---
type: source
id: eiserbeck-2023-deepfake-smiles
title: Deepfake smiles matter less—the psychological and neural impact of presumed AI-generated faces
authors:
- Anna Eiserbeck
- Martin Maier
- Julia Baum
- Rasha Abdel Rahman
year: 2023
venue: Scientific Reports
volume: '13'
pages: '16111'
doi: 10.1038/s41598-023-42802-x
peer_review: peer-reviewed-journal
evidence_tier: B
design: 'EEG (64 ch) belief-manipulation study: real face photographs labelled on screen as ''REAL'' or ''FAKE'' (cover story), valence rating on 100-point scale; label and emotion counterbalanced'
modality:
- image-face
stimuli: '540 real photographs: 180 identities × angry/neutral/happy (RaFD, KDEF, FACES databases); no image was actually synthetic'
n_participants: 30
n_note: 21 female, 9 male; mean age 25.9; 7 datasets excluded and replaced
topics:
- uncanny-valley
- neural
- labelling
- emotion
summary: Simply labelling a real smiling face as 'fake' made it rated less positive and removed its early neural smile effects (P1/N170/EPN), while increasing late, effortful processing (LPP); angry faces were unaffected by the label.
results:
- metric: Valence rating, happy faces
  value: fake 73.12 vs real 74.97 (0-100)
  note: label effect only for happy faces
  source_location: null
- metric: Rating RT, happy faces
  value: ≈1,366 ms fake vs ≈1,290 ms real
  note: slower for presumed fakes
  source_location: null
- metric: Overall valence
  value: fake 49.80 vs real 50.78
  note: small
  source_location: null
- metric: P1 / N170 / EPN smile effect (happy vs neutral)
  value: present for real-labelled only
  note: EPN for fake p=.066
  source_location: null
- metric: LPP smile effect
  value: present for fake-labelled only
  note: more positive LPP → more effortful evaluation
  source_location: null
- metric: Angry vs neutral
  value: N170, EPN, LPP effects for both labels
  note: negative expressions unaffected by label
  source_location: null
moderators: []
limitations:
- Label × emotion interaction significant behaviourally but not in ERPs (authors' caveat)
- Small N
- Belief manipulation only; no actual synthetic faces
mechanism_supported:
- top-down-belief
- dehumanization
verification:
  status: verified-fulltext
  confidence: high
  checked: 2026-10-07
  via: nature.com article page (open access); author list on page truncated
url: null
pmid: null
pmcid: null
arxiv_id: null
openalex_id: null
oa_url: null
cited_in:
- MR-0002
updated: '2026-10-07'
---

# Deepfake smiles matter less—the psychological and neural impact of presumed AI-generated faces

> **Source record** `eiserbeck-2023-deepfake-smiles` · 2023 · Scientific Reports  
> **Evidence tier:** B (peer-reviewed-journal) · **Verification:** verified-fulltext  
> **Bottom line:** Simply labelling a real smiling face as 'fake' made it rated less positive and removed its early neural smile effects (P1/N170/EPN), while increasing late, effortful processing (LPP); angry faces were unaffected by the label.

## Citation {#citation}

Anna Eiserbeck, Martin Maier, Julia Baum, Rasha Abdel Rahman (2023). Deepfake smiles matter less—the psychological and neural impact of presumed AI-generated faces. *Scientific Reports*, 13, 16111. https://doi.org/10.1038/s41598-023-42802-x

Verification entry (identifiers, resolvers, claims with source locations): [/data/citations.json#eiserbeck-2023-deepfake-smiles](/data/citations.json#eiserbeck-2023-deepfake-smiles)

## Study design {#design}

| Field | Value |
|---|---|
| Design | EEG (64 ch) belief-manipulation study: real face photographs labelled on screen as 'REAL' or 'FAKE' (cover story), valence rating on 100-point scale; label and emotion counterbalanced |
| Modality | image-face |
| Stimuli | 540 real photographs: 180 identities × angry/neutral/happy (RaFD, KDEF, FACES databases); no image was actually synthetic |
| Participants (N) | 30 |
| N detail | 21 female, 9 male; mean age 25.9; 7 datasets excluded and replaced |
| Topics | uncanny-valley, neural, labelling, emotion |

## Results {#results}

| Metric | Value | Note |
|---|---|---|
| Valence rating, happy faces | fake 73.12 vs real 74.97 (0-100) | label effect only for happy faces |
| Rating RT, happy faces | ≈1,366 ms fake vs ≈1,290 ms real | slower for presumed fakes |
| Overall valence | fake 49.80 vs real 50.78 | small |
| P1 / N170 / EPN smile effect (happy vs neutral) | present for real-labelled only | EPN for fake p=.066 |
| LPP smile effect | present for fake-labelled only | more positive LPP → more effortful evaluation |
| Angry vs neutral | N170, EPN, LPP effects for both labels | negative expressions unaffected by label |

## Mechanism relevance {#mechanism}

`top-down-belief`, `dehumanization`

## Limitations {#limitations}

- Label × emotion interaction significant behaviourally but not in ERPs (authors' caveat)
- Small N
- Belief manipulation only; no actual synthetic faces

## Cited in {#cited-in}

- [MR-0002](/metareports/uncanny-valley-and-synthetic-faces.md): The Uncanny Valley and Synthetic Media: Does It Exist, What Causes It, and Have Deepfakes Crossed It?

## Verification {#verification}

| Field | Value |
|---|---|
| Status | `verified-fulltext` |
| Compiler confidence | high |
| Checked on | 2026-10-07 |
| Via | nature.com article page (open access); author list on page truncated |