It's almost impossible to escape AI being used in our articles. My experience has been that the following applications use it including:
Microsoft Word
Hemingway Editor
Grammarly (of course)
Microsoft Notepad
Google docs
I turn Grammarly off when I'm writing in Word, but MIcrosoft picks up (to some extent) what Grammarly would have told me about. It's a losing battle. :-(
I am currently using GPTZero as a light check against my writing. ChatGPT is the biggest offender. For example, even if it tells me that it is my voice and tone, I'll paste that into GPTZero, and it is immediately flagged as 100% AI-written.
As a test, I typed the statement, "I’m testing whether Google Doc uses AI checkers like Grammarly to help me catch errrrors. It not only caught it in Google Doc that error was misspelled, but it's being caught in my test sentence here. I have to assume that it's because I have Grammarly installed on my laptop. According to Seattle University ("Grammarly stores all text entered into its service on their servers. This means any high risk data you type could be at risk of being accessed or misused. Disabling Grammarly when working on sensitive documents ensures that this information remains secure.") (https://seattleuniversity.atlassian.net/servicedesk/customer/article/2002190356) (Or does it? What about Microsoft in Word, for example)
Using AI to make our lives easier is proving to be a challenge. Is there no escaping it?
Am I ranting? Yes, I'm sorry. I do that sometimes. AI Detection Tools are a vexing issue.
Thank you for this detailed review on what is happening on the Medium platform. It is very sad to see that a platform that worked so well before for the writers are now only earning cents...I hope that things will improve as I do really like the community. AI detectors do not work very well and many times can give false results...in my experience.
>For example, some stories, even from writers with sizable followings on a large pub like Illumination, with 189K followers, get under 10 views, usually from direct links shared by writers or editors.
Do you think that this issue is also, in some part, the problem?
Most people who subscribe to a publication only see a couple stories from that pub per day at most, through their Daily Digest (which seems to aim to provide a broad overview, instead of featuring lots of stories from one single pub). If ILLUMINATION is publishing dozens of stories per day, won't it be pretty unlikely that a reader will get recommended that story, or that they'll see it on ILLUMINATION's front page when it's quickly going to disappear under the constant stream of new articles?
I feel like this is another consequence of Medium not thinking too far ahead on how large publications work.
Hello! Your points trigger me and my response will be an article long. This is to show my respect and my good intention to inform authentic writers like you. This argument has been made since Illumination became a prominent publication in 2020. I am the owner of this pub, but I am also a consulting editor for many large—even larger—publications. The publication interface has almost no impact on views.
We analyzed over a million stories from multiple angles, including internal and external views/read/engagement. The quality of content and presentation has minimal impact. This might sound shocking, but 99% of a story’s success depends on whether Medium’s human curators like it and add it to the distribution system. If they do, large publications like Illumination work wonders and can make a story go viral. If they don’t, no matter how impactful, memorable, or well-written the post is—even if it comes from a distinguished professional—it will struggle to surpass 100 views from direct links.
For example, a new writer with a very small following wrote a story about "old person smell" on Illumination. The story received over 30,000 views overnight because Medium loved it. Meanwhile, another writer with more followers and substantial qualifications, who wrote about the same topic from a scientific perspective—making it more accessible and appealing to the public—received only 22 views on the same publication through direct links. Yes, this is the writer you characterize as "a crying wolf." Checkout his story to gain insights: https://medium.com/p/98166f12a488
Why is this happening? Because Medium selectively amplifies certain types of content from specific writers. I was a Boost nominator for 12 months, working with 30 senior editors who curated content for me. We followed the criteria Medium published—but Medium’s curators did not follow their own criteria. Stories that exceeded the stated requirements were rejected, while those that barely met them were boosted. Even stories that didn’t meet the criteria at all were organically boosted—and still are.
That was one of the reasons I resigned from the program and ultimately felt compelled to close my best publication. I link those stories here if you missed them:
I was not the only one who left the Boost program. Several pub owners like Dr Giles also resigned for the same reason. This program, which looks great in theory, only supported 1% of good writers, disadvantaging 99%.
It is not about the platform but the mismanagement of it, lack of transparency, and strategic approach. It is the reason so many great writers like Tim Denning stopped contributing, and some even deleted their stories and sme entirely left the platform. Now, those who can use AI and create sensational content and present them with humanizers in Grade 3 level shine.
Hi Dr. Yildiz, thanks for replying! And I hope that my comments here come across as earnest and open to engaging and figuring out what works best for all of us as a community moving forward, and not as an attack on you or Dr. Broadly in any way. I follow you both and do miss your insights on Medium.
When you say:
We analyzed over a million stories from multiple angles, including internal and external views/read/engagement. The quality of content and presentation has minimal impact. This might sound shocking, but 99% of a story’s success depends on whether Medium’s human curators like it and add it to the distribution system. If they do, large publications like Illumination work wonders and can make a story go viral. If they don’t, no matter how impactful, memorable, or well-written the post is—even if it comes from a distinguished professional—it will struggle to surpass 100 views from direct links.
But then you and Dr. Broadly both speak out against using AI curation... what would you perceive to be the best way forward? If human moderation is biased and AI curation is unable to perceive AI from human writing, what would be your envisioned solution?
>Meanwhile, another writer with more followers and substantial qualifications, who wrote about the same topic from a scientific perspective—making it more accessible and appealing to the public—received only 22 views on the same publication through direct links. Yes, this is the writer you characterize as "a crying wolf." Checkout his story to gain insights: https://medium.com/p/98166f12a488
I agree that this isn't fair, but I also don't see how it's Medium's fault. Are they supposed to remember every story on a topic that gets published, to only pick the subjectively best one to Boost? Should they revoke the Boost from the other author to award it to Dr. Broadly?
Do you feel that the entire Boost program should be cut? I'm interested in your thoughts.
AI, although still in its infancy, is used by all of us in many facets of our lives. And its growing pains seem to be creating more problems than solutions. I cannot honestly answer the Question posed - Do you feel comfortable with your content being checked by AI detectors without your explicit consent? - with a simple: YES/NO/UNSURE. The question itself raises too many other more important questions. Like - If the detector finds some AI generated content in my article, does it know why I inserted it? - Does the detector know how AI helped me to improve the article? - For what reason are you using an AI detector (unreliable anyway) to scan my article? Ask me what you want to know and I'll tell you. It is irrelevant whether I feel COMFORTABLE or not. If someone wants to carry out a fruitless exercise, it's up to them, I can't stop them, so why should I care.
I will never support AI. As an artist - AI is unjustly damaging the art world. I do not use AI in any part of life. Not for art, not for copy. I’ve completely boycotted. I refuse to post of full art on any social platform that won’t let me opt out of AI training - that’s stealing and it’s an infringement on my copyright. AI is also destroying our ecosystem- it use SO much energy and water. This efforts of AI could be reallocated to fix hunger, homelessness, fix our educational system…legit ANYTHING other than what we are doing…
AI policy on Medium and blind faith in AI detectors has a lot to answer for, when AI can be totally hidden. The only outcome is that genuine writers will suffer. Who is benefitting from this loophole? Those who use AI and disguise it are! They think everyone is dumb enough to believe it's not AI, and that an AI detector is smarter.
It isn't surprising, although shocking, to learn that genuine articles aren't been distributed widely due to the use of AI detectors. Maybe writers will now stop submitting to publications known to use AI detectors. The catch being that the publication editors who do so, are adhering to Medium rules! It's a vicious cycle, and the victims are genuine writers.
To be honest, AI is here to stay and is revolutionizing the future of content creation, data analysis, marketing, and much more. It is up to creators to use it responsibly and ensure high-quality content that is fully guided by humans who have also created these tools. We should embrace it and use it ethically and responsibly, not abuse it. AI tools are your assistant, just like a human assistant and you are the delegator, manager, mentor, and guide. You should make it your own unique content with your brand and voice.
Just like humans can write poor quality, too much fluff, inaccurate, senseless, poorly formatted, poorly edited, plain bad content, so can AI... But who is to blame if AI creates substandard garbage? The human behind the steering wheel is to blame and should upskill themselves before using AI. You cannot just get into a Boeing, truck, yacht, car, or any vehicle without the necessary skills to control and drive it. AI tools are professional tools that need professionals in the cockpit to control it properly and safely.
Hey James, thank you for your kind words. That is true and AI is here to help make our lives and business easier, much easier. I forgot how to knit because I never continued practising it after my grandma taught me how to 😃... but I have a few other skills I can practice.
I have a paid subscription to GPTZero to monitor content on pubs I edit. Notoriously inaccurate. The same article can flag as 100% human one day and 100% AI another. With no changes to the article. There are other ways to flag AI but a human has to be educated and use the flags appropriately. Great article. And my Medium postings have dwindled to almost nothing since you left Medium. Leaving a big hole that AI cannot fill :)
Readers can easily smell AI content. Beginner writers like someone like me, who usually writes in schoolboy English, and orginality.ai gives a tight slap - "I already know this stuff, do you have any more neurons to make it better?". For this reason, I purchased an " orginality.ai " subscription. Writing becomes, it's like pulling off cactus fruits with the naked hand from the cactus plant. Camel lips don't get affected by their fiery sting attacks. I am not an such experienced writer. AI detector made life too hard for me on medium.
Isn't it ludicrous that Medium is attempting to completely block AI generated articles by using the very same tool - AI apps and detectors - and a very poor one at that!
I conducted an experiment last year on this same subject. I took 10 AI detector apps, including 4 paid options. I submitted 6 AI generated articles and 6 articles I had personally written some 30 years ago. For the AI generated articles, there was a mixture of answers, ranging from totally human to 80% AI. Strangely the self generated articles faired no better, from 100% human to 65% AI written, Even worse, when I submitted the same piece 3 times to the same detector, I got 3 different answers. I submitted the findings to Medium. The answer? nil, no reply! 2nd submission, ditto. I am a retired research scientist, used to producing academic papers. If I submitted an article to Medium in the proper academic manner, it would not pass their AI detection checks, because it would be flagged as "Too correct and mechanical"
I have been rejected as a writer for a publication because according to him, my articles are Al-generated. Hours of my own research, and years of writing on Medium, but they trust an Al detection tool. I will probably leave Medium for good.
I don't know how to solve this issue in regards to writing because by now the AI is really good at replicating human speech. The only thing that could work now as it worked before is trust. For example, I want to properly state and explain what parts of my content/story/image is created with the help of AI and which isn't. The reader must receive this info and then evaluate. I hate seeing Youtube inundated with thousands of AI videos without the video and its creator telling me it's AI. What this means is a stronger emphasis on the "Story behind the Story" and the artist/writer themselves. So it starts with us (or a system) properly vetting and stating what is and what isn't AI. Censorship won't work. Maybe some AI stuff is really good and worth reading. But I'd rather know what I'm reading and by whom.
Thanks for sharing Dr Yidiz, it's definitely something I have thought of as a reason for the reduction in views. I've noticed my earnings have dropped because my articles don't seem to reach the same audience they use to (pointing to a distribution problem).
I'm curious what the impact of tools like grammarly (that have heavy AI in them) have on my posts. Recently I've started using grammarly more, in line with the recent drop of earnings and lower distribution. I've also noticed that grammarly has moved beyond just word correction to whole paragraph rewriting.
I'm going to test not using it to see if that is flagging my work as "inauthentic" to see if my views increase.
While I've only ever published brief articles on Linkedin where they got good traction until some years back an algorithm change, it strikes me there is also another issue at play, and that's the formulas and formats that get rewarded.
It's not surprising there are so many false detections as many authors use specific formats depending on their topics - formats designed to charm algorithms.
I guess what I'm alluding to is that formats may cause false results.
Although I voted unsure if AI detection should require the consent of the author it is not that I have any faith in the accuracy of these tools.
It's almost impossible to escape AI being used in our articles. My experience has been that the following applications use it including:
Microsoft Word
Hemingway Editor
Grammarly (of course)
Microsoft Notepad
Google docs
I turn Grammarly off when I'm writing in Word, but MIcrosoft picks up (to some extent) what Grammarly would have told me about. It's a losing battle. :-(
I am currently using GPTZero as a light check against my writing. ChatGPT is the biggest offender. For example, even if it tells me that it is my voice and tone, I'll paste that into GPTZero, and it is immediately flagged as 100% AI-written.
As a test, I typed the statement, "I’m testing whether Google Doc uses AI checkers like Grammarly to help me catch errrrors. It not only caught it in Google Doc that error was misspelled, but it's being caught in my test sentence here. I have to assume that it's because I have Grammarly installed on my laptop. According to Seattle University ("Grammarly stores all text entered into its service on their servers. This means any high risk data you type could be at risk of being accessed or misused. Disabling Grammarly when working on sensitive documents ensures that this information remains secure.") (https://seattleuniversity.atlassian.net/servicedesk/customer/article/2002190356) (Or does it? What about Microsoft in Word, for example)
Using AI to make our lives easier is proving to be a challenge. Is there no escaping it?
Am I ranting? Yes, I'm sorry. I do that sometimes. AI Detection Tools are a vexing issue.
Thank you for this detailed review on what is happening on the Medium platform. It is very sad to see that a platform that worked so well before for the writers are now only earning cents...I hope that things will improve as I do really like the community. AI detectors do not work very well and many times can give false results...in my experience.
>For example, some stories, even from writers with sizable followings on a large pub like Illumination, with 189K followers, get under 10 views, usually from direct links shared by writers or editors.
Do you think that this issue is also, in some part, the problem?
Most people who subscribe to a publication only see a couple stories from that pub per day at most, through their Daily Digest (which seems to aim to provide a broad overview, instead of featuring lots of stories from one single pub). If ILLUMINATION is publishing dozens of stories per day, won't it be pretty unlikely that a reader will get recommended that story, or that they'll see it on ILLUMINATION's front page when it's quickly going to disappear under the constant stream of new articles?
I feel like this is another consequence of Medium not thinking too far ahead on how large publications work.
Hello! Your points trigger me and my response will be an article long. This is to show my respect and my good intention to inform authentic writers like you. This argument has been made since Illumination became a prominent publication in 2020. I am the owner of this pub, but I am also a consulting editor for many large—even larger—publications. The publication interface has almost no impact on views.
We analyzed over a million stories from multiple angles, including internal and external views/read/engagement. The quality of content and presentation has minimal impact. This might sound shocking, but 99% of a story’s success depends on whether Medium’s human curators like it and add it to the distribution system. If they do, large publications like Illumination work wonders and can make a story go viral. If they don’t, no matter how impactful, memorable, or well-written the post is—even if it comes from a distinguished professional—it will struggle to surpass 100 views from direct links.
For example, a new writer with a very small following wrote a story about "old person smell" on Illumination. The story received over 30,000 views overnight because Medium loved it. Meanwhile, another writer with more followers and substantial qualifications, who wrote about the same topic from a scientific perspective—making it more accessible and appealing to the public—received only 22 views on the same publication through direct links. Yes, this is the writer you characterize as "a crying wolf." Checkout his story to gain insights: https://medium.com/p/98166f12a488
Why is this happening? Because Medium selectively amplifies certain types of content from specific writers. I was a Boost nominator for 12 months, working with 30 senior editors who curated content for me. We followed the criteria Medium published—but Medium’s curators did not follow their own criteria. Stories that exceeded the stated requirements were rejected, while those that barely met them were boosted. Even stories that didn’t meet the criteria at all were organically boosted—and still are.
That was one of the reasons I resigned from the program and ultimately felt compelled to close my best publication. I link those stories here if you missed them:
https://medium.com/illumination/i-withdraw-from-the-pilot-boost-program-intentionally-to-create-better-outcomes-for-the-community-7096b0411fc7
https://medium.com/illumination/goodbye-beloved-illumination-curated-welcome-new-curated-newsletters-6cbfe5399d84
I was not the only one who left the Boost program. Several pub owners like Dr Giles also resigned for the same reason. This program, which looks great in theory, only supported 1% of good writers, disadvantaging 99%.
I am not an anti-Medium, far from itit. It is an excellent platform with a wonderful community. Therefore, I donated 40 hours of my precious time for its growth for 6 years. But it is no longer a platform supporting authentic writers therefore, I had to say goodbye until it became a great one again. https://medium.com/illumination/this-is-my-last-story-on-medium-until-the-algorithm-is-fairly-adjusted-fba50c38543c
It is not about the platform but the mismanagement of it, lack of transparency, and strategic approach. It is the reason so many great writers like Tim Denning stopped contributing, and some even deleted their stories and sme entirely left the platform. Now, those who can use AI and create sensational content and present them with humanizers in Grade 3 level shine.
Hi Dr. Yildiz, thanks for replying! And I hope that my comments here come across as earnest and open to engaging and figuring out what works best for all of us as a community moving forward, and not as an attack on you or Dr. Broadly in any way. I follow you both and do miss your insights on Medium.
When you say:
We analyzed over a million stories from multiple angles, including internal and external views/read/engagement. The quality of content and presentation has minimal impact. This might sound shocking, but 99% of a story’s success depends on whether Medium’s human curators like it and add it to the distribution system. If they do, large publications like Illumination work wonders and can make a story go viral. If they don’t, no matter how impactful, memorable, or well-written the post is—even if it comes from a distinguished professional—it will struggle to surpass 100 views from direct links.
But then you and Dr. Broadly both speak out against using AI curation... what would you perceive to be the best way forward? If human moderation is biased and AI curation is unable to perceive AI from human writing, what would be your envisioned solution?
>Meanwhile, another writer with more followers and substantial qualifications, who wrote about the same topic from a scientific perspective—making it more accessible and appealing to the public—received only 22 views on the same publication through direct links. Yes, this is the writer you characterize as "a crying wolf." Checkout his story to gain insights: https://medium.com/p/98166f12a488
I agree that this isn't fair, but I also don't see how it's Medium's fault. Are they supposed to remember every story on a topic that gets published, to only pick the subjectively best one to Boost? Should they revoke the Boost from the other author to award it to Dr. Broadly?
Do you feel that the entire Boost program should be cut? I'm interested in your thoughts.
You are so generous Dr Yildiz.
Amen! Could not agree more with you Mehmet!
AI, although still in its infancy, is used by all of us in many facets of our lives. And its growing pains seem to be creating more problems than solutions. I cannot honestly answer the Question posed - Do you feel comfortable with your content being checked by AI detectors without your explicit consent? - with a simple: YES/NO/UNSURE. The question itself raises too many other more important questions. Like - If the detector finds some AI generated content in my article, does it know why I inserted it? - Does the detector know how AI helped me to improve the article? - For what reason are you using an AI detector (unreliable anyway) to scan my article? Ask me what you want to know and I'll tell you. It is irrelevant whether I feel COMFORTABLE or not. If someone wants to carry out a fruitless exercise, it's up to them, I can't stop them, so why should I care.
I will never support AI. As an artist - AI is unjustly damaging the art world. I do not use AI in any part of life. Not for art, not for copy. I’ve completely boycotted. I refuse to post of full art on any social platform that won’t let me opt out of AI training - that’s stealing and it’s an infringement on my copyright. AI is also destroying our ecosystem- it use SO much energy and water. This efforts of AI could be reallocated to fix hunger, homelessness, fix our educational system…legit ANYTHING other than what we are doing…
AI policy on Medium and blind faith in AI detectors has a lot to answer for, when AI can be totally hidden. The only outcome is that genuine writers will suffer. Who is benefitting from this loophole? Those who use AI and disguise it are! They think everyone is dumb enough to believe it's not AI, and that an AI detector is smarter.
It isn't surprising, although shocking, to learn that genuine articles aren't been distributed widely due to the use of AI detectors. Maybe writers will now stop submitting to publications known to use AI detectors. The catch being that the publication editors who do so, are adhering to Medium rules! It's a vicious cycle, and the victims are genuine writers.
To be honest, AI is here to stay and is revolutionizing the future of content creation, data analysis, marketing, and much more. It is up to creators to use it responsibly and ensure high-quality content that is fully guided by humans who have also created these tools. We should embrace it and use it ethically and responsibly, not abuse it. AI tools are your assistant, just like a human assistant and you are the delegator, manager, mentor, and guide. You should make it your own unique content with your brand and voice.
Just like humans can write poor quality, too much fluff, inaccurate, senseless, poorly formatted, poorly edited, plain bad content, so can AI... But who is to blame if AI creates substandard garbage? The human behind the steering wheel is to blame and should upskill themselves before using AI. You cannot just get into a Boeing, truck, yacht, car, or any vehicle without the necessary skills to control and drive it. AI tools are professional tools that need professionals in the cockpit to control it properly and safely.
Great response, Hein. We'll never be free from AI paranoia unless we all get on with our knitting!!
Hey James, thank you for your kind words. That is true and AI is here to help make our lives and business easier, much easier. I forgot how to knit because I never continued practising it after my grandma taught me how to 😃... but I have a few other skills I can practice.
I have a paid subscription to GPTZero to monitor content on pubs I edit. Notoriously inaccurate. The same article can flag as 100% human one day and 100% AI another. With no changes to the article. There are other ways to flag AI but a human has to be educated and use the flags appropriately. Great article. And my Medium postings have dwindled to almost nothing since you left Medium. Leaving a big hole that AI cannot fill :)
Readers can easily smell AI content. Beginner writers like someone like me, who usually writes in schoolboy English, and orginality.ai gives a tight slap - "I already know this stuff, do you have any more neurons to make it better?". For this reason, I purchased an " orginality.ai " subscription. Writing becomes, it's like pulling off cactus fruits with the naked hand from the cactus plant. Camel lips don't get affected by their fiery sting attacks. I am not an such experienced writer. AI detector made life too hard for me on medium.
Is this the same for substack too?
Isn't it ludicrous that Medium is attempting to completely block AI generated articles by using the very same tool - AI apps and detectors - and a very poor one at that!
I conducted an experiment last year on this same subject. I took 10 AI detector apps, including 4 paid options. I submitted 6 AI generated articles and 6 articles I had personally written some 30 years ago. For the AI generated articles, there was a mixture of answers, ranging from totally human to 80% AI. Strangely the self generated articles faired no better, from 100% human to 65% AI written, Even worse, when I submitted the same piece 3 times to the same detector, I got 3 different answers. I submitted the findings to Medium. The answer? nil, no reply! 2nd submission, ditto. I am a retired research scientist, used to producing academic papers. If I submitted an article to Medium in the proper academic manner, it would not pass their AI detection checks, because it would be flagged as "Too correct and mechanical"
I have been rejected as a writer for a publication because according to him, my articles are Al-generated. Hours of my own research, and years of writing on Medium, but they trust an Al detection tool. I will probably leave Medium for good.
I don't know how to solve this issue in regards to writing because by now the AI is really good at replicating human speech. The only thing that could work now as it worked before is trust. For example, I want to properly state and explain what parts of my content/story/image is created with the help of AI and which isn't. The reader must receive this info and then evaluate. I hate seeing Youtube inundated with thousands of AI videos without the video and its creator telling me it's AI. What this means is a stronger emphasis on the "Story behind the Story" and the artist/writer themselves. So it starts with us (or a system) properly vetting and stating what is and what isn't AI. Censorship won't work. Maybe some AI stuff is really good and worth reading. But I'd rather know what I'm reading and by whom.
Thanks for sharing Dr Yidiz, it's definitely something I have thought of as a reason for the reduction in views. I've noticed my earnings have dropped because my articles don't seem to reach the same audience they use to (pointing to a distribution problem).
I'm curious what the impact of tools like grammarly (that have heavy AI in them) have on my posts. Recently I've started using grammarly more, in line with the recent drop of earnings and lower distribution. I've also noticed that grammarly has moved beyond just word correction to whole paragraph rewriting.
I'm going to test not using it to see if that is flagging my work as "inauthentic" to see if my views increase.
While I've only ever published brief articles on Linkedin where they got good traction until some years back an algorithm change, it strikes me there is also another issue at play, and that's the formulas and formats that get rewarded.
It's not surprising there are so many false detections as many authors use specific formats depending on their topics - formats designed to charm algorithms.
I guess what I'm alluding to is that formats may cause false results.
Although I voted unsure if AI detection should require the consent of the author it is not that I have any faith in the accuracy of these tools.