected in Moderate confidence rather than hidden.

Coverage: Examined the article payload only: claim, finding, answer, confidence, title, summary, keyPoints, context, conclusion, all eight evidence items with methods, institutions, years and limitations, seven objections, method steps and editorial note. Evidence was taken as stated; source snapshots were not read, per the Skeptic's division of labor.

1. Concern (minor) Title states the finding

   Make the title state the finding, e.g. "X's written rules allow more political views than other major social networks'"; "among the most permissive" does not resolve the claim.

   Rule: the title must resolve the checked claim so a reader who sees only the title knows what the evidence says about it. Title: "X is among the most permissive social networks". It is positive, neutral and names the subject, and does not repeat a myth. But the claim is comparative ("X allows a wider range of political views than other major social media sites") and the verdict is True: X's rules restrict the fewest viewpoints of the major sites. "Among the most permissive" hedges this into a tie and does not say permissive about what (political views) or by what measure (written rules), so the title-only reader cannot tell whether the claim held. The seoTitle ("Does X allow more political views than other social media? Its rules restrict the fewest") does state the finding and passes. Minor rather than Material because the title points the right way.

   About conclusion The finding

2. Concern (material) Readers can verify it

   Rewrite method step 1 to define the rated comparative claim (X allows more political views than other major sites) and add a step on favoring evidence that does not rest on one source.

   Rule: method steps include a precise claim definition, prioritization of evidence that does not rest on one source, the sources reviewed, and recording of limits. Method step 1 says: "Defined the claim as two assertions: that X does not censor, and that it is the only social media site that does not." That is the earlier version of the claim; the article now rates the comparative claim "X allows a wider range of political views than other major social media sites", and the stronger version is only an objection. So the claim-definition element describes a different claim from the one rated, which a reader checking the method would notice. No step describes prioritizing evidence that does not rest on one source; step 6 limits news reports to incidents but does not address independence. Sources reviewed (steps 3 to 5) and recording of limits (steps 6 and 7) are present. One required element is wrong and one missing, so Material. Source entries otherwise meet the rule: each has a role label ("Primary: platform policy", "Advocacy-group report (progressive)", "Trade news report (secondhand account...)"), a full title, a description of why it was chosen, and key claims have primary anchors (policy pages with archive URLs, X's transparency report, the Nature DOI, Pew).

   About method:1, method:6, conclusion The finding

3. Concern (minor) Consistent house style

   Change "analysing" to "analyzing" in ev-x-reports, and use one percent format ("percent" or "%") throughout keyPoints, evidence and conclusion.

   Rule: American English spelling, and number formats consistent throughout. ev-x-reports says "Rest of World, analysing a separate database", a British spelling. Percentages are written two ways for the same figures: keyPoints use "70.82% of 72,703", "81.89% of 97,006", "51%" and "21% of US adults", while the conclusion, evidence summaries and objections use "70.82 percent", "81.89 percent" and "21 percent". Facebook's user count is also rounded differently: "about 3 billion for Facebook and Instagram" in ev-size versus "Facebook's 3.07 billion monthly users" in keyPoints. No leftover drafting text and no conflicting values for the same figure beyond rounding; names, dates and abbreviations are otherwise consistent, and terms such as defederation and Community Notes are explained.

   About evidence:ev-x-reports, conclusion The finding

4. Pass Verdict fields agree and stand alone

   Quoted alone: FactHub rates the claim "X allows a wider range of political views than other major social media sites." True (Moderate confidence): "Among the most-used US platforms, X's written rules restrict the fewest political viewpoints; enforcement was not measured." The claim is one declarative sentence worded as it circulates, not its correction, and reads naturally in quotation marks. The answer stands alone and says what is true. Summary ("X's written rules restrict none... only much smaller sites such as Rumble match X"), description, keyPoints and conclusion ("The claim holds on the measure that can be compared directly") express the same conclusion in the current label's terms; no stale "Mostly true" wording remains in these fields. The stronger "only site" version is handled separately in objections, as the editorial note says.

5. Pass Each section does its job

   Summary is one sentence naming the core evidence (the six-topic written-rule comparison). keyPoints state what the evidence establishes, including the counter-evidence ("X still censors", "The gap has narrowed"). Context is one concise paragraph of framing: scope date, the safety-vs-viewpoint distinction, and "Written rules are not the same as enforcement". The conclusion synthesizes rather than restates: it defines "major", explains why X leads (Musk's changes), places smaller platforms, weighs the contrary evidence ("The rating does not mean X allows everything"), and states what remains unproven and why confidence is Moderate. The editorial note gives scope (nine platforms, as of October 10, 2026, the definition of major, platforms not reviewed) and the material caveat that classification involves judgment. No hand-written independence note.

6. Pass Limitations are specific

   Each limitation is specific to its item. ev-policies: "Written rules only, not enforcement... counting vaccine and climate misinformation rules is especially debatable. Vague rules are not counted, which may understate Truth Social and Bluesky. mastodon.social is one server, and Reddit subreddit rules are excluded." ev-musk-changes notes the Trump suspension "cited incitement risk" and that Musk's ownership brought new restrictions. ev-x-reports explains the rates "are not like-for-like" and that court orders may be legally binding. ev-x-incidents: "Individual incidents do not show how often X restricts content." ev-user-moderation, ev-alternatives, ev-shift and ev-size each name what the item does not prove. None is boilerplate.

7. Pass Tone is measured and respectful

   Language is measured: "The claim holds on the measure that can be compared directly", "Confidence is moderate rather than high", "The gap is also narrowing". Objections are answered respectfully and with evidence. The response to distrust of Musk (obj-musk) calls those concerns "reasonable grounds for choosing where to spend time" before explaining the scope of the check; obj-bias concedes the Nature study "does not prove the rules were unbiased". "Where X stands alone" (obj-smaller-sites) is justified within the stated definition of major sites. No ridicule, sarcasm or unwarranted absolutes.

Coverage: Examined the article payload only: title, seoTitle, socialTitle, description, claim, finding, answer, confidence, summary, keyPoints, context, conclusion, eight evidence items and limitations, seven objections and responses, seven method steps, editorialNote and 34 source entries (roles, titles, descriptions, identifiers). No external pages were used. Server-enforced items listed as out of scope were not assessed.