Partly. Google can reliably tell you whether an image was made with Google’s own AI tools: since November 2025 the Gemini app checks uploaded images for SynthID, Google’s invisible watermark. Google Search’s “About this image” can also show whether an image carries C2PA metadata saying it was created or edited with AI, whichever tool made it. What Google doesn’t offer is a general detector for unlabeled images from other generators, which is most of what people want to check.
What Google can check today #
| Google feature | What it checks | What it covers |
|---|---|---|
| Gemini app verification | SynthID watermark | Images made or edited with Google AI |
| SynthID Detector portal | SynthID watermark | Early testing with journalists and media professionals |
| “About this image” in Search, Lens, Circle to Search | C2PA metadata, when present | Any tool that adds C2PA AI labels |
| Google Photos details | Its own IPTC edit notes | Photos edited with Google AI tools |
| Visible Gemini watermark | A mark you can see | Images made or edited in the Gemini app |
Gemini and SynthID #
On November 20, 2025, Google added image verification to the Gemini app. Upload an image and ask something like “Was this created with Google AI?” and Gemini checks for SynthID (Google). Google says more than 20 billion pieces of content have been watermarked with SynthID, and that it plans to extend verification to video and audio and to more surfaces, such as Search. Google DeepMind’s SynthID page now describes checking images, video and audio clips in Gemini (Google DeepMind).
The key limit: this check only detects Google AI content. An image from Midjourney, Stable Diffusion, or most other generators won’t have SynthID, so a “no” tells you nothing about whether it’s real.
“About this image” and C2PA #
In September 2024, Google said that if an image contains C2PA metadata, people can use “About this image” in Search, Google Lens and Circle to Search to see whether it was created or edited with AI tools (Google). That works for any tool that writes C2PA AI labels, not just Google’s.
The limit here is different: it only works while the metadata survives. Most social platforms strip metadata on upload, and screenshots never carry it.
Google’s own labels on its images #
Google marks its own AI output in several ways. Images made or edited in the Gemini app carry a visible watermark plus SynthID (Google). Images from Nano Banana Pro (Gemini 3 Pro Image) in the Gemini app, Vertex AI and Google Ads also get C2PA metadata. Google Photos notes when a photo was edited with Google AI tools such as Magic Editor and Magic Eraser (Google).
What Google can’t tell you #
- Whether an unlabeled image from another generator is AI. Without SynthID or surviving C2PA metadata, Google’s tools have nothing definitive to read.
- Whether a real photo’s caption is true. Google Lens can find older copies of an image, which helps with miscaptioned photos, but that’s reverse search, not AI detection.
- Whether a chatbot’s opinion is reliable. If you ask a general chatbot whether an image “looks AI,” its answer is visual reasoning, not a calibrated detection score. The SynthID check is a different thing from asking for an opinion.
How to check an image fully #
Use Google for what it does well, and fill the gaps with other tools:
- Google Lens or Circle to Search: look for older copies of the image and any “About this image” information. See reverse image search vs AI detector.
- Gemini: if Google’s tools are a plausible source, ask Gemini whether the image was made with Google AI.
- A general AI detector: for everything else. Expose AI reads Content Credentials, XMP and generator tags (including Google Imagen and Gemini tags), then runs an on-device model trained on the fingerprints of many generators. You get Likely real, Uncertain or Likely AI-generated with a confidence meter, and the image never leaves your phone. It doesn’t read SynthID, so the Gemini check and the detector complement each other.
- Source and context checks: who posted it first, and does the scene fit the claimed time and place?
For step-by-step instructions on a phone, see checking a photo for AI on Android.
Why Google doesn’t offer a general detector #
Google hasn’t explained publicly why it favors watermarks and provenance over a public classifier for all AI images. The trade-offs are clear, though. Watermark checks are close to certain for the content they cover and produce very few false alarms. General classifiers cover everything but are probabilistic: they miss some AI images and flag some real ones, and at Google’s scale, even a small false-positive rate would mean mislabeling a huge number of genuine photos. How accurate AI image detectors are explains that base-rate problem. For how the two provenance systems compare, see SynthID vs C2PA.
Frequently asked questions #
Can Google Lens detect AI images? #
Google Lens finds visually similar images and pages. Through “About this image,” it can show AI information when an image carries C2PA metadata. It doesn’t give an AI verdict for images without that metadata.
Can Gemini tell if an image is AI-generated? #
Gemini can check whether an image was made or edited with Google AI by looking for SynthID. It can’t confirm images from other generators this way, so a negative answer isn’t proof the image is real.
Does Google label AI images in search results? #
Google has said “About this image” can show whether an image with C2PA metadata was created or edited with AI. Images whose metadata was stripped, or that never had any, won’t show that information.
What is SynthID? #
SynthID is Google DeepMind’s technology for watermarking and identifying AI-generated content. It embeds imperceptible signals in images, video, audio and text from Google’s AI tools, designed to survive common edits like cropping and compression.