A photo of a famous investor “endorsing” a get-rich-quick app. A video of a CEO announcing an urgent bank transfer. A viral picture of a disaster that never happened. A glowing product review that reads a little too perfectly.
In 2026, AI can create images, voices, videos and text that look and sound real. Researchers have found that even people, and many detection tools, struggle to tell the best AI images from real photos. So the old advice, “just look for weird hands”, is no longer enough.
This guide shows what still works: visual clues, context checks, free tools and the new “content credentials” systems that platforms are rolling out.
Quick Answer: A 6-Step Check
- Pause. Strong emotion and urgency are red flags.
- Look closely for visual and audio glitches (but don’t rely on this alone).
- Check the source. Who posted it first, and can you trust them?
- Run a reverse image search to find the original and earlier versions.
- Check provenance tools, such as Content Credentials and SynthID checkers.
- Confirm with trusted news outlets or fact-checkers before you share or pay.
Why It Is Harder Than Ever
- Quality has improved. Many telltale errors, like extra fingers, are now rare in newer tools.
- Detectors aren’t perfect. AI detection tools can give false alarms and miss fakes, so don’t treat any one score as the final answer.
- Labels can disappear. Screenshots, re-uploads and compression can strip information that marks an image as AI-generated.
- Not every tool adds marks. Some generators, including open-source ones that run on personal computers, add no watermark at all.
The practical lesson: combine several checks and focus on context, not only on pixels.
Step 1: Look for Visual Clues (Helpful, But Not Proof)
These signs can still appear, especially in older or cheaper AI tools.
In images:
- Hands, teeth and ears that look slightly off.
- Text and logos that are garbled, misspelled or inconsistent. Sign text and small print are common weak spots.
- Lighting and shadows that don’t match across the scene.
- Reflections in glasses, mirrors or water that don’t make sense.
- Skin that looks too smooth, with few pores or blemishes.
- Backgrounds with warped lines, strange objects or repeating patterns.
- Jewellery, glasses and clothing details that merge or change shape.
In videos:
- Lips that don’t match the words.
- Unnatural blinking, facial expressions or head movement.
- A face that blurs at the edges, hairline or neck.
- Odd lighting on the face compared with the rest of the scene.
In audio:
- A flat or overly even tone, strange pauses or odd breathing.
- Background noise that cuts in and out unnaturally.
A caution: The absence of these clues doesn’t prove something is real. Many new images have none of them.
Step 2: Check the Context
This is often the most reliable step.
- Who shared it? Check the account’s age, history and followers. New accounts with lots of sensational posts are suspicious.
- Is there an original source? Real events usually have several independent reports, photos and witnesses.
- Does it match the facts? Check the date, location, weather, uniforms, signs and language.
- Does it seem designed to make you angry, scared or excited? Fake content often pushes strong emotion.
- Is it only on social media? If major news outlets aren’t covering a “huge” event, be cautious.
Step 3: Use Reverse Image Search
A reverse image search shows where else an image has appeared.
- Google Lens or Google Images: Upload the image or paste its link.
- TinEye: Good for finding older copies.
- Bing Visual Search is another option.
Look for the earliest version, a different caption or an original photo that has been edited. If the “news photo” matches an old event, an artwork or a stock image, you have your answer.
Step 4: Check Content Credentials and Watermarks
Two main systems now help show whether something was made with AI.
| C2PA Content Credentials | SynthID | |
|---|---|---|
| What it is | Signed information attached to a file, showing the tool, date and edits | An invisible watermark embedded in the content itself |
| What it tells you | Where the file came from and how it was edited | That the content likely came from a tool that uses SynthID |
| Can it be lost? | Yes. Screenshots and some uploads can strip it | It can survive some edits, but not all, and heavy edits may remove it |
| How to check | A Content Credentials viewer, such as verify.contentauthenticity.org | Google’s SynthID checking in the Gemini app, or OpenAI’s Verify tool |
Some AI companies now run public checkers. For example, OpenAI offers a Verify tool (openai.com/verify) that looks for its own provenance signals, including Content Credentials and SynthID, in supported images and audio files.
The big limit: These tools work only for content from tools that add the signals. If a check finds nothing, that does not mean the image is real. It may simply have come from a tool without a watermark, or the signal may have been stripped.
Step 5: Look for Platform Labels
Many platforms, such as Meta’s apps, TikTok, YouTube and X, add “AI-generated” or “altered content” labels in some cases. These are useful hints, but they can be missing or wrong, especially after content is downloaded and re-uploaded.
Step 6: Verify With Fact-Checkers and Trusted Sources
- Search for the claim, not just the image.
- Check established fact-checking sites such as Snopes, AFP Fact Check, Reuters Fact Check, or regional ones like BOOM and Alt News in India.
- Look for coverage by several independent, reputable news outlets.
How to Spot Other Types of Fake Content
Fake screenshots
Screenshots of posts, chats or news headlines are easy to fake. Check the original account or website directly, and don’t trust a screenshot as proof.
Deepfake voice calls
Criminals can clone a voice from short audio clips. If a relative or boss calls with an urgent money request, hang up and call back on a known number. Many families agree on a private “safe word”.
Fake reviews
Watch for lots of five-star reviews posted at the same time, vague wording, repeated phrases or reviewers with only one review. Check several sources, not just the seller’s website.
Fake news sites
Check the web address carefully, the “About” page, named authors and whether other trusted outlets report the same story.
AI-written text
AI text can read smoothly but may contain wrong facts, made-up sources or generic phrases. Check names, numbers and quotes yourself. Text-detection tools are unreliable, so don’t judge a person or student by a detector score alone.
Fake documents
AI can create realistic-looking invoices, receipts, IDs and bank statements. Verify through official channels, not through a file or link sent to you.
Why This Matters for Your Money
Fake content is often used to take money, not just to mislead.
- Deepfake endorsements: Fake videos of celebrities or experts promoting investments, crypto or “trading bots”.
- Executive impersonation: A fake video or voice call that appears to come from your boss, asking for an urgent payment.
- Romance and friendship scams: AI-generated profile photos and chats.
- Fake shops and fake reviews: Realistic product pictures and glowing reviews for products that don’t exist.
- Fake invoices and payment requests.
Simple protection rule: Never move money based on a message, video or call alone. Confirm through a second, trusted channel, such as a phone number you already have.
For more, read our guides on [15 Common Online Scams and How to Avoid Them] and [How to Keep Your Social Media Accounts Secure].
Tips for Sharing Safely
- Pause before sharing, especially breaking news or shocking images.
- Add context, such as the source, when you share.
- Correct mistakes if you shared something false.
- Report fake accounts and misleading content to the platform.
What If You Create AI Content Yourself?
- Label AI-generated images clearly, especially if they look realistic.
- Keep Content Credentials turned on if your tool supports it.
- Don’t use AI to impersonate real people or to mislead others.
Frequently Asked Questions
How can I tell if an image is AI-generated? Use several checks together: look for visual clues, check the source, run a reverse image search and use provenance tools such as Content Credentials and SynthID checkers. No single check is enough.
Are AI image detectors accurate? Not reliably. They can miss new fakes and flag real photos. Use them as one clue among many.
What is a Content Credential? It is signed information attached to a file that shows where it came from and how it was edited. It follows an open standard from the C2PA group.
What is SynthID? SynthID is an invisible watermark that Google DeepMind and some partners embed in AI-generated content, so special tools can detect it.
If a checker finds no AI watermark, is the image real? No. Many AI tools add no watermark, and signals can be removed, so a missing watermark proves nothing.
Can I trust a screenshot? Not by itself. Screenshots are easy to fake and also strip provenance information.
What should I do if I spot a deepfake? Don’t share it, report it to the platform and, if it involves fraud or harassment, report it to the relevant authorities in your country.
Final Thoughts
No detector can tell you with certainty whether something is real, and that is why habits matter more than tools. Slow down, check the source, search for the original, use provenance tools and confirm with trusted outlets. And when money or personal details are involved, always verify through a second channel before you act.
Sources and Further Reading
- Coalition for Content Provenance and Authenticity (c2pa.org) and Content Credentials (contentauthenticity.org)
- OpenAI Help Center: Provenance signals in OpenAI-generated content
- Google DeepMind: SynthID
- Fact-checking networks: Snopes, AFP Fact Check, Reuters Fact Check