Photos you post online can be collected by crawlers and used to train AI image and language models. Photographers, artists and ordinary people ask the same question: can I stop it? The honest answer is "not completely, but you can reduce it". This guide explains each practical step, what it actually does, and what it cannot do, so you do not rely on a measure that gives a false sense of safety.
Key Takeaway: Combine several small protections and never rely on one. Removing EXIF data protects your privacy (for example your location). It does not stop an AI company from training on the picture itself.
The 5 Layers at a Glance
| Layer | What it does | What it cannot do |
|---|---|---|
| robots.txt rules | Asks AI crawlers not to collect your site | It is a request. It does not force a crawler to obey, and it does not touch copies already collected. |
| Platform opt-outs | Limits use of your posts on that platform | Applies to that platform only, often going forward only, and the rules differ by country. |
| Metadata cleanup | Removes GPS, device and date details | Does not stop training on the picture. |
| Watermark and smaller copies | Signals ownership and deters casual copying | AI tools can remove watermarks, and low resolution is a common tip with little evidence behind it. |
| Glaze and Nightshade | Add changes meant to disrupt AI training or style copying | Not permanent, can alter flat-colour art, and researchers have reported ways to weaken them. |
1. Block AI Crawlers on Your Own Website
If you host your photos on your own site, your robots.txt file is the first line of defence. Three user agents matter most:
- GPTBot is OpenAI's crawler for content that may be used to train its models. OpenAI says disallowing it indicates your content should not be used for training.
- Google-Extended is a token that controls whether content Google crawls may be used for training future Gemini models and for grounding. Google states that it does not affect inclusion in Google Search and is not a ranking signal.
- CCBot is the crawler of Common Crawl, a large public web archive that many AI datasets draw from.
User-agent: GPTBot
Disallow: /
User-agent: Google-Extended
Disallow: /
User-agent: CCBot
Disallow: /
Two details to know. First, blocking GPTBot does not remove you from ChatGPT search: that is a separate crawler called OAI-SearchBot, and each setting is independent. If you want to appear in AI search answers, leave OAI-SearchBot allowed. Second, robots.txt is a convention. Well-behaved crawlers follow it and others may not, so it deters rather than blocks.
2. Use Platform Opt-Outs and Objection Forms
If you post on social platforms, check each one's privacy settings for an AI or data-sharing option, because they change often. For example, Meta has said it uses public posts and photos from adult accounts to improve its generative AI, and it offers an objection route for adults in the EU and UK. Reports say the objection applies going forward, does not remove data already used, and does not cover photos other people post of you. Outside those regions the options are narrower and not guaranteed.
Making an account private reduces what can be used in future, but it does not undo past public posts. Open the platform's own Privacy Center for the current steps rather than trusting an old screenshot.
3. Remove Private Metadata Before You Post
Photos can carry a GPS location, device model, date and sometimes your name. Removing it is good practice for family photos, home and school pictures, and anything posted publicly. Use the EXIF Remover: it shows every field first, removes the data in your browser and does not upload the image. The step-by-step guide is in how to remove EXIF data from a photo.
One trade-off: a metadata field called IPTC "Data Mining" lets a rights owner state whether an image may be used for AI training. It travels inside the file, so stripping all metadata also removes it. For a portfolio you want to keep that statement and your copyright line. For private pictures, remove everything. Even when it is present, it is a stated preference and does not grant or deny permission by itself.
4. Watermark and Post Smaller Copies
A visible watermark makes ownership clear and stops most casual reposting. Add one with the Watermark tool, and read how to add a watermark to photos for placement tips. Be realistic: AI tools can erase watermarks, so a watermark is a deterrent and proof of origin, not protection.
Posting a reduced-size copy with the Resize tool and keeping the full-resolution original offline is another common suggestion. It has a clear benefit for page speed and for stopping others from grabbing your best file. Whether it makes a photo less useful for AI training is a general claim with little evidence behind it, so treat it as a bonus.
5. Cloaking Tools for Artwork: Glaze and Nightshade
Glaze and Nightshade come from researchers at the University of Chicago and are free to download. Glaze adds small changes that people barely notice but that are meant to disrupt an AI model copying an artist's style. Nightshade is designed to corrupt training data. Their own developers describe them as deterrents and not a permanent solution.
- The changes can be visible on flat colours and smooth backgrounds.
- Security researchers have reported methods that weaken these protections, including blurring and a Cambridge proof of concept called LightShed. These findings come through press coverage, and protection may not hold as AI methods change.
- They are aimed at artists. For ordinary photos, steps 1 to 3 give better value.
What Does Not Work
- Removing EXIF alone as a defence against training. It only removes metadata.
- Lowering JPEG quality until the picture looks bad. You damage your own photo and gain little.
- A copyright notice in the page footer. It does not stop a crawler.
- Assuming "private" means safe. Anything you share publicly can be copied, and past copies stay copied.
Check and Prove What You Own
Keep your original files and their dates. If you worry that your work is in a public training dataset, the "Have I Been Trained" search lets you look through some open datasets by image or text. Legal questions about training on public images are still being decided in different countries, so for a serious dispute talk to a lawyer or a rights-monitoring service.
A Simple Routine
- Add the robots.txt rules above to your own site.
- Open the AI or privacy settings of each platform you post on and use any opt-out or objection available to you.
- Before posting a photo, remove its location data, resize it and add a watermark if it is your work.
- For art, decide whether Glaze or Nightshade fits, accepting the limits.
- Keep the originals and revisit your settings every few months.
Frequently Asked Questions
Does removing EXIF data stop AI from training on my photos?
No. Removing EXIF data protects your privacy, for example by hiding your GPS location, but it does not stop an AI company from using the picture itself. Use it together with the other steps.
Does blocking GPTBot remove my site from ChatGPT search?
No. GPTBot relates to training. ChatGPT search uses a separate crawler, OAI-SearchBot, and each setting is independent.
Does blocking Google-Extended hurt my Google Search ranking?
No. Google states that Google-Extended does not affect a site's inclusion in Google Search and is not used as a ranking signal.
Is robots.txt enough to stop AI scraping?
No. It asks crawlers to stay away. Well-behaved crawlers follow it and others may not, and it does not remove images that were already collected.
Can AI remove a watermark from my photo?
Yes, AI tools can often remove or inpaint visible watermarks. A watermark helps show ownership and deters casual copying, but it is not real protection.
Are Glaze and Nightshade free, and do they work?
Both are free tools from University of Chicago researchers. Their developers call them deterrents, not permanent solutions, and researchers have reported ways to weaken them, so do not rely on them alone.
Can I opt out of Instagram and Facebook using my photos for AI?
Meta offers an objection route for adults in the EU and UK, found in its Privacy Center. Elsewhere the options are narrower and not guaranteed, and an objection applies going forward. Check Meta's current page for the steps.
Summary: You cannot fully stop AI training on photos you share, but you can reduce it: block GPTBot, Google-Extended and CCBot on your own site, use platform opt-outs, remove private metadata with the EXIF Remover, watermark and downsize public copies, and use Glaze or Nightshade for art with eyes open about their limits.
Sources
- OpenAI: Crawlers and user agents (GPTBot, OAI-SearchBot)
- Google Search Central: Common crawlers (Google-Extended)
- Pixsy: How to protect your images from generative AI
- MIT Technology Review: Four ways to protect your art from AI
- The Register: Tech to protect images against AI scrapers can be beaten
- Meta: Building AI technology for the UK
Remove location data from your photos
Free, no signup, and your image stays in your browser.
Open the EXIF Remover →