---
type: source
id: diel-2024-deepfake-detection-meta
title: 'Human performance in detecting deepfakes: A systematic review and meta-analysis of 56 papers'
authors:
- Alexander Diel
- Tania Lalgi
- Isabel Carolin Schröter
- Karl F. MacDorman
- Martin Teufel
- Alexander Bäuerle
year: 2024
venue: Computers in Human Behavior Reports
volume: '16'
pages: '100538'
doi: 10.1016/j.chbr.2024.100538
peer_review: peer-reviewed-journal
evidence_tier: A
design: Systematic review and meta-analysis of human deepfake-detection studies (k = 137 effect sizes)
modality:
- image-face
- video
- audio
- text
stimuli: Heterogeneous across included studies
n_participants: 86155
n_note: 56 papers; 137 effect sizes
topics:
- human-detection
- meta-analysis
- modality
- training
summary: Pooled human deepfake-detection accuracy was 55.54% (95% CI 48.87-62.10), not significantly above chance; detection-improvement strategies raised it to 65.14%.
results:
- metric: Overall accuracy
  value: 55.54% [48.87, 62.10]
  note: k=67; CI includes 50%
  source_location: abstract
- metric: Audio
  value: 62.08% [38.23, 83.18]
  note: k=8
  source_location: abstract
- metric: Video
  value: 57.31% [47.80, 66.57]
  note: k=26
  source_location: abstract
- metric: Image
  value: 53.16% [42.12, 64.64]
  note: k=18
  source_location: abstract
- metric: Text
  value: 52.00% [37.42, 65.88]
  note: k=15
  source_location: abstract
- metric: Odds-ratio analysis
  value: OR 0.64 [0.52, 0.79], ≈39% detection
  note: k=62
  source_location: abstract
- metric: With improvement strategies (feedback training, AI support, caricaturization)
  value: 65.14% [55.21, 74.46]
  note: k=15; strongest for video
  source_location: abstract
- metric: Sensitivity (d')
  value: not significantly different from chance
  note: ''
  source_location: abstract
moderators:
- Detection-improvement strategies raised performance above chance, with strongest effects for video.
limitations:
- Heterogeneous paradigms and generators
- Many included studies use pre-2022 generators
- Wide CIs, especially audio
verification:
  status: verified-abstract
  confidence: high
  checked: 2026-10-07
  via: Semantic Scholar API abstract; DOI resolves to Elsevier PII S2451958824001714
url: null
pmid: null
pmcid: null
arxiv_id: null
openalex_id: null
oa_url: null
cited_in:
- MR-0001
- MR-0003
updated: '2026-10-07'
---

# Human performance in detecting deepfakes: A systematic review and meta-analysis of 56 papers

> **Source record** `diel-2024-deepfake-detection-meta` · 2024 · Computers in Human Behavior Reports  
> **Evidence tier:** A (peer-reviewed-journal) · **Verification:** verified-abstract (compiler confidence: high)  
> **Bottom line:** Pooled human deepfake-detection accuracy was 55.54% (95% CI 48.87-62.10), not significantly above chance; detection-improvement strategies raised it to 65.14%.

## Citation {#citation}

Alexander Diel, Tania Lalgi, Isabel Carolin Schröter, Karl F. MacDorman, Martin Teufel, Alexander Bäuerle (2024). Human performance in detecting deepfakes: A systematic review and meta-analysis of 56 papers. *Computers in Human Behavior Reports*, 16, 100538. https://doi.org/10.1016/j.chbr.2024.100538

Verification entry (identifiers, resolvers, claims with source locations): [/data/citations.json#diel-2024-deepfake-detection-meta](/data/citations.json#diel-2024-deepfake-detection-meta)

## Study design {#design}

| Field | Value |
|---|---|
| Design | Systematic review and meta-analysis of human deepfake-detection studies (k = 137 effect sizes) |
| Modality | image-face, video, audio, text |
| Stimuli | Heterogeneous across included studies |
| Participants (N) | 86155 |
| N detail | 56 papers; 137 effect sizes |
| Topics | human-detection, meta-analysis, modality, training |

## Results {#results}

| Metric | Value | Note |
|---|---|---|
| Overall accuracy | 55.54% [48.87, 62.10] | k=67; CI includes 50% |
| Audio | 62.08% [38.23, 83.18] | k=8 |
| Video | 57.31% [47.80, 66.57] | k=26 |
| Image | 53.16% [42.12, 64.64] | k=18 |
| Text | 52.00% [37.42, 65.88] | k=15 |
| Odds-ratio analysis | OR 0.64 [0.52, 0.79], ≈39% detection | k=62 |
| With improvement strategies (feedback training, AI support, caricaturization) | 65.14% [55.21, 74.46] | k=15; strongest for video |
| Sensitivity (d') | not significantly different from chance |  |

## Moderators {#moderators}

- Detection-improvement strategies raised performance above chance, with strongest effects for video.

## Limitations {#limitations}

- Heterogeneous paradigms and generators
- Many included studies use pre-2022 generators
- Wide CIs, especially audio

## Cited in {#cited-in}

- [MR-0001](/metareports/human-detection-accuracy.md): Human Detection of Deepfakes: Accuracy, Bias, Confidence, and Training
- [MR-0003](/metareports/voice-cloning-human-detection.md): Voice Clones and Synthetic Speech: How Well Do People Detect Them?

## Verification {#verification}

| Field | Value |
|---|---|
| Status | `verified-abstract` |
| Compiler confidence | high |
| Checked on | 2026-10-07 |
| Via | Semantic Scholar API abstract; DOI resolves to Elsevier PII S2451958824001714 |

Values on this page have not all been checked against the full text. Confirm them against the primary source before high-stakes use.