Guide

ZeroGPT in 2026: How Accurate It Is, What It Gets Wrong, and Who Runs It

A plain-English 2026 guide to ZeroGPT: what the 98.4% accuracy claim is measured on, what peer-reviewed testing found instead, why ZeroGPT is not GPTZero and neither is zerogpt.net, what its own terms of use disclaim, who operates the site, why its image detector is licensed from TruthScan, and what the score should and should not be used for.

September 27, 2026 · The AI Rankings

Quick answer: ZeroGPT is a free, no-signup AI text detector at zerogpt.com that returns a single percentage — “your text is 87.2% AI/GPT generated” — and highlights the sentences it believes were machine-written. Its published claim is 98.4% detection accuracy with a false positive rate under 1%, produced by something it calls DeepAnalyse Technology (ZeroGPT). No peer-reviewed study has ever reproduced that figure. The best-documented independent measurement of ZeroGPT on academic prose put its overall accuracy at 64.35%, its false positive rate at 16.67% and its false negative rate at 45.15% (Pratama, PeerJ Computer Science, 23 June 2025), and an earlier 14-tool study found that none of the detectors it tested, ZeroGPT included, reached 80% accuracy (Weber-Wulff et al., International Journal for Educational Integrity, 25 December 2023). Four things follow. ZeroGPT is not GPTZero — they are separate companies, and a third site at zerogpt.net uses the same brand name with a different accuracy claim. ZeroGPT’s own terms of use disclaim any warranty that the site will be “accurate, reliable” and cap the company’s total liability at fifty US dollars (ZeroGPT terms). Its image and video detectors are not its own technology; they are licensed from TruthScan under a partnership announced on 22 January 2026. And the same site offers an AI Humanizer alongside the detector.

This guide covers ZeroGPT specifically: the product, the score, the evidence and the ownership. For the underlying detection mechanism across all tools, see how AI detectors work. For ranked alternatives with independent accuracy scores, see best AI detectors. For the institutional detector most students are actually assessed by, see Turnitin AI detection.

What ZeroGPT actually is

ZeroGPT is a free consumer web tool. You paste text into a box, press a button, and get a percentage back.

Four facts about the product shape everything else on this page.

It is free and needs no account. ZeroGPT’s pricing page lists a Free plan at “$0 forever” with no credit card required (ZeroGPT pricing).

It is a single-percentage tool, not a report. Where Turnitin produces an indicator inside an institutional similarity report visible only to staff, ZeroGPT gives the person holding the text a number and a set of highlighted sentences, and offers a PDF export. There is no institutional version, no roster and no audit trail.

It is a suite, not a detector. The site lists an AI Humanizer, AI Image Detector, AI Video Detector, Plagiarism Checker, AI Paraphraser, AI Grammar Check, AI Summarizer, AI Translator, Word Counter, Dictionary, AI Email Helper, Background Remover, Citation Generator and Pomodoro Timer alongside the text detector.

It is operated by a Wyoming company, not by an academic-integrity vendor. ZeroGPT’s terms of use name the operating entity as OLIVE WORKS LLC, at 5830 E 2nd St, Ste 7000 #9537, Casper, Wyoming (ZeroGPT terms). ZeroGPT is a consumer software business rather than an assessment company, and it makes no institutional commitments.

ZeroGPT is not GPTZero, and neither is zerogpt.net

ZeroGPT and GPTZero are different products from different companies. GPTZero is at gptzero.me and was founded as an academic-integrity tool; ZeroGPT is at zerogpt.com and is operated by Olive Works LLC. They have different detection models, different accuracy claims, different pricing and different independent test results. The names are near-anagrams of each other and are often confused. If a colleague, marker or client tells you “the detector said 92%”, the first question is which detector.

There is also a third site. zerogpt.net uses the ZeroGPT brand name, carries its own copyright line (”© 2026 ZeroGPT. All rights reserved.”), names no operating entity, and advertises a 0.0% false positive rate — a different claim from zerogpt.com’s “under 1%”. We could not establish any relationship between the two sites. Treat a result as coming from the specific domain you used, and check the address bar before quoting a number to anyone.

For ranked comparisons with independent scores rather than vendor claims, see best AI detectors.

How ZeroGPT’s detector works

ZeroGPT attributes its results to DeepAnalyse Technology, a trademarked name that appears on the homepage and throughout the marketing copy (ZeroGPT).

ZeroGPT has published no technical description of DeepAnalyse. There is no model card, no architecture note, no training-corpus description, no released evaluation set and no paper. Compare that with Turnitin, which documents that it segments text, scores segments between 0 and 1, pools to sentence level and aggregates, and states outright that it does not use perplexity or burstiness. ZeroGPT states the outcome and nothing about the method.

The interface shows that it scores at the sentence level and aggregates: the interface highlights individual sentences as AI-generated and returns a document-level percentage, which is the same broad shape most classifier-based detectors use. Anything beyond that is inference, and this page does not make it. For the mechanisms consumer detectors generally use — token predictability, sentence-length variation and trained classifiers — see how AI detectors work. Do not assume ZeroGPT implements any specific one of them, because ZeroGPT has not said.

ZeroGPT’s own FAQ frames detection results as something to use “within a fair review process” rather than as evidence on their own (ZeroGPT FAQ).

What ZeroGPT will and will not scan

The published limits are thin, and two of them conflict between pages on ZeroGPT’s own site.

The minimum-length figure appears only in the FAQ. ZeroGPT’s FAQ says the tool needs 150 to 200 or more words for a reliable reading, and that 500 to 1,000 or more gives better stability (ZeroGPT FAQ). The detector interface itself imposes no such floor and will happily return a confident-looking percentage on two sentences. By the vendor’s own account, a score on a short passage is outside the range where the tool is designed to work.

“All languages” and a six-language selector are not the same claim. ZeroGPT’s homepage says it supports “all the languages with the highest accuracy rate” and the FAQ repeats the “all available languages” framing, while the interface offers six European languages. No per-language accuracy figure is published for any of them. Every independent result cited on this page concerns English.

How to read a ZeroGPT score

The number is a model’s confidence, not a measurement of what happened when the text was written.

A percentage is not a proportion of the document. “87.2% AI/GPT generated” does not mean 87.2% of your words came from a chatbot. It is an aggregate of sentence-level probabilities. It should not be read as a word count.

The highlighted sentences are where the score came from, and they are the only part you can check. If ZeroGPT highlights four sentences in a 900-word piece and returns a high percentage, look at those four sentences. ZeroGPT’s own FAQ says that “highly polished, formulaic, or low-entropy writing can resemble AI” (ZeroGPT FAQ).

Two runs of the same text can differ, and two detectors almost always will. ZeroGPT publishes no versioning, no changelog and no model-release notes, so a score you got in March 2026 was produced by a system you cannot now identify. Nothing on the site lets you reproduce an old result.

Which AI models ZeroGPT says it can detect

ZeroGPT’s marketing lists “20+ LLM families covered” and names ChatGPT, GPT-3, GPT-4, GPT-5, GPT-6, Gemini, Grok, Perplexity AI, Claude, DeepSeek and Llama models (ZeroGPT).

Three qualifications belong with that list.

No per-model detection rate is published for any of them. The list says which model families ZeroGPT claims to recognise; it does not say how often it recognises each one. That figure is data not available for every model named.

The list is a claim of coverage, not evidence of it. The most recent independent studies below tested ZeroGPT against ChatGPT o1 and Gemini 2.0 Pro output (Pratama, 2025) and earlier ChatGPT versions. No published study we could find has tested it against GPT-5, GPT-6, Claude or Grok output.

Detecting a family is not the same as detecting a version. Detector performance is known to degrade against models newer than the detector’s training data — the PeerJ study found exactly this pattern of “poor performance” on the latest model versions across the three tools it examined (Pratama, 2025).

What ZeroGPT claims about accuracy

There are two different vendor claims on two different ZeroGPT pages, and the gap between them matters.

The homepage claim is specific: “98.4% Detection accuracy” and “<1% False positive rate.” No methodology, no test set, no date, no sample size and no definition of “accuracy” accompanies either figure (ZeroGPT).

The FAQ claim is softer and names its own source: “best-in-class performance — pushing toward >98% on internal evaluations.” That is ZeroGPT describing its own headline number as the output of internal evaluations (ZeroGPT FAQ). “Pushing toward >98% on internal evaluations” and “98.4% detection accuracy” are not the same statement, and a reader who only sees the homepage will not know the difference.

What is missing from both is the thing that makes an accuracy figure meaningful. An accuracy percentage on a detector is a function of the test set: which models generated the AI half, how long the passages were, how much human writing was included, and whether any of it was paraphrased. ZeroGPT publishes none of them. The 98.4% figure cannot be checked as published, and every independently measured figure is lower.

What independent testing actually found

Five independent measurements are summarised below. None of them reaches 98%.

The 2023 and 2025 studies measured different things. Weber-Wulff and colleagues tested 14 detectors in 2023 and found that none exceeded 80% accuracy, that the tools leaned towards classifying text as human-written, and that paraphrasing made results worse (Weber-Wulff et al., preprint at arXiv:2306.15666). Pratama’s 2025 study put ZeroGPT’s overall accuracy on journal abstracts at 64.35%, with a 16.67% false positive rate and a 45.15% false negative rate (Pratama, PeerJ Computer Science 11:e2953). ZeroGPT publishes no version history, so it is not possible to say which version either study tested.

The most favourable independent result is also the most recent. Cheng and colleagues, testing manuscript introduction sections in November 2025, found ZeroGPT returned only 1.6% to 6.5% AI content on human-written text and 50% to 92.5% on AI-generated text, and could separate five different AI-use conditions at p < 0.001 (Cheng et al., Advances in Simulation). The same authors concluded that these tools “should not be relied upon alone at this time.”

Paraphrasing changed the result sharply in one test. Kar and colleagues found ZeroGPT scored a QuillBot-paraphrased version of AI text at 32.31% AI, down from 95.03% for the original, while a Grammarly-paraphrased version still scored 96.74% (Kar et al., Indian Journal of Psychological Medicine). That is a three-fold swing driven by which editing tool touched the text, on a single sample of under 500 words. The n=1 sample size means this is a signal, not a rate — but the direction matches every other study of paraphrased text across the category, covered in best AI humanizers and best AI paraphrasing tools.

One widely cited figure does not belong to ZeroGPT alone. The Stanford study by Weixin Liang and colleagues, published in Patterns in July 2023, reported a mean false positive rate of 61.22% on TOEFL essays written by non-native English speakers, against roughly 5.19% on US eighth-grade essays, and ZeroGPT was one of the seven detectors tested (Liang et al., Patterns, preprint at arXiv:2304.02819). That 61.22% is the seven-detector mean, not ZeroGPT’s individual score, which the published text does not break out. Pratama’s 2025 study, by contrast, found no significant difference between native and non-native authors in ZeroGPT’s results.

And one 2026 study that people will reach for did not test ZeroGPT at all. The peer-reviewed evaluation published in the International Journal for Educational Integrity on 29 June 2026 by Van Vlasselaer, Van Droogenbroeck and Spruyt — the study that found 0% strict accuracy on fully AI-generated papers for two named tools — tested GPTZero, Copyleaks, Turnitin and Pangram (Van Vlasselaer et al.). ZeroGPT was not among them.

The false-positive problem, and who it falls on

ZeroGPT’s own FAQ says regular writing is most at risk: “highly polished, formulaic, or low-entropy writing can resemble AI” (ZeroGPT FAQ).

The discipline effect is measurable. Pratama found ZeroGPT’s performance on AI-assisted abstracts varied significantly by field, with a median of 0% detected in technology and engineering abstracts against 30.48% for interdisciplinary work, at p < 0.001 (Pratama, 2025). The same paper found no significant difference between native and non-native authors in ZeroGPT’s results — a point in ZeroGPT’s favour, and one that runs against the general pattern in the category.

ZeroGPT’s advice for reducing a false positive is about style. The FAQ tells users to add “concrete details, sources, and varied sentence structures” and to avoid generic language (ZeroGPT FAQ) — advice that addresses how text reads rather than who wrote it, and does not help with work already submitted.

There is no appeal process. Turnitin has documented thresholds and an administrator who can switch the feature off; ZeroGPT publishes a support email address. Nothing in ZeroGPT’s product or terms provides a route to contest a score.

What ZeroGPT’s own terms of use say about accuracy

The terms disclaim accuracy outright. ZeroGPT’s terms of use state that the site is provided “on an ‘as-is’ and ‘as available’ basis,” that the company “expressly disclaim[s] any and all warranties,” and specifically that it does not guarantee the site “will be accurate, reliable” (ZeroGPT terms of use). Disclaimers of this kind are common in consumer software; the terms, not the homepage, are the binding document.

Total liability is capped at fifty dollars. The terms limit the company’s liability “for any damages arising from or related to this agreement” to “a maximum of fifty U.S. dollars (u.s. $50).”

The jurisdiction clause names “Netherlands County, California”, which is not one of California’s 58 counties.

The image and video detectors are licensed from TruthScan

ZeroGPT’s AI image detection is not ZeroGPT technology. On 22 January 2026 ZeroGPT announced a partnership with TruthScan, a deepfake-detection company, to provide image detection to ZeroGPT users, expanding the site beyond text (press release issued by EZ Newswire on behalf of ZeroGPT). The release placed the detector at both zerogpt.com/ai-image-detector and truthscan.com, and quoted ZeroGPT chief executive Rawad Baroud saying: “People use our platform to detect AI content because they trust the accuracy of our product.”

This reframes the best-documented public test of ZeroGPT’s image side. NewsGuard’s audit, published 8 May 2026 and run in late April and early May 2026, tested five image detectors against 45 images and found ZeroGPT wrongly flagged 20% of authentic news photographs as AI-generated, while catching 93.33% of deliberately manipulated ones (NewsGuard, cited from our own best AI image detectors analysis). Those results describe a detector launched three months earlier through the TruthScan partnership. A ZeroGPT image score and a ZeroGPT text score therefore come from different systems.

A ZeroGPT video detector is also listed on the site. Whether it shares the TruthScan engine is data not available; ZeroGPT has published nothing about it and we found no independent test of it.

The detector and the humaniser

ZeroGPT lists an AI Humanizer and an AI Paraphraser in the same tool menu as its AI detector. Offering both is common across free-tool sites; best AI humanizers covers the tools themselves. ZeroGPT publishes nothing about how its humaniser and detector relate.

What it costs

Paid-tier prices are data not available here. ZeroGPT’s pricing page renders its plan prices client-side and did not return figures when fetched on 27 September 2026, and the USD prices on third-party software directories are undated and disagree with each other. Check ZeroGPT’s pricing page directly, and note that prices may be shown in your local currency.

What is confirmed is the shape of the offer: the free tier is genuinely free, needs no account, and caps input at 15,000 characters per check. For most individual uses, the free tier is the product.

How ZeroGPT compares with the tools it gets confused with

The figures come from different studies on different text, so they are not a like-for-like ranking. The 2026 peer-reviewed comparison found that every tool it tested classified human writing correctly and that three of the four detected none of the fully AI-generated papers (Van Vlasselaer et al.). Full rankings with current scores are on best AI detectors.

What to do, by situation

If you are a student who has been flagged by ZeroGPT. A ZeroGPT score is not an institutional finding, and no university assesses with it. If someone has used it against you, the strongest material you have is process evidence: version history from Google Docs or Word, drafts with timestamps, notes, outlines and references. The legal position on detector-based findings has hardened — in Matter of Newby v Adelphi University, decided 28 January 2026, a New York court annulled an integrity finding built on a Turnitin AI score, in a case where the student had submitted Grammarly and ZeroGPT reports showing his essay as human-written. Best AI detectors sets out what each tool’s evidence is actually worth.

If you are a teacher considering ZeroGPT. The 2025 PeerJ study measured a 16.67% false positive rate on journal abstracts; there is no institutional version, audit trail or appeal route; and ZeroGPT’s terms warrant no accuracy. ZeroGPT’s own FAQ frames results as one input “within a fair review process”. Best AI for teachers covers the assessment-design approaches that survive the detector problem.

If you are a writer or freelancer checking your own work before a client sees it. The free tier covers this. Paste more than 200 words, read the highlighted sentences rather than the percentage, and treat a flag on your own prose as a note about how formulaic that passage reads. Expect a different answer from a different tool.

If you are checking an image. Use a purpose-built image detector and know that the ZeroGPT-branded one is TruthScan’s. On NewsGuard’s audit no tool was good at both authentic and manipulated images; see best AI image detectors for the trade-off.

If you are deciding whether to pay. Paid prices, methodology and model versions are unpublished; test the free tier on text whose origin you know first.

Who runs ZeroGPT

The operating entity named in ZeroGPT’s terms of use is OLIVE WORKS LLC, at 5830 E 2nd St, Ste 7000 #9537, Casper, Wyoming (ZeroGPT terms). The chief executive is Rawad Baroud, named in ZeroGPT’s own January 2026 press release (EZ Newswire on behalf of ZeroGPT), who also spoke to NewsGuard for its May 2026 image-detector audit, attributing false positives on real photographs to resizing, compression, unusual lighting, high contrast and blur.

Beyond that, the public record is thin. Founding date, headcount, funding, revenue and any corporate parent are data not available: ZeroGPT publishes no about page with company history, no newsroom and no leadership page, and we found no filing or credible primary report establishing any of them. The site’s only published contact route is support@zerogpt.com.

Summary

• The product: a free, no-account text detector at zerogpt.com returning a percentage and highlighted sentences, capped at 15,000 characters per check, alongside image, video, humaniser and other free tools.
• The claim: 98.4% accuracy and under 1% false positives, which ZeroGPT’s FAQ describes as from internal evaluations; no method or test set is published.
• Independent results: 64.35% overall accuracy on journal abstracts (PeerJ, 2025); 1.6% to 6.5% AI detected on human scientific prose and 50% to 92.5% on AI text (Advances in Simulation, 2025); no tool above 80% accuracy in a 2023 14-tool study.
• The terms: no warranty of accuracy, and liability capped at $50.
• Operator: Olive Works LLC of Casper, Wyoming; chief executive Rawad Baroud. Image detection is licensed from TruthScan.
• Not GPTZero, and not zerogpt.net, a separate site using the same name.

Frequently asked questions

Is ZeroGPT accurate?

Not at the level it claims. ZeroGPT advertises 98.4% detection accuracy and a false positive rate under 1%, and its own FAQ describes that figure as coming from internal evaluations. The best-documented independent measurement, published in PeerJ Computer Science on 23 June 2025, put ZeroGPT’s overall accuracy at 64.35% on journal abstracts, with a 16.67% false positive rate and a 45.15% false negative rate. An earlier 14-tool study published in December 2023 found no tool above 80% accuracy. A more favourable November 2025 study found it returned only 1.6% to 6.5% AI content on human scientific prose. None of the independent figures approaches the vendor’s claim.

Is ZeroGPT the same as GPTZero?

No. They are separate products from separate companies with separate detection models and separate accuracy claims. ZeroGPT is at zerogpt.com and is operated by Olive Works LLC of Casper, Wyoming; GPTZero is at gptzero.me. The names are near-anagrams and are often confused. If a score is being quoted at you, establish which of the two tools generated it before you respond to it, because their independent test results differ substantially.

Is ZeroGPT free?

Yes, and the free tier needs no account or credit card. ZeroGPT’s pricing page lists a Free plan at “$0 forever” that includes text detection with sentence highlighting and a PDF report, capped at 15,000 characters per check. Paid Personal, Business, EDU and Enterprise plans exist with monthly and annual billing and an advertised 30% annual discount, but ZeroGPT’s pricing page renders its figures client-side and did not return prices when checked on 27 September 2026, so the paid rates are not published here. For most individual uses the free tier is the whole product.

Why does ZeroGPT say my writing is AI when I wrote it myself?

Because the detector responds to how predictable and regular your prose is, not to who typed it. ZeroGPT’s own FAQ states that “highly polished, formulaic, or low-entropy writing can resemble AI,” which describes careful edited writing, technical and methodological passages, rigid disciplinary conventions and the work of writers using a second language. A 2023 Stanford study published in Patterns found seven detectors, ZeroGPT among them, flagged TOEFL essays by non-native English writers at a mean false positive rate of 61.22%; the 2025 PeerJ study found no significant native/non-native difference in ZeroGPT’s results.

Can ZeroGPT detect ChatGPT?

Sometimes, and less reliably on longer or edited text than the marketing suggests. ZeroGPT lists ChatGPT, GPT-3, GPT-4, GPT-5 and GPT-6 among the “20+ LLM families” it claims to cover, but publishes no detection rate for any individual model. One 2024 study scored an unedited ChatGPT article at 95.03% AI, while the 2025 PeerJ study, using ChatGPT o1 and Gemini 2.0 Pro, recorded a 45.15% false negative rate on AI-generated abstracts, meaning it missed nearly half of them. No published study we could find has tested ZeroGPT against GPT-5 or GPT-6 output.

Does ZeroGPT detect paraphrased or humanised text?

Inconsistently, and the tool used to paraphrase changes the answer more than anything else. In a 2024 study in the Indian Journal of Psychological Medicine, ZeroGPT scored an AI-written article at 95.03% AI, a QuillBot-paraphrased version of the same text at 32.31%, a ChatGPT-paraphrased version at 90.61% and a Grammarly-paraphrased version at 96.74%. That was a single sample of under 500 words, so treat it as a signal rather than a rate, but the direction matches the wider literature: machine paraphrasing degrades detector performance substantially and unevenly. The 2023 Weber-Wulff study reached the same conclusion across 14 tools.

How many words does ZeroGPT need to give a reliable result?

ZeroGPT’s FAQ says 150 to 200 or more words, and recommends 500 to 1,000 or more for stable results. That figure appears only in the FAQ: the detector interface itself sets no minimum and will return a confident-looking percentage on a single sentence. By the vendor’s own account, a score on a short passage is outside the range where the tool is designed to work, so short-text results should be discarded rather than discounted. The maximum input on the free tier is 15,000 characters per check.

Is ZeroGPT safe to use with my own unpublished writing?

ZeroGPT’s FAQ states that submitted text is “not saved, shared, published online, or used to train” its model and is “discarded” after the result is produced. That is the vendor’s own statement and we have no independent verification of it. Its terms of use provide the usual counterweight: the service is supplied “as-is,” all warranties are disclaimed, and total liability is capped at fifty US dollars. For genuinely confidential material — unpublished research, client work under NDA, legal drafts — a stated retention policy backed by a fifty-dollar liability cap is thin protection, and the safer choice is not to paste it into a free web tool.

Do teachers use ZeroGPT?

Some do, though it is not built for institutional use. ZeroGPT has no institutional version, no roster or class integration, no audit trail and no appeal process, and its terms of use warrant nothing about accuracy. Universities that run automated AI detection generally do it through Turnitin, which at least documents its thresholds and lets administrators disable the feature. ZeroGPT’s own FAQ frames its results as something to use “within a fair review process” rather than as evidence on their own.

How does ZeroGPT compare to Turnitin?

They are different kinds of tool. Turnitin is an institutional product whose AI writing indicator is visible only to instructors and administrators, is licensed by the institution, and publishes a documented false-positive claim of under 1% on documents scoring above 20%. ZeroGPT is a free consumer tool that anybody can point at anybody’s text and that publishes no methodology. On independent evidence neither is strong: the June 2026 peer-reviewed comparison found Turnitin missed 100% of fully AI-generated master’s-level papers, while the 2025 PeerJ study put ZeroGPT’s overall accuracy at 64.35%. If you are a student, the distinction that matters is that Turnitin’s score can trigger a process and ZeroGPT’s cannot.

Who owns ZeroGPT?

ZeroGPT’s terms of use name the operating entity as OLIVE WORKS LLC, registered at 5830 E 2nd St, Ste 7000 #9537, Casper, Wyoming. Its chief executive is Rawad Baroud, named in ZeroGPT’s own January 2026 press release announcing an image-detection partnership with TruthScan. Founding date, headcount, funding and any corporate parent are not published, and we found no filing or credible primary report establishing them. The site’s published contact route is a support email address.

Is zerogpt.net the same as zerogpt.com?

No, and this is a third layer of the naming confusion. zerogpt.net is a separate site using the same brand name, carrying its own ”© 2026 ZeroGPT” copyright line, naming no operating entity, and advertising a 0.0% false positive rate — a different claim from zerogpt.com’s “under 1%”. We could not establish any relationship between the two. Every figure and every independent study on this page concerns zerogpt.com. If you are quoting a score to anyone, check which domain produced it.

Does ZeroGPT work in languages other than English?

ZeroGPT says so, but publishes nothing to support it. Its FAQ states that “all available languages are supported across all the tools” and its homepage claims support for “all the languages with the highest accuracy rate,” while the site interface offers a six-language selector covering English, French, German, Spanish, Italian and Portuguese. No per-language accuracy figure is published for any language, and every independent study cited on this page tested English text only. Treat a non-English ZeroGPT score as unmeasured.

← All guides