The AI Detector Directory: Every Checker We Know, One Page

If you have spent any time trying to figure out whether a piece of text, an image, a voice clip, or a chunk of code was generated by a machine, you have probably discovered the same thing everyone else does: there are far too many AI detectors, they all claim to be the accurate one, and nobody hands you a map. You search for a name you half-remember, land on a marketing page, and then find three competitors with almost identical promises. It is genuinely confusing, and the confusion is not an accident. Detection is a crowded, fast-moving market where the incentive is to sound authoritative, not to be honest about limits.

This page is the map. It is a plain, navigational directory of the AI detectors we know about, organized by what they actually detect, with a short, honest description of each one. Think of it as the index at the back of the book rather than a chapter. If you want the deep dive on any given tool, follow the link at the end of its entry to the full review. If you just want to understand the landscape and find the handful of tools worth your attention, read straight through. This is our attempt at a genuine ai detector directory — a single list of ai detectors that treats each one like a real thing with real caveats, not a logo in a comparison grid.

A few ground rules before the list. First, no detector is a lie detector. Every tool below produces a probability, and probabilities are wrong sometimes. Second, the honest description matters more than the score. A detector that tells you "likely AI, 78% confidence" is being more truthful than one that flashes a red "100% AI" verdict, because the second one is pretending to a certainty that does not exist. Third, categories bleed into each other. Several text tools also try to do images now; several "detectors" are really plagiarism suites that bolted on an AI module. We have sorted things by their primary job. With that said, here is the whole board.

Text detectors: the crowded main event

This is where the noise is loudest. Text is the easiest content to generate at scale and the hardest to detect reliably, which is exactly why the category is packed with contenders. When people say "AI detector" without qualification, they almost always mean a text detector, and they usually mean one of the dozen or so names below. This is the core of any list of ai detectors, so it gets the most room.

The institutional heavyweight

• Turnitin — The detector most students actually meet, because schools buy it in bulk and it lives inside the assignment-submission workflow rather than as a standalone website you can visit. Its AI-writing indicator is bundled with the plagiarism check most institutions already pay for, which is the real reason it dominates campuses. The caveat is significant: it runs behind an instructor dashboard you cannot see, its scores have been quietly walked back and re-tuned more than once, and a student flagged by it usually cannot inspect or contest the underlying reasoning. If you want to understand exactly which engine your school is running and how it behaves, our breakdown of what AI detector Turnitin uses is the place to start.

The names you have seen in headlines

• GPTZero — Probably the most famous consumer-facing name in the category, launched into a media frenzy and now a full product with browser extensions and an API. It is genuinely usable and reasonably transparent about giving sentence-level highlights, but its early reputation for accuracy outran the reality, and it produces false positives on human writing that happens to be clean and predictable. See GPTZero explained for how it actually scores.
• ZeroGPT — A free, ad-supported site that shows up near the top of every search and is often confused with GPTZero despite being a separate, unrelated product. It is fast and costs nothing, which explains its traffic, but it is also one of the less consistent tools we have looked at, and its confident percentages should be treated as loose signals rather than findings. The full ZeroGPT review covers where it goes wrong.
• Originality.ai — Built explicitly for web publishers, SEO teams, and content agencies that need to vet freelance writers at volume. It tends to score aggressively toward "AI," which its target buyers often want, but that aggression is precisely the problem if you are on the receiving end of a false flag. Our Originality.ai review digs into that trade-off.
• Copyleaks — An enterprise-flavored platform that pairs plagiarism detection with AI detection and sells hard into education and business. It supports a lot of languages and integrations, which is a real strength, but "supports" a language is not the same as "is accurate in" that language, and its verdicts can swing on lightly edited text. Details in the Copyleaks review.
• Winston AI — Marketed with some of the highest accuracy claims in the business, aimed at educators and writers, with a clean interface and OCR so it can read handwriting and PDFs. Treat the headline accuracy number as marketing until proven otherwise; the interesting question is how it behaves on your text, not on the vendor's benchmark. See the Winston AI review.
• Sapling — Originally a writing-assistance and customer-support company that added a detector, offered free on its site with a simple probability output. It is competent and pleasant to use, but it is a smaller player without the scale of the leaders, and it is best treated as one signal among several. More in the Sapling review.

The academic-adjacent writing tools

• Scribbr — Best known to students as a citation and proofreading service, Scribbr layered an AI detector onto its academic-help brand, which is why it appears in a lot of student searches. It leans on a partner engine under the hood rather than fully home-grown technology, and it inherits both the strengths and the false-positive risks of that engine. The Scribbr review unpacks what is really running.
• Crossplag — A plagiarism-first tool with a bilingual and cross-language focus that added AI detection to its suite. It is a smaller, less-hyped option that can be useful for multilingual contexts, but the AI module is not its center of gravity and should be weighted accordingly. See the Crossplag review.
• Pangram — A newer, research-forward entrant that has drawn attention for a different training approach and some genuinely strong independent showings. We are cautiously more impressed by it than by most of the field, but "newer and promising" still means you verify before you rely on it for anything consequential. Read the Pangram review for why it stands out.

The writing platforms that also detect

• Grammarly — Yes, the grammar tool now waves at AI detection, which matters mostly because so many people already have it installed and assume its opinion is authoritative. It is not built primarily as a detector, and its signal is best understood as a convenience feature rather than a verdict you would defend in a dispute. We put that to the test in is Grammarly's AI detector accurate.
• QuillBot — Famous as a paraphraser, QuillBot also ships a free detector, which is a mild irony given that paraphrasers are one of the main ways people try to defeat detectors in the first place. It is easy to reach and free, but it sits squarely in the "quick sanity check, not evidence" tier. See is QuillBot's AI detector accurate.
• Writer.com — Once home to a widely shared free detector that a lot of people bookmarked, Writer.com then de-emphasized and effectively retired that public tool as its business pivoted toward enterprise generation. It is a cautionary tale about depending on any free detector that a company can pull at will. The story is in what happened to Writer.com's detector.
• Content-at-Scale / Brandwell — A content-generation company that offered its own detector, then rebranded from Content at Scale to Brandwell as its focus shifted. A generator selling a detector is a structural conflict worth keeping in mind, though it is hardly the only tool with that shape.

If your only question is "which of these text detectors is least bad for my situation," the sensible move is to stop treating the category as a leaderboard and read our comparative take in the best AI detectors ranked, which weighs transparency and false-positive behavior alongside raw accuracy claims. That is the single most useful page to pair with this directory if you are trying to actually choose something.

Free and consumer tools: the ones you will actually try first

Most people do not start with an enterprise contract. They paste text into whatever free box ranks first, get a percentage, and either relax or panic. That instinct is understandable and mostly harmless as long as you know what free detectors are and are not. Several tools already listed above — ZeroGPT, QuillBot, Sapling, GPTZero's free tier — live in this world, and there is a whole ecosystem of smaller free sites that are functionally interchangeable.

The honest description of the free tier as a whole is this: free detectors are fine for a first-pass gut check and dangerous as a basis for any decision that affects a person. They are typically ad-supported or lead-generation funnels for a paid product, they rarely explain their reasoning, and their confident-looking percentages are the least reliable numbers in the category precisely because there is no accountability behind them. If a free tool tells you something is "100% AI," that number is theater. We walk through the whole tier — what is worth using, what to ignore, and how to read the outputs without fooling yourself — in the free AI detectors guide.

There is also a specific subspecies worth calling out: detectors embedded inside SEO and content-marketing platforms. If you run a content operation, your existing writing suite may already include an AI-detection feature, which is convenient but comes with the same generator-adjacent conflicts noted above. We cover which platforms bundle detection and how much to trust the bundled version in SEO tools with built-in AI detectors.

Image detectors: telling generated pictures from photographs

Text detection gets the headlines, but image detection is arguably where the stakes have gotten highest fastest, because a convincing fake photo travels further and does more damage than a paragraph of synthetic prose. The image category is younger, the tooling is less standardized, and the honest caveat is stronger: image generators improve so quickly that a detector accurate on last year's model can be near-useless on this year's. Here is the map for pictures.

• Hive — One of the more serious names in the space, Hive sells AI-image and deepfake detection as an API to platforms and enterprises rather than as a toy you paste a JPEG into. It is a real, well-resourced player, but it is built for integration and moderation pipelines, and its consumer-facing surface is limited. See the Hive AI detector review.
• Was It AI (wasitai.com) — A simple, free, upload-a-picture consumer tool for the "is this photo real?" question, which is exactly the question ordinary people have. It is approachable and useful for a quick check, but its confidence should be read loosely, especially on edited, compressed, or screenshotted images. Full notes in the Was It AI review.
• Illuminarty — A detector aimed at spotting AI-generated art and imagery, popular with people trying to police generated art in communities and marketplaces. It is a reasonable option in a thin field, but like all image detectors it struggles as generators evolve and as images are post-processed.
• SynthID — Google's watermarking-and-detection approach, which is philosophically different from the rest of this list: instead of guessing after the fact, it embeds an invisible signal at generation time. That is a genuinely more robust idea, but it only works on content produced by systems that chose to watermark, so it says nothing about images from models that did not participate.

Because the image field is moving so fast and the tools are so uneven, we keep a dedicated overview that explains how image detection actually works, what watermarking changes, and why "it looked obviously fake to me" is not a reliable method. Start with AI image detectors explained before you trust any single verdict on a picture.

Video and deepfake detectors: the hardest problem on the board

If image detection is young, video and deepfake detection is younger still and considerably harder. Video is heavy, it is usually compressed and re-encoded on its way through social platforms, and the manipulations range from fully synthetic clips to subtle face swaps grafted onto real footage. The honest description of this entire category is that it is more research than product: much of the strongest work lives in labs, platform trust-and-safety teams, and specialist vendors rather than in a website you can use this afternoon.

Some of the image-focused vendors above — Hive in particular — extend into deepfake video detection, and there are academic and forensic tools that analyze temporal inconsistencies, unnatural blinking, lighting mismatches, and compression artifacts. For an ordinary person, the practical reality is that no free, reliable, general-purpose "paste a video, get a verdict" tool exists at the quality people imagine, and claims otherwise deserve heavy skepticism. We lay out what genuinely works, what is still lab-bound, and how to reason about a suspicious clip in AI video detectors explained.

Voice and music detectors: the audio frontier

Synthetic voice has crossed the line into "good enough to fool your relatives on a phone call," which makes voice detection both urgent and immature. AI music generation, meanwhile, has moved fast enough that streaming platforms and rights holders now care intensely about spotting machine-made tracks. These are two distinct problems — cloned speech versus generated songs — but they share the same audio-forensics DNA and the same core caveat.

That caveat: audio detection is easier to break than most people assume, because a bit of background noise, a phone-line codec, or a re-recording can strip away the very artifacts a detector relies on. There are specialist voice-clone detectors and emerging music-provenance tools, and some rely on watermarking rather than blind analysis, which — as with images — only helps when the generator cooperated. Rather than crown a single winner in a field this unsettled, we survey the landscape, the vendors, and the honest limits in AI music and voice detectors explained.

Code detectors: was this written by Copilot?

A quieter but real category: detecting AI-generated source code. This matters in classrooms teaching programming, in technical hiring where take-home assignments are supposed to reveal a candidate's own ability, and in some open-source and licensing contexts. Code detection is a strange beast because code is both more structured and more repetitive than prose — there are only so many ways to write a standard loop — which cuts both ways for a detector.

The honest description here is that code detection is even less mature and less standardized than text detection, and the false-positive risk is arguably worse: a competent human writing idiomatic, conventional code can look exactly like a model writing the same idiomatic, conventional code, because that is the whole point of idiomatic code. Tooling exists, some of it academic and some bolted onto broader plagiarism systems, but treating a code-detector verdict as proof of misconduct is a mistake. We go through what is out there and why to be careful in AI code detectors explained.

Humanizer-adjacent tools: read this section twice

Here is a corner of the market that deserves a flashing warning light. A growing number of companies sell both a "humanizer" — a tool that rewrites AI text to slip past detectors — and, often on the same website or under the same brand family, a detector. Sometimes they sell a detector openly and quietly offer bypass features; sometimes they are humanizer-first and offer a detector as a companion. Either way, there is a structural conflict of interest baked into the business model: a company that profits when detection fails has an incentive that points away from honest detection.

We are not saying every tool in this bucket is useless or dishonest. We are saying that when the same brand sells you the shield and the sword, you should read its detector's verdicts with extra suspicion and understand what its actual business is. Here are the ones people ask about most.

• JustDone — Marketed as an all-in-one AI content suite that includes both detection and humanizing/rewriting features, which is exactly the dual-role setup to be wary of. It may be perfectly pleasant to use, but its detector lives inside a product whose broader purpose includes making AI text harder to detect. See the JustDone review.
• Phrasly — Primarily known as a humanizer aimed at helping AI text pass as human, with detection offered alongside. Reading its own detector's "you're safe" verdicts is a bit like asking the locksmith who sold you the lockpick whether your door is secure. The Phrasly review spells out the conflict.
• Walter Writes — Another humanizer-forward brand that also fields a detector, positioned to help writing evade AI checks. Same structural caveat: a bypass-oriented company's detector is not a neutral referee. Details in the Walter Writes review.

The reason we include these in the directory at all is that people search for them and deserve a straight answer rather than either a fake endorsement or a pretend that they do not exist. A complete ai detector tools list has to acknowledge the humanizer economy, because it shapes how well every other detector on this page performs. Bypass tools and detectors are locked in an arms race, and today the bypass side is winning more often than the detection industry likes to admit.

Academic detectors beyond Turnitin

Turnitin dominates the North American campus imagination, but it is not the only integrity system in the world, and if you are outside its orbit — in Europe, in specific institutions, or in publishing and research-integrity workflows — you will meet others. These tools share Turnitin's core shape: they live inside institutional systems, instructors or editors see the results, and the person being evaluated often cannot inspect the reasoning.

• iThenticate — Turnitin's sibling aimed at researchers, journals, and publishers rather than undergraduates, used to screen manuscripts and dissertations before publication. It is a serious tool in academic publishing, but it inherits the same "you cannot see inside it" opacity as its consumer cousin.
• Compilatio — A plagiarism-and-AI system with strong adoption in French-speaking and European academic institutions, less familiar to English-only audiences but very real where it is deployed. Its AI detection carries the usual caveats about false positives on clean student writing.
• SafeAssign — Bundled into the Blackboard learning-management system, which is how many students encounter it without ever choosing it. It is primarily a plagiarism tool, and its AI-detection posture is weaker than the specialists, so a low or high signal from it should be read with care.

If you are a student, an educator, or a researcher trying to understand what runs in your specific institution and how to respond to a flag from something other than Turnitin, our roundup of academic AI detectors beyond Turnitin covers the whole institutional field and the appeals reality that comes with it.

Open-source detectors: for the curious and the technical

Finally, a category for people who want to look under the hood rather than trust a black box. Open-source and research-grade detectors are not usually the most accurate option on this page, but they are the most honest in a specific way: you can see how they work, which is a rare and valuable thing in a field built on proprietary confidence.

• GLTR — The Giant Language model Test Room, an early academic visualization tool that colors each word by how predictable it was to a language model. It does not hand you a verdict; it shows you the texture of the text and lets you reason about it yourself, which is both its charm and its limitation for anyone wanting a simple yes-or-no.
• Hugging Face detectors and models — A range of open detector models and demos hosted on the community machine-learning platform, spanning old GPT-2-era classifiers and newer efforts. Quality varies wildly from model to model, and many are dated relative to the generators they now face, so they are best treated as educational and experimental rather than authoritative.

The value of the open-source corner is not that it will win your dispute — it usually will not — but that it demystifies the whole enterprise. Spend an afternoon with a tool that shows its reasoning and you will never again be quite so impressed by a commercial detector's confident percentage. For a tour of what is available and what it teaches, see open-source AI detectors.

How to actually use this directory

You now have the whole board: text, free and consumer, image, video and deepfake, voice and music, code, the humanizer-adjacent tools to treat with suspicion, the academic systems beyond Turnitin, and the open-source options for the curious. That is genuinely most of the meaningful detectors a normal person will encounter, which is the point of building a single all ai detectors reference instead of scattering the information across a hundred pages.

The way to use it is not to run your text through every tool and average the scores — that feels rigorous but is not, because the tools share failure modes and a majority vote among biased judges is still biased. Instead, use this page to identify which category you are actually in, pick one or two of the more transparent tools in that category, and read their outputs as probabilities with error bars rather than verdicts. If the stakes are high — a grade, a job, an accusation — remember that no detector on this list is designed to be evidence, and the honest ones will tell you so themselves.

Bookmark this page as your starting point and treat every linked review as the real substance. The directory tells you what exists; the reviews tell you whether to trust it. And if you take one thing away from the whole exercise, let it be a healthy allergy to certainty: in a field where the generators improve every few months and the detectors scramble to keep up, the tool that admits its doubt is almost always more trustworthy than the one that does not. Come back whenever you meet a name you do not recognize — that is what a directory is for, and we will keep expanding this one as new checkers appear and old ones fade.