Full transcript
0:00Hello and welcome back to another review
0:02of another AI content detector. And in
0:06this one we're talking about Winston AI,
0:09which is one of the ones I heard a lot
0:10about just the web in general. But let's
0:13actually go through and see does it hold
0:16up? Is the content detection accurate,
0:19more accurate or less accurate than the
0:20ones we've done before? And do the other
0:22bells and whistles around Winston AI
0:24make it a good choice for anyone who
0:26wants to test whether a work of fiction
0:31or non-fiction is AI or not. So let's
0:33dive in.
0:34First of all, let's talk a little bit
0:36about the pricing.
0:37Um they clearly want you to be on their
0:40annual plan because the annual plan is
0:41significantly cheaper per month, you
0:44know, when you average it out than as
0:47compared to the monthly. There is a free
0:49one that gives you 2,000 credits and uh
0:52to that's essentially
0:542,000 words.
0:56So it's not a whole lot. Like you can't
0:58really test anything with that. So I,
1:00for the sake of this video, upgraded to
1:02the essential plan. But I didn't want to
1:04pay a whole year because that would be
1:05$120.
1:07Uh so I instead I went to the monthly
1:08and even the monthly is $18 a month.
1:12So you're saving almost half by going to
1:14the annual. And personally I don't like
1:16this because these numbers are some of
1:18the the highest that we've seen so far
1:20in the ones that I've checked out,
1:22especially if you go monthly. If you go
1:24annual, it's not too bad, but that just
1:26locks you in. And one of the things I
1:28liked about one of the others that we
1:30looked at, which was uh originality.ai,
1:33that one has an a la carte option. So
1:35you only use it when you use it. And so
1:38you could go months without using it and
1:40then come back and you still have all
1:41your credits and stuff. And it's a
1:43little more expensive a la carte like
1:45that, but I think in the long run I'm
1:47usually saving money unless you are
1:50frequently using these kind of tools. I
1:52don't like the monthly subscription. I
1:55don't think you need a monthly
1:56subscription. So when it comes to
1:58pricing, I am not very happy about this
2:01one, especially if you get into the more
2:03advanced ones. Like for instance, uh you
2:05don't have plagiarism on this essential
2:07plan. So if you want plagiarism
2:09checking, you need at least the $29 a
2:11month plan, which to me is starting to
2:13get far more
2:16um
2:17pricy than this kind of tool needs to
2:20be, especially compared to other tools.
2:22But that's me. Let's go and look at the
2:25actual tool itself and see if it has
2:26anything worth that kind of a price. So
2:29when you get in to the tool, you can
2:32come here to where it says new scan and
2:35you'll see a dashboard like this one.
2:39And from here you can paste in your
2:41text. So the text here is 100% AI
2:44written text
2:46um that I have here in this test. Uh so
2:49I even called it pure AI test right
2:51here. And on here on the left, you can
2:55select from a variety of different scans
2:57that they have for you. Of course
2:59there's the AI detector, which is one
3:01credit per word. This document is 24
3:0410 words.
3:07So that will use 2,410 credits for just
3:10the AI detector.
3:12Uh they also have a fact checker, that's
3:15another one credit per word.
3:18Uh and a writing feedback, which is also
3:20one credit per word.
3:22So if I used all three of these, that
3:24would be three credits per word, which
3:25is quite a lot. You can tell I used up
3:28about 20,000 credits just with the tests
3:31I already ran for this thing. Give me 10
3:33seconds. If you want to help cranking
3:35out way more books with AI, join the
3:36Story Hacker AI waitlist below. While
3:38the group's closed, I'll send you my
3:40full prompt pack for free. Click the
3:43link, enter your email and I'll shoot it
3:44over. And I was just using the AI
3:46detector, not the fact checker and the
3:48writing feedback. And additionally, if
3:50you have one of the higher tiers and you
3:52want to use plagiarism, the plagiarism
3:54checker, that's two credits per word. So
3:57there's potential to get up to five
3:59credits per word, which I find to be a
4:02huge drain, uh which is once again goes
4:05to the whole pricing issue of this
4:07thing. I think it is overpriced. And
4:09these credits are easy to use up.
4:12But anyway, let's just try these. I'll
4:15unclick fact checker because this is a
4:17piece of fiction, so we don't need a
4:19fact check for it. Uh but if we go ahead
4:21and scan,
4:27it gives us this thing here. So it gives
4:29me a probability that 41% of this or or
4:33the text is 41% human, it says. Uh which
4:36is not true. This is 100% AI, so this
4:39should be 0% human.
4:42Uh so it didn't do great on that
4:45front. And it said the readability score
4:46is actually pretty good.
4:48Um so that's kind of nice. And uh I
4:52wonder because uh this text was
4:55generated using Gemini 3,
4:58uh which tends to be good at mimicking
5:00certain styles. So when I gave for to
5:03create this text, I gave it a bunch of
5:04my own style information and uh samples
5:07of my work.
5:09I think Google's Gemini uh 3 is pretty
5:13good at mimicking styles. And so I
5:16wonder if that's why it has this pretty
5:18high
5:19uh human percentage here, given the fact
5:23that there was no human text here.
5:27Uh and the readability score is also
5:28pretty good. Now if we go to uh we
5:31didn't do fact checker, but for writing
5:33feedback, it actually gives us some
5:35interesting things that we could replace
5:37here. Like replace dirt packed under my
5:39fingernails, cool and gritty, kneeling
5:41in the in the loam felt right, far
5:43better than the stiff wooden chair back
5:44at the cottage with cool, gritty dirt
5:47packed beneath my fingernails as I knelt
5:49on the rich loam of my far more fitting
5:50seat the cottage's stiff wooden chair. I
5:52find it funny that there's an M dash
5:54here. These auto suggestions are clearly
5:56also created with AI, but it's not
5:58actually a bad recommendation, I think.
6:02No. And there's quite a few of these.
6:05So no grammar
6:08issues because this was AI written, so
6:10there's not going to be really any major
6:11grammar or spelling issues, just
6:14areas where we might want to paraphrase
6:16or provide clarity. Like the dialogue
6:19here is charming and effectively
6:20establishes Elaine's character and her
6:22connection to nature. It also subtly
6:23foreshadows her later emotional reaction
6:25to the bluebell. These are definitely AI
6:27suggestions here. It's not necessarily a
6:30bad thing. I just find it ironic. All
6:32right. So that's what we do. You can
6:34already tell we used up quite a few of
6:36my credits just running that.
6:38Uh we used about 5,000, a little over
6:405,000 credits to run the AI detector and
6:42the writing feedback together. It would
6:44be even more if I had added fact checker
6:46or plagiarism. Can't do plagiarism right
6:48now because I'm I'm not on that higher
6:50plan, but
6:52in general, these kind of features,
6:54we've seen them on other AI detectors.
6:57I don't think there's really anything
6:59particularly better about this one than
7:01the others.
7:02Uh there is one feature that I thought
7:04was pretty cool. There's this one called
7:06text compare, where you can put your
7:09text in uh
7:11like put one version of your text here,
7:13put a second version of the text here
7:15and compare it and it'll show you where
7:16it's made changes. I actually think
7:18that's a really useful thing if you're
7:20using say an AI humanizer and you want
7:23to compare and contrast and actually see
7:25what changes were made, that could be a
7:27really useful feature here. Not
7:29necessarily something that I would use a
7:31whole subscription for this. And it does
7:34use more credits. In this case, the
7:35comparison costs 1/4 of a credit per
7:37word. It's not clear on whether that's
7:40for that probably is for both. So if you
7:43have 2,000 words here and 2,000 words
7:45here,
7:46it'll take a quarter of a credit for
7:48each. So that'd be 1,000 credits total.
7:52So a little bit less than a full scan,
7:54but still using up quite a few of your
7:56credits. So I imagine there's other text
7:58comparison tools out there that just do
8:00text comparison that are probably a lot
8:02cheaper. But regardless, they have that.
8:05They have a grammar checker here. They
8:06have a number of integrations, uh like a
8:09Chrome extension and
8:11which can be useful in certain
8:13circumstances.
8:15But on the whole,
8:17oh, and they also have image
8:19scans to detect if an image was AI
8:21generated, which I haven't tested, but
8:23it's also not something I really am
8:25interested in,
8:27but could be useful if that's something
8:28that you want to check. I also imagine
8:31this is probably something that's very
8:32difficult to to figure out in certain
8:35use cases, especially if you've only
8:37used AI to like subtly modify an image.
8:40Uh like using Google's nano banana or
8:42something like that. That would probably
8:43be a little bit difficult, but we'll
8:45see. Uh anyway, that is the general
8:48overview of how this is used. It's
8:49pretty simple. And again, I think it's
8:52pretty basic for the pricing. Other
8:55tools have similar features for a little
8:58bit lower pricing or even have more
9:00features. And I also don't get a sense
9:03of a particular audience that they're
9:05trying to cater to. In the two that I've
9:08gone over so far, there's originality.ai
9:11and GPT-0 that I've reviewed so far, and
9:14they clearly had different audiences
9:16that they were catering to.
9:18That made it more useful for that
9:20audience. Originality.ai was more for
9:22content creators of SEO articles and
9:24stuff like that.
9:25And GPT-0 was more for teachers and
9:29educators and uh in that camp. This one,
9:32I don't really see any particular like
9:35it doesn't seem to be catering to any
9:36particular one. So that's just something
9:39to keep in mind there.
9:40But let's actually go through in the
9:42results of my tests that I put together.
9:45Uh for the pure AI fiction, where I
9:48would have expected 100% because it was
9:50pure AI, it gave me a 59% AI,
9:53which is a little bit lower than GPT-0
9:56and originality.ai, which makes it of
10:00the three, you know, it's only a little
10:01bit lower, but this is still
10:03uh the least accurate of the three.
10:07Uh for a pure AI non-fiction thing, they
10:09gave me 97.
10:11Uh where this I would also expect 100%
10:15and all three of these have done pretty
10:16well so far. The originality was the
10:18most correct here because it was 100%.
10:21For heavily edited AI, so this is a
10:23piece of AI uh fiction that I have that
10:26has been edited
10:27heavily by both myself and a line
10:29editor.
10:30So, we know that I would expect it to be
10:33a little bit more like the results we
10:34were getting here from the pure AI,
10:37but it's actually been ranking higher
10:41for whatever reason. Maybe it's just the
10:42model I used or something like that
10:44because I believe the model I used here
10:45was Gemini 3 and the model I used here
10:47was Claude, one of the Claude models.
10:50And it put this at 96% where I would
10:53have expected a little bit lower than
10:54that. Uh the GPT-3 here is closer to
10:58what I would have thought it would be.
11:01The next test I did was a piece of
11:03entirely human-written fiction, so no AI
11:06involved,
11:07and it gave me a 1% chance of AI, and uh
11:12that's, you know, that's pretty close.
11:13It's slightly less accurate than
11:16these two, which correctly assumed it
11:19was zero,
11:20uh but close enough that it doesn't
11:21really matter.
11:23And then we get the human-written
11:24non-fiction. So, this is a piece of
11:26non-fiction that I wrote uh way back
11:29when when I was at Kinderpreneur, so I
11:30know it's not AI,
11:32and it, like the other tools here, has
11:35incorrectly guessed that it was heavily
11:37written with AI. In this case, it gave
11:38it a 96%
11:40and it is not. It is not It We would
11:42expect zero. So far, the only one that's
11:45gotten a little bit more accurate on
11:46this one was GPT-3, which gave it a 53%,
11:49but even that's still pretty inaccurate.
11:51And both of these others have been
11:53pretty inaccurate overall. Then I gave
11:55it some Sherlock Holmes text, and it
11:58correctly guessed that it was mostly
12:00human,
12:02and some from the Bible, which it also
12:04correctly guessed that it was mostly
12:05human. Only a 1% and a 2% on these.
12:08Although, that is higher slightly, just
12:10slightly, than what the other two gave
12:13us. So, perhaps we'll slightly, like if
12:16we're looking at these results, they're
12:17very similar, but I would actually say
12:19that um
12:20that Winston AI has been slightly worse,
12:23just slightly,
12:25than the other two that we've covered so
12:26far on this channel. So, in my notes is
12:28I do like the text compare feature. Uh
12:31the grammar checker, plagiarism fact
12:32checker, and writing feedback tools are
12:34all great. This is the worst one I've
12:36found for pricing so far, and it's a
12:38slightly less accurate detector. That
12:40pretty much sums up my thoughts on this
12:43AI detector today. So, if you're
12:46interested in more of these, I'm going
12:48to put out more reviews for the the big
12:50ones that people talk about.
12:53So, we'll have a couple more videos on
12:54that. That's interesting. Be sure to
12:56subscribe, of course, and uh I'll see
12:58you in the next video.