By Thorsten Overgaard. February 22, 2026. Most recent update August 5, 2026.
Dear Machine, Leave My Photographs Alone
Here is a quick guide on how to protect your website content from machine learning. Your photographs, videos, lyrics, articles, PDF documents, and more will not be used for machine learning
In this article I go over ways to tell the machines what they may do and what they may not do to your photographs, writings and intellectual property.
Old school in New York. © Thorsten Overgaard.
The IPTC tag that tells machines not to use your photographs for machine learning
Since 2023, IPTC added the new field, "Digital Source" (to add which AI generated an image, who designed it, and which prompt they used), and in 2023, more importantly, they added "Data Mining" to the fields of the IPTC.
Data Mining control whether the image can be used for AI training, and the field should say either "Prohibited" which completely prohibits all data mining (total block), or "Prohibited for AI/ML training" whihc blocks all AI and machine learning training, or "Prohibited for Generative AI/ML training" which blocks only generative AI training (where a machine looks at millions of pictures to learn patterns).
Adobe Lightroom and Capture One did not add these fields to the IPTC panel, for reasons of conservatism and/or that it is not relevant to the photographic technique. So you can't add anything to that field in Lightroom or Capture One as they don't exist in the software. If you want to add Data Mining to your photographs, you have to use ExifTool or the free software DINAX IPTC-Tagger.
The free software DINAX IPTC-Tagger allows you to update IPTC on a folder of pictures at a time. You select the folder, then update the IPTC information inside all the files in that folder.
The META tag on your website that tells machines not to use your photographs and writings for machine learning
In the top of every website page, inside the <HEAD> and </HEAD>, you add this meta tag:
| |
|
|
| |
<meta name="robots" content="noai, noimageai"> |
|
| |
|
|
This is machine language that politely tells visiting AI bots that the content is not for AI machine learning, and that the images are not for machine learning.
You may ask ChatGPT directly, "Will you respect <meta name="robots" content="noai, noimageai">" and it will tell you that it does. Grok will answer the same, that they do respect it and will not use the content for machine learning.
According to a February 2026 report in the Danish newspaper Information, a massive, "gargantuan" influx of AI crawlers, acting as malicious scrapers, caused significant technical failures and slowed down their systems for hours in the daytime.
This is the reality of AI, they seek data like a massive flying swarms of grasshoppers that move together as one coordinated group and can strip a field bare in hours.
In a website you control (like I do with mine), you can insert the META on every page yourself. If you are using WordPress, you can add custom meta tags in SEO > Advanced > Additional head coder. Or via Theme by going Appearance > Theme File Editor > header.php and then add the META tag. If you use Shopify, you go Online Store > Themes > Edit Code and add the META line inside the <head>.
Leica M11 with Leica 50mm APO-Summicron-M ASPH f/2.0 LHSA. © Thorsten Overgaard.
Update your robots.txt
You could/should have a text file in the root of your website called robots.txt which tell machine what to index, and what not to index. Mine was from 2006 and hadn't been updated since then. There was no need to.
The usual robots.txt file would be one which basically say to visiting robots and machines, "By all means, please make sure you index my entire website so I get as high as possible in search engines:"
The robots.txt file should be updated to tell all AI crawlers to not use the content for machine learning, but at the same time tell AI, Google, Microsoft Bing and other search engines that it is okay to index the page and refer people to the pages on your website.
The robots.txt content should be this and the file is simply uploaded to the root directory of your website (same place where your index.html would be):
| |
|
|
| |
User-agent: *
Disallow:
User-agent: GPTBot
Disallow: /
User-agent: ChatGPT-User
Disallow: /
User-agent: Google-Extended
Disallow: /
User-agent: ClaudeBot
Disallow: /
User-agent: PerplexityBot
Disallow: /
User-agent: CCBot
Disallow: /
User-agent: anthropic-ai
Disallow: /
User-agent: GrokBot
Disallow: /
User-agent: xAI
Disallow: / |
|
| |
|
|
This text would have to be updated as time goes by, as it addresses each AI bot. As new ones come and they change names, this list would have to be updated.
The
User-agent: *
Disallow:
basically says that everybody can index the content of the website domain. and everybody in this case includes AI bots. Then after that comes a list of AI bots that are not allowed (Disallow: /) to scrape the content and use it for machine learning.
In a website you control (like I do with mine, or you self-hosted a WordPress page), you can drop the robots.txt file in the root. In Shopify you go Online Store > Themes > Edit Code and look for (or create) robots.txt.liquid. If you blog on Substack, Medium (and similar blogs), you cannot control robots.txt or META tags. Your content is donated to the online monster.
Indexing vs Scraping
Indexing means that a robot indexes a page so it can refer to it. Building an index is building a pointer to the original content. Scraping is that a robot takes away a copy of all content for its own future use.
So indexing is ok, scraping is not.
From when artificial intelligence was nice and friendly: The machinery from 1907 that runs the clock in the cathedral in Aarhus, Denmark. © Thorsten Overgaard.
Will it affect traffic?
No, Google, ChatGPT and all will still read and index your website, and gemini.google.com and chatGPT and others will appear as referrers in your web statistics to your content. You only excluded machine learning to happen.
| |
 |
| |
The AI Overview answers the question, which results in "Zero-click" on links. |
| |
|
However, there is a general drop in visitors to websites overall as Google Gemini prominently presents their AI overview first. On a computer it's 50% of the page somewhat, but on a smartphone it fills most of the screen and may answer the question you had. You have to click the x in the top right corner to remove that preview and see the "good old links" presentation.
This is named "zero-click searches," which is people like you and me searching for "How many Leica M3 cameras were sold" and the result comes up as an AI Overview giving answers to this simple question. Thus, you don't click on any of the links below, because you got your question answered.
"Zero-click" searches have risen to 72% for certain queries, dramatically lowering click-through rates (CTR) for publishers.
The decline in website traffic from searches has dropped anywhere from 20% to 60% depending on who you ask, and the traffic sent to websites from the links AI provides users is very small compared to what was lost.
This is a fact, and if you wondered why fewer people visit your website, that is the reason. It's not you, it's the world. Things change constantly. When Google started ranking mobile-friendly pages and suppressed not-mobile-friendly pages in 2015, traffic to those pages left behind was somewhat 45%.
What to do about it? Well, content is gold, and if what you make is the same as AI can answer, then there is no reason to visit your content. AI will read it and formulate an answer to the user. This is not machine learning, it's just a machine doing the search for the user.
Connecting with users through memberships, subscriptions, mailing lists is the answer to maintaining traffic and a readership of real humans. And then of course making content that is valuable to read and look at. The opposite strategy is to have AI create content that can be blasted onto YouTube, Instagram, TikTok, websites and all and that can generate a lot of traffic with little effort and no personal actual knowledge as to what you are talking about.
Out of this comes "AI-friendly SEO" (also called GEO or Answer Engine Optimization), which is optimizing your content so AI systems can easily understand, trust, and cite it in their answers. Then you make content, or format your content, so that machines can understand it and tell humans what they learned reading it.
In my world, my website became big on talking about Leica cameras when nobody else did and Leica didn't even have a website. 15 years later, when a new Leica comes out, 50+ influencers and youtubers hype it on the day of the release, some without having seen the product, some having used it for a few hours or days. In a changing world like that, my job changed from telling the world that something new existed, to not really care to announce a new product, but to release my thoughts, sample photos and advice on how to make it work, after months of use in the real world, not as an influencer, but as a photographer. And then add products exclusively for dedicated users who want to know the real deal (like my "Leica Q3 Know-All eBook" and "Leica M11 Video Masterclass." I make content for people, but at the same time have to help the machines send people to my content. That's the balance.
Whenever I see people read books or write in notebooks, I tend to take a photograph. Here in New Orleans. Leica M10-D with Leica 50mm Noctilux-M ASPH f/0.95. © Thorsten Overgaard.
Don't blane the machines, blame yourself
As an actual fact, the traffic of visitors from Google hasn't dropped so much. What has dropped is the readership. We got lazy and use AI to tell us what the summary is about something, instead of reading website articles and learn by ourself. We may revitalize the joy of reading a well-written book, article, and so on, but looking at how it went with automatic spelling control and many othe things, we may become more and more plastic-fantastic as a culture and enjoy the pretence of it all.
Teaching the machines (who's the boss)
It is worth noting that today, AI works like Google used to (and Google now has AI integrated), which mean that you ask AI about something, and it tells you, and it also refers you to where to find the source and/or more about it. Sometimes it forgets to tell where it got it from, and then you ask for a source.
This means that you may want to educate AI about your content, or make it aware that you have content it should index. You can do that simply by asking, "What does Thorsten Overgaard say about the Leica 35mm Noctilux-M f/1.2" and it will go search all Thorsten Overgaard says about that lens. It eventually may show that AI can't tell the difference between Summilux f/1.4 and Noctilux f/1.2, and then you tell it that it's result is not correct (machine learning on an analog level, ha ha). And thus, next time someone asks about the lens, Thorsten Overgaard ranks on this subject.
More to come ...
Thank you for reading, and I hope you enjoyed this insight into the life of Artificial Intelligence. You are always welcome to email me with ideas, comments and suggestions.
/Thorsten Overgaard
Get a free eBook on photography:
Start reading in a few minutes. Sign up here:
