AI image detector checklist: use these 12 practical checks to verify a suspicious picture before trusting a detector score. An image may produce different results after it is cropped, compressed, edited, resized or captured as a screenshot, so the source, history, provenance and surrounding evidence matter too.
Last reviewed: July 24, 2026
Quick answer: This AI image detector checklist treats a detector result as a clue, not proof. Preserve the original image, investigate its source, run a reverse image search, review available Content Credentials or supported watermarks, compare multiple detectors and mark uncertain evidence as inconclusive.
Important: Never publicly accuse a person, creator, seller or business of deception based only on an AI detector percentage. High-stakes images require stronger independent evidence.
The purpose of this AI image detector checklist is not to promise perfect detection. It gives everyday users a repeatable process for answering questions such as “How can I tell if an image is AI-generated?” and “Can I trust this detector result?”
Can You Trust an AI Image Detector?
An AI image detector analyzes patterns that its system associates with generated images. The output may be a label, confidence score or percentage, but it remains an estimate produced by one model.
A detector may create a false positive by labeling an authentic photograph as AI-generated. It may also create a false negative by missing an AI-generated image. Performance can change according to the generator, detector training data, image subject and file quality.
A 2026 preprint evaluated 23 pretrained detector variants across 12 datasets containing about 2.6 million images from 291 generators. The researchers found no universal winner. The best detector averaged 75% accuracy across the benchmark, while several modern generators reduced average detector accuracy to approximately 18%–30%. Read the original AI-generated image detector benchmark.
Key point: A high score means the detector found patterns it associates with AI. It does not independently establish who created the picture, how it was edited or whether its caption is true.
AI Image Detector Checklist: 12 Checks Before You Trust a Result
Follow this AI image detector checklist in order whenever possible. Some investigations may reveal a clear source quickly. Others will remain uncertain even after every check.
Save the Original Image and Its Context
Begin with the highest-quality and most complete version available. Screenshots, reposts and cropped copies may lose metadata, credentials, watermarks or contextual information.
- Save the original post or page URL.
- Record the account, caption and publication date.
- Preserve surrounding comments or claims.
- Note whether the file is a screenshot, collage or edited copy.
Identify the Exact Claim
Define what the image is supposed to prove. A real photograph may be paired with a false caption, while an AI illustration may be harmless when clearly labeled.
- Is it presented as photography, AI art, illustration or satire?
- Does it claim to show a real person, product, place or event?
- Is a specific date or location provided?
- Could believing it affect money, safety or reputation?
Investigate the Original Uploader
Look beyond the pixels. A clear creator credit, transparent label and consistent account history may be more useful than a detector percentage.
- Review the account biography and previous posts.
- Look for a named photographer, artist or publication.
- Check whether the post links to an original source.
- See whether independent reliable accounts confirm the claim.
A new or anonymous account is not automatically deceptive. Account history is supporting evidence, not proof.
Run a Reverse Image Search
A reverse image search may uncover earlier copies, original sources and different captions. It does not directly classify the picture as AI-generated, but it helps establish its history.
- Search using the complete image.
- Repeat the search with a crop around the main subject.
- Review similar images as well as exact matches.
- Compare dates, captions, locations and creator credits.
Google explains that Lens may show similar images and websites containing the picture. Follow the official Google Lens reverse image search instructions.
Find the Earliest Available Version
When several copies appear, look for the earliest credible publication. Later versions may have been cropped, compressed or attached to a different story.
- Compare publication dates.
- Look for a higher-resolution or uncropped copy.
- Check whether the earliest page identifies the creator.
- See whether the picture predates the claimed event.
No earlier search result does not prove AI generation. A new, private or poorly indexed image may have no searchable history.
Check Content Credentials and Provenance
Content Credentials use the C2PA standard to provide cryptographically signed information about a digital file’s origin and history. Depending on the supported product, this may identify creation tools, edits or AI involvement.
- Look for available Content Credentials.
- Check who signed the information.
- Review listed creation and editing tools.
- Watch for invalid, missing or incomplete history.
- Do not treat missing credentials as proof of fakery.
C2PA states that provenance can describe an asset’s origin and history, but provenance alone cannot determine whether its content is true or factual. Read the official C2PA Content Credentials explainer.
This AI image detector checklist therefore combines provenance with context, reverse searching and independent confirmation.
Check Supported Invisible Watermarks
Some AI providers embed signals that their verification systems can recognize. These checks are useful for identifying content from a supported provider, but they are not universal AI detectors.
The OpenAI image verification tool checks for supported C2PA and SynthID signals associated with images created using OpenAI tools. OpenAI explains that a detected signal indicates likely OpenAI origin, but does not establish whether the picture is accurate or correctly presented.
If no OpenAI signal is found, the image could still have lost its metadata, contain a degraded watermark, come from an older model or have been created by another provider. OpenAI provides further detail in its C2PA and SynthID guidance.
Google’s Gemini verification feature uses SynthID for supported content created or edited by Google AI and may also display compatible Content Credentials. Review Google’s official AI-content verification guidance.
Test More Than One AI Image Detector
When the source and provenance checks do not settle the question, compare two or three reputable detectors using the same file.
- Use the cleanest available version in every tool.
- Record the detector name, result and date.
- Read how each provider defines its score.
- Do not mechanically average unrelated percentages.
- Do not select only the result that confirms your suspicion.
- Review privacy terms before uploading sensitive images.
| Check | Example Result | Safe Interpretation |
|---|---|---|
| Detector A | Likely AI | The system found patterns associated with generated images. |
| Detector B | Uncertain | The evidence did not support a confident classification. |
| Provider verification | No supported signal | No signal from that supported provider was detected. |
The AI image detector checklist does not require every tool to agree. Conflicting results should increase caution rather than encourage a forced conclusion.
Inspect Visual and Physical Consistency
Zoom in and examine whether the scene behaves consistently. Visual clues are useful, but a convincing image may contain none of the classic errors.
- Compare lighting, reflections and shadow direction.
- Inspect signs, labels and background text.
- Review hands, teeth, jewelry and small accessories.
- Look for repeated textures or duplicated objects.
- Check perspective, scale and object boundaries.
- Ask whether objects interact in physically possible ways.
Avoid outdated claims such as “AI images always have bad hands.” Authentic photos may contain blur or editing artifacts, while modern generators can create convincing anatomy.
Account for Editing, Screenshots and Compression
Social platforms, messaging services and editing apps can alter the file before you receive it. Resizing, recompression, sharpening, filters and screenshots may change detector results or remove useful metadata.
- Prefer the highest-resolution and least edited copy.
- Test the original and screenshot separately when available.
- Consider whether only part of the picture was AI-edited.
- Do not apply one verdict to every element in a collage.
Seek Independent Confirmation
If the image claims to document a real event, product, person or location, look for evidence that does not trace back to the same original post.
- Check reputable news or fact-checking organizations.
- Review official statements from relevant authorities.
- Search for photographs from different angles.
- Compare landmarks, weather, timing and product details.
- Contact the relevant organization when the stakes justify it.
Record an Honest Final Verdict
Finish the AI image detector checklist with the conclusion that best matches all available evidence. Do not force a simple “real or fake” answer.
Inconclusive is a responsible answer. Stating uncertainty is safer than making an unsupported accusation.
How to Interpret an AI Detector Percentage
This AI image detector checklist treats every percentage as a tool-specific output. Different providers may calculate and describe their scores differently.
| Output | Safer Interpretation | Unsafe Interpretation |
|---|---|---|
| 95% AI | Strong AI-associated patterns were found by that detector. | The image is definitely fake. |
| 52% AI | The result is uncertain or near the detector’s threshold. | The image is exactly 52% fake. |
| 5% AI | Limited AI-associated evidence was found in this file. | The image has been proven authentic. |
| No supported signal | A provider-specific signal was not detected. | No AI system was involved. |
Why Do AI Image Detectors Disagree?
Detectors may use different training datasets, classification thresholds, generator coverage and image-processing methods. One tool may perform well on portraits but struggle with illustrations, screenshots or images from a newly released generator.
Common causes of disagreement include:
- different training images and generator coverage;
- different thresholds for labeling content;
- resizing, compression or format conversion;
- partial AI editing rather than full generation;
- new generators that were absent from training data;
- different performance across portraits, landscapes and graphics.
Best practice: When tools disagree, return to the source, provenance and independent context instead of testing detectors until one supports your preferred answer.
Three Practical AI Image Verification Examples
The following examples show how to apply the AI image detector checklist to everyday situations.
Example 1: Suspicious Product Photography
- Reverse-search the product picture.
- Compare it with the manufacturer’s official images.
- Check labels, accessories, dimensions and packaging.
- Review seller history and customer-uploaded photographs.
- Ask for a new photograph showing the actual product.
- Do not label the seller fraudulent from a detector score alone.
Example 2: A Viral News or Disaster Image
- Identify the claimed date and location.
- Search for older uses of the picture.
- Check reputable local reporting and official alerts.
- Compare buildings, weather and lighting.
- Look for independent images from other angles.
- Do not repost while the evidence remains uncertain.
Example 3: A Dating or Social-Media Profile Picture
- Reverse-search several profile images.
- Review account history and inconsistent details.
- Request a live video conversation.
- Never send money or identity documents based on an online relationship.
- Report suspicious behavior privately through the platform.
When children or teens are interacting with highly human-like AI services, our AI companion chatbot safety guide explains related trust, privacy and emotional-dependence risks.
Common AI Image Detection Mistakes
Use the AI image detector checklist to avoid these common errors:
- Trusting one detector: One result cannot replace source verification.
- Treating a score as certainty: Model output is not independent proof.
- Checking only hands or text: Modern images may avoid obvious errors.
- Ignoring the caption: A real photograph may be presented with a false story.
- Assuming missing metadata proves AI use: Screenshots and uploads often remove metadata.
- Assuming metadata proves truth: Provenance does not validate every claim depicted.
- Uploading private images carelessly: Check privacy and retention terms first.
- Making public accusations: Incorrect allegations may cause serious harm.
- Ignoring partial edits: A real photograph may include an AI-generated object or background.
What to Do When an Image Is High-Stakes
Take additional precautions when an image may affect money, identity, safety, employment, insurance, medical decisions, legal claims, school discipline, reporting or someone’s reputation.
High-stakes rule: Pause before sharing or acting. Preserve the original file, document its source and contact the relevant organization, authority, platform or qualified professional. A public AI detector should not decide a serious case by itself.
AI Image Detector Checklist Worksheet
Copy this shortened AI image detector checklist into your notes when investigating a suspicious picture:
| Check | What to Record | Result |
|---|---|---|
| Source and context | Account, URL, caption, date and original claim | Reliable / Weak / Unknown |
| Reverse search | Earlier versions, creators and different captions | Found / Not found |
| Content Credentials | Signer, creation tool and edit history | Present / Missing / Invalid |
| Watermark verification | Provider and exact result | Detected / Not detected / Unclear |
| Detector comparison | Tool names, scores and dates | Consistent / Conflicting |
| Independent confirmation | Reliable reports or evidence from other sources | Confirmed / Unconfirmed |
| Final verdict | Likely AI-generated / Likely authentic / Manipulated or out of context / Inconclusive | |
Frequently Asked Questions
Can an AI image detector be trusted?
It can provide useful supporting evidence, but it should not be treated as conclusive proof. Results depend on the detector, generator, image type, file quality and any editing or compression.
What does “90% AI-generated” mean?
It generally means the detector found patterns strongly associated with AI according to its model. It does not necessarily mean there is an independently established 90% probability that the image is fake.
Can a real photograph be flagged as AI-generated?
Yes. False positives may occur with heavily edited, compressed, unusual or synthetic-looking photographs. Check the original source and surrounding evidence before reaching a conclusion.
Can an AI-generated image pass a detector?
Yes. A detector may not recognize an unfamiliar generator, and cropping, screenshots, editing or compression may weaken the patterns it analyzes.
Why do AI image detectors give different answers?
They use different training datasets, methods, thresholds and generator coverage. Conflicting results should be recorded as uncertainty rather than combined into an invented average.
Does missing metadata mean an image is AI-generated?
No. Metadata may disappear when a file is edited, downloaded, screenshotted, sent through messaging services or uploaded to a social platform.
Can reverse image search detect AI-generated images?
Reverse image search does not directly classify AI content. It helps find earlier versions, original sources, similar pictures and pages using the same image.
What should I do when the evidence is uncertain?
Mark the result inconclusive, avoid public accusations and seek additional evidence before sharing or acting. Preserve the original file when the issue is important.
Can OpenAI verify whether an image came from ChatGPT?
OpenAI’s verification tool checks for supported provenance signals associated with images generated by OpenAI tools. No detected signal does not prove that the image is authentic or that no AI was involved.
How should I use this AI image detector checklist?
Start with the original source, complete all available verification steps and select the verdict supported by the combined evidence. Do not rely on the detector percentage alone.
Final Takeaway
Use this AI image detector checklist whenever a suspicious picture could influence what you share, purchase, report or believe. Start with the source and claim, find earlier versions, review available provenance information, compare detector results and seek independent confirmation.
The safest conclusion is not always “real” or “AI.” When the evidence is missing or contradictory, mark the result as inconclusive and avoid making an accusation.
Sources and editorial note: This guide was manually reviewed using current official documentation from OpenAI, Google and C2PA, together with a February 2026 detector-benchmark preprint. Tool capabilities and interfaces may change, so review each provider’s current documentation before making an important decision.
Check the Evidence, Not Just the Percentage
Save this AI image detector checklist and use all 12 checks before sharing, reporting or acting on a suspicious picture. A detector result is supporting evidence—not final proof.
