Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Repair Windows errors before they cause bigger problems3Scan for outdated or missing drivers - takes under a minuteiTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
People can sometimes spot manipulated media, but unaided judgment is not a dependable way to verify whether an online image, video, audio clip, or text is authentic. In a 2024 meta-analysis, average detection of deepfake material was not significantly above chance. That does not mean every person is fooled or that every task is equally difficult; it means confidence alone cannot establish that a piece of media is real.
What does “near chance” mean for spotting deepfakes?
Chance is the result expected from guessing under a particular test’s conditions. A result near chance means that, on average, participants did not reliably distinguish deepfakes from authentic material in those studies. It does not mean every participant scored exactly 50%, nor does it tell you whether a specific post is genuine.
Alexander Diel and co-authors’ 2024 systematic review and meta-analysis combined 137 effects from 56 papers involving 86,155 participants. For deepfake stimuli, pooled accuracy was 55.54% (95% confidence interval 48.87%–62.10%; 67 effects). Because that interval includes 50%, the estimate is not evidence of reliable above-chance performance. The authors also found that overall detection sensitivity was not significantly above chance. Read the 2024 review and meta-analysis.
Recommended Free Tools
These are pooled results across studies, not a prediction for every viewer, platform, or new synthetic technique. The test materials, tasks, and participants differed, so a single percentage should not be treated as a universal human-detection score.
#1 Best Overall
Does the type of media make a difference?
The same review reported different pooled accuracy estimates across deepfake formats. Each interval crossed chance, so the small differences between point estimates do not establish that one format is reliably easier to recognize.
| Deepfake format | Pooled accuracy | How to interpret it |
|---|---|---|
| Audio | 62.08% | The confidence interval crosses 50%; this is not evidence that listeners reliably detect deepfake audio. |
| Images | 53.16% | The confidence interval crosses 50%; the pooled estimate is close to chance. |
| Text | 52.00% | The confidence interval crosses 50%; the pooled estimate is close to chance. |
| Video | 57.31% | The confidence interval crosses 50%; this does not establish reliable detection. |
All four estimates are pooled results from Diel and co-authors’ 2024 review; the reported confidence intervals cross 50%. They summarize varied studies rather than a head-to-head test under identical conditions. A higher point estimate for one modality should not be read as a dependable advantage.
Rank #2
Can confidence tell you whether your judgment is right?
No. Feeling certain is a subjective assessment, not proof of authenticity. Accuracy measures whether a participant’s judgment matched the test material; confidence measures how sure that participant felt. They are different quantities, and a confident judgment can still be mistaken.
A 2025 study by Chen and Goh examines how confidence and individual traits relate to human deepfake-video detection accuracy, and calls for further investigation into miscalibrated confidence. Its findings do not justify a blanket claim that confidence is always unrelated to accuracy. They do support treating confidence as something to assess separately from correctness. Read Chen and Goh’s 2025 study.
Do training or AI tools improve detection?
Some tested strategies helped participants in study settings. In Diel and co-authors’ review, studies using approaches such as feedback training, AI support, and deepfake caricaturization had pooled accuracy of 65.14% (95% confidence interval 55.21%–74.46%; 15 effects). This result suggests that targeted assistance can improve performance in the tested settings; it does not show that a generic checklist, a short training session, or a particular tool can authenticate arbitrary content or keep working against every new technique.
Automated detector results need similar restraint. A 2024 ACM survey identifies transferability, interpretability, and robustness as challenges, and notes that reliable evidence is lacking for some real-world uses. A detector’s score is an output from a system, not a verdict that a clip is genuine or fake. Read Wang and co-authors’ 2024 survey.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What should you do when an online clip matters?
Research on detection accuracy does not establish a universal recipe for verifying every post. When the stakes are high, a prudent response is to seek corroborating context and provenance rather than relying on visual impressions or a detector score alone.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Quick Recap
Best Value
- Look for independent reporting or other reliable sources that confirm the event and its context.
- Check whether the original source, date, location, and publication context can be established.
- Distinguish whether the question is “Was this media synthetically altered?” from “Is the claim made with it true?” They are separate questions; evidence about deepfake detection does not by itself determine whether a claim is true.
- When evidence is incomplete, avoid sharing the item as confirmed fact.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
