On this page

Outlet assessment

The Guardian

85/ 100

Reliability

Assessment confidence
medium
Check
full assessment · final
Assessed
2026-11-16
Reevaluation
due after 15 May 2027 (180 days; only if the outlet is still in use. How)
Kind
news, GB
Leaning
center left; climate-action framing

How scores are set: outlet assessment.

British national news publisher, online at theguardian.com with UK, US, Australian and European editions; owned through Guardian Media Group by The Scott Trust Limited.

Summary#

Reliability 85 (70-89 band). In a random 72-item sample (24 months to November 2026), 57 of 59 checkable main claims were accurate or mostly accurate (96.6%, Wilson 88.5-99.1 among checkable items), with 0 false items. The score sits below 90 because errors are not rare at the material level: 9 graded claims carry errors (3 minor, 6 material, 0 central), and both misleading items are research findings whose causal or translational reach the article overstated (2 of 4 research findings accurate). Event reports (37/37) and official positions (8/8) were all accurate; a revealed COP29 count was overstated about twofold. Orientation: center left on UK politics in editorial and opinion selection, with climate-action framing; both low confidence. Assessment confidence is medium: 59 checkable items, but the tool’s uncertainty interval (30.8 points, including frame-weight uncertainty and missing-outcome sensitivity) stays above the 25-point target at the 72-item ceiling; 8 of 8 wording comparisons are complete; blind second grading agreed on 17 of 18 items. Status final.

Ownership#

The Scott Trust Limited is the sole shareholder of Guardian Media Group plc, which publishes Guardian journalism, and of Scott Trust Endowment Limited, the endowment that funds it; the consolidated accounts report 100% ownership and voting in Guardian News & Media. The intermediate immediate-shareholder links were not each verified. The Trust has no beneficial owner: five directors hold its shares non-beneficially and the accounts report no ultimate controlling party. Its stated core purpose is “to secure the financial position and editorial independence of the Guardian in perpetuity”; the editor-in-chief, Katharine Viner, sits on the Trust board, and the Trust appoints the global readers’ editor. Chair Ole Jacob Sunde has long ties to Schibsted, the Tinius Trust and Formue. The Observer was sold to Tortoise Media on 22 April 2025; the group kept an 11.2% stake and a board seat in Tortoise. Editorial independence: unknown, because the method’s test (shared content or direction across sister outlets) has nothing to compare (the Guardian editions are one outlet), and mission statements are not evidence of non-interference. Details: The Scott Trust Limited.

Accuracy#

  • Overall: 57 of 59 checkable main claims accurate or mostly accurate (50 accurate, 7 mostly accurate, 2 misleading, 0 false); 96.6%, Wilson 88.5-99.1. Reweighted to the frame’s estimated own-interest share (9.3%) the rate is 96.3% (approximate stratified envelope 77.6-99.4). These are rates among checkable claims, not a probability that any Guardian item is true.
  • Own-interest items: 13 confirmed (media competitors such as the BBC and ABC Australia, ad and AI platforms such as Meta, TikTok and OpenAI, the Guardian’s own awards and investigations, the Observer). 10 were checkable, all accurate or mostly accurate (72.2-100); 76.9% checkable. Other items: 47 of 49 (95.9%, 86.3-98.9). Gap -4.1 points (own-interest items scored slightly higher); with 10 checkable items the difference is not meaningful.
  • By type: event reports 37/37; official positions 8/8; statistics 6/6 (one minor and two material errors, all mostly accurate); quotes 3/3; research findings 2 of 4 (both misleading). By subject (8+ checkable): business 13 of 14, world 8 of 8.
  • Errors by severity: 3 minor (a climate share given as “atmospheric” CO2, an OBR scenario presented as a £20bn hole, a “record” British Gas settlement the regulator does not call a record), 6 material: the COP29 investigation’s “at least 132” oil and gas executives and staff with host-country badges, where the UN provisional list (1,880 host guests) gives 34 at oil and gas companies, 47 with oilfield services and industry bodies and 72 with ACWA Power, although every named delegation (Aramco 10, ACWA 25, BP 8, ExxonMobil 4) matched exactly; a study of home working presented as showing it “has failed to level up” the economy, when the study is underpowered on geography; a brachyury structural study presented as pointing to a near-term chordoma drug, although the binders were not functional inhibitors; Gordon Brown’s op-ed calling the 50%-of-median line the official absolute-poverty baseline; a column generalizing a non-representative poll to national publics; and a headline calling 13,460 mostly voluntary returns “deportations” (the body says most were voluntary). No central errors.
  • 13 substantive not-checkable items (opinion, country diaries, first-person accounts, anonymous-source causal accounts, forecasts). The two items first blocked were read on retry: the COP29 list through an archived copy and Mel Stride’s Sky News remark through Politico’s report of it (accurate).
  • Corrections seen in the sample: none.
  • Blind second grading: 18 items, agreement 17 of 18 (94.4%). The disagreement (British Gas “record” settlement) went to a third blind grader; the majority grade is mostly accurate with a minor error.

Outside record#

  • Noel Clarke v Guardian News & Media (High Court, 22 August 2025, [2025] EWHC 2193 (KB)): the libel claim was dismissed; the Guardian established truth and public-interest defenses for all eight articles, though one allegation was not established (court press summary, which is not the judgment).
  • Manafort–Assange report (November 2018): the Guardian reported that Paul Manafort held secret talks with Julian Assange in the Ecuadorian embassy. Manafort and WikiLeaks denied it; CJR noted that no other outlet could confirm it, and FAIR documented later edits adding “sources say” to the headline. No court or regulator ruled on it; it remains unconfirmed.
  • Corrections policy: “GNM policy is to correct substantial errors as soon as possible”; an independent global readers’ editor (Elisabeth Ribbans since January 2020), appointed by the Scott Trust, handles complaints.
  • Wikipedia perennial sources (editor consensus from 21 discussions): “generally reliable”; op-eds to be handled as opinion; “Some editors believe The Guardian is biased or opinionated for politics.” One reference, not a verdict.

Orientation#

  • Left-right (GB), center left, moderate, low confidence. The sample’s editorials and selected opinion pieces argue from center-left positions: lower VAT on public charge points and incentives for EV buyers (“a recharge is needed”), a climate-finance deal in which rich countries pay for historic emissions, the Observer view that barring refugees who arrived irregularly from citizenship is “cruel and misjudged”, and a Gordon Brown op-ed demanding budget action on child poverty. News headlines are mostly neutral; a counter-example is the deportations story, which adopted the government’s tougher framing. Low confidence: about 19 political or opinion items and no wording comparison of a UK political event with measured comparison outlets.
  • Climate and fossil-fuel industry, moderate, low confidence: news items label Heartland a “US climate denial group” (Heartland calls itself skeptical) and present COP29 host badges as “red carpet” treatment for fossil-fuel bosses; editorials advocate climate action. Three to four items only.

Wording#

Eight complete comparisons from 20 randomly drawn events (12 had no coverage meeting the bucket and orientation rules); most other-outlet texts are headline-level. Pattern: Guardian headlines tend to state the strongest version of a development, while the body carries the qualifier. “TikTok breached EU advertising transparency laws” (body: “preliminary verdict”; LexisNexis and Engadget lead with “preliminary”); “UK joins European deal to send rejected asylum seekers to third-country hubs” (BBC, Reuters and Politico Europe describe a non-binding political declaration; the Telegraph’s “backs Rwanda-style migrant camps” is the other strong framing); and a “brutal” BBC cost-saving drive where the Independent, Straits Times, BBC and Reuters say “savings” or “cost-cutting”. On Dela Rosa, “evading arrest” and “slipped out” match Philstar, between Al Jazeera’s “flees” and Reuters’ quoted “escape”. Starmer’s final PMQs (“emotional”, “defends his record”) matches the tribute framing of the BBC, AP and Al Jazeera; only the Telegraph is critical. Culture, science and obituary items (Munch, gulls, Tony Slattery) differ only in emphasis. Orientation evidence in the political comparisons is thin (BBC impartiality pledge, Al Jazeera profile, ENNHRI’s mission), so the comparisons do not test the UK left-right reading.

Perception#

Nine dated comments from three platforms (2025). Hacker News, 7 comments from one discussion (March-April 2025): views range from “a truly great paper with excellent writers” and “its reporting is generally factual, even if they make mistakes sometimes” to “often outrage-bait or inaccurate information” and “I stopped reading the Guardian after one too many straight up lies”; two describe it as “an establishment, social liberal paper” or criticize typos without a trust judgment. Free Republic, 1 comment (March 2025): “I trust this deep state hack and The Guardian about as much as l would trust Schiff”. Bluesky, 1 post (August 2025): “ONLY The Guardian has provided solid, unbiased climate reporting” (a self-described supporter).

By audience (platform profiles, 16 November 2026): center left audiences, 7 posts (Hacker News 7): 1 trusts, 1 mostly trusts, 2 mixed, 1 mostly distrusts, 2 not stated; mean trust 60, gap -25 points against measured reliability 85. Right audiences, 1 post (Free Republic): distrusts; mean 0, gap -85. Left audiences, 1 post (Bluesky): trusts; mean 100, gap +15. Label: divergent by audience (trust from the left, distrust from the right), but the right and left views rest on one post each and the Hacker News comments come from one thread, so this is a weak signal, not a measured audience split. Two audience sides are covered, so perception is medium. X, Reddit and YouTube comments were not inspected; Lemmy and Free Republic material outside the 24-month window was excluded. Platform pages: Hacker News, Free Republic, Bluesky.

How it is weighed#

Rely on the Guardian for event reports, official statements, court and parliamentary records, and published figures, which were accurate throughout the sample. Take more care with reports of studies and science breakthroughs, where both misleading grades fell (check the study’s own limits), with counts from the Guardian’s own analyses of datasets (the COP29 figure did not reproduce), with category words in headlines (deportations), with strong verbs in headlines whose bodies qualify them (“breached” for a preliminary finding, “deal” for a non-binding declaration), and with opinion pieces and editorials, which carry the center-left and climate-action positions. Claims touching its own interests (the BBC and other publishers, ad and AI platforms, the Observer and Tortoise, press regulation) were accurate in 10 of 10 checkable items, but that stratum is small. Treat the UK, US, Australian and European editions as one source.

Evidence sufficiency and missingness#

  • Frame: 753 items from monthly 15th-day section archives (11 sections, November 2024 to November 2026), not a census. 193 of 275 first archive requests were rate-limited; recovered months are uneven, so 40 of 72 drawn items fall between November 2024 and March 2025 and 2026 is under-represented. Results describe that frame best.
  • No technical gaps remain: both blocked items were read on retry (an archived UN spreadsheet; a second outlet’s report of the Sky broadcast, which itself stays unreadable). The COP29 count rests on the grader’s classification of 997 home organizations; the Guardian’s own classification is unpublished.
  • Two sampled items describe the same 15 June 2026 market move, so the items are not fully independent.
  • Main-claim grades checked authenticity, attribution and inference separately; for study items the study’s design and limits were read (the two misleading grades rest on the studies’ own stated limits).
  • Only one grade rests solely on unprofiled references.

Used in posts#