sparklessszzz/InstaArt-HumanAI

Instagram AI Art vs Human Art: Engagement & Comment Dataset

Dataset Summary

This dataset was created and contributed by Akshaya, Cynthia, Grace, and Soham as part of a project at UC San Diego.

This dataset supports research into how audiences engage with AI-generated art versus human-made art on Instagram, with a specific focus on comment sentiment, reaction types, and engagement patterns. It consists of 40 matched pairs of Instagram posts - one human art post and one AI-disclosed art post per pair - along with up to 50 top comments per post, their timestamps, and full engagement metrics (likes, comment count, reposts, shares).

All posts were collected from a fresh Instagram account with no prior following history, created specifically for this study to avoid algorithmic bias from existing follower networks or interaction patterns.

Dataset Structure

The dataset of 5 dataframes:

Posts_Primary columns

Human post fields:

• —Post ID - Instagram post shortcode
• —Human Post Link - Direct URL to the post
• —Type (Reel/Post) - Content format (Post or Reel)
• —Art Description - Short thematic tags describing the artwork (e.g. nature, oil painting, portrait, watercolor)
• —Instagram Account - Creator account handle
• —Followers - Follower count of the human creator at time of collection
• —Posts - Total number of posts on the human creator's account
• —Likes - Total like count on the post
• —Total Comment Count - Total number of comments on the post
• —Reposts - Repost count
• —Shares - Share count

AI post fields (mirrored structure):

• —Post ID.1, AI Post Link, Type (Reel/Post).1, Art Description.1, AI Art Insta Account, Followers.1, Posts.1, Likes, Total Comment Count, Reposts, Shares

Pairing field:

• —Discovery Hashtag (used to find the post) - The hashtag used to surface the AI post for this pair

Comment datasets

Each comment dataset has one row per post (indexed by Post ID) and up to 50 columns (1–50), where column n contains the n-th comment in the collected set. The date sheets follow the same structure, with datetime values instead of text.

Important: Comments are not stored in chronological order. Column position reflects collection order (filtered by highest engagement - see below), not posting time. The corresponding timestamp sheets must be used to sort comments by actual posting time for any temporal analysis.

Collection Methodology

Account setup

A new Instagram account was created prior to data collection to prevent the platform's recommendation algorithm from being influenced by prior interests, follows, or engagement history. This was done to ensure posts were discovered organically through hashtag search rather than through a personalised feed.

AI post discovery - hashtag-based search

AI posts were discovered by searching Instagram using a curated set of AI art hashtags. Hashtags were selected to cover a range of community sizes, from large general tags to smaller niche communities:

A post was included as an AI post only if it explicitly disclosed AI usage - either in the caption, as a hashtag, or in the post description. Posts where AI use was implied but not stated were excluded.

Human post discovery - thematic matching

For each AI post selected, a human art post was manually identified to serve as its pair. Human posts were matched on visual theme and subject matter - for example, an AI post depicting a fantasy portrait was paired with a human post depicting a traditionally rendered portrait. This matching was done to control for content category when comparing engagement and comment patterns.

Human posts were selected from artists with no explicit AI disclosure, sourced through art-focused hashtags and Instagram's explore feature.

Post type

Both conditions include a mix of static image posts and Reels:

The analysis primarily focuses on posts (static images) given their greater representation in the dataset.

Comment collection

For each post, the top 50 comments were collected. Comments were filtered and ranked using a composite criterion of highest likes, highest comment reply count, and overall engagement, reflecting Instagram's own ranking of "most relevant" comments. This means the dataset captures the comments most visible to a typical viewer, rather than a chronological sample.

Up to 50 comments per post were retained. Posts with fewer than 50 comments have NaN-padded columns for the remaining positions.

Dataset Statistics

Scale

• —40 matched pairs (40 human posts, 40 AI posts)
• —Up to 50 comments per post
• —80 posts total, spanning a wide range of follower sizes and engagement levels

Engagement metrics summary

Both conditions show high variance driven by a small number of viral posts. Median values are more representative of typical posts in the dataset.

Creator account size

Research Questions This Dataset Addresses

This dataset was originally collected to investigate three research questions:

RQ1 - Engagement analysis: Do AI-disclosed art posts receive statistically different engagement (likes, comments, reposts, shares) compared to matched human art posts?

RQ2 - Comment classification: What types of reactions appear in comments on AI art posts versus human art posts? Categories examined include: praise/admiration, authenticity policing, ethics/copyright concerns, hostility/harassment, support/defense, neutral questions, humor/memes, emoji-only responses, and non-English comments.

RQ3 - Sentiment progression: Does sentiment in comment sections change over time as more people comment? Specifically, does negativity escalate in AI art posts (bandwagon effect) compared to human posts?

Limitations

• —Instagram-specific: Results may not generalise to platforms with different moderation policies, comment ranking algorithms, or user demographics.
• —Explicit disclosure only: Only posts that explicitly disclosed AI use were included. Posts where AI use is hidden or ambiguous - which may attract qualitatively different reactions - are not represented.
• —Comment moderation: Instagram's content moderation filters out the most extreme comments before they can be collected. This means the dataset likely underrepresents highly toxic or hostile comments.
• —Comment ranking: Comments were collected based on engagement rank (likes + replies), not chronological order. This introduces a selection bias toward comments that resonated with the existing audience of each post.
• —Hashtag search ordering: Instagram does not sort hashtag search results by a single transparent criterion. The search order had an element of unpredictability, which introduces some randomness into which posts were selected - beneficial for diversity, but limiting for reproducibility.
• —Single-hashtag search constraint: Instagram's search interface does not support combining multiple hashtags in a single query. Each AI post was discovered through one hashtag at a time.
• —Skew toward positive content: The dataset has a positivity bias - the majority of comments across both conditions are positive or neutral. This reflects both the nature of the communities discovered through AI art hashtags and the effect of comment moderation.
• —Dataset size: With 40 matched pairs, the dataset is suited for exploratory analysis and hypothesis generation. Studies requiring high statistical power for subgroup comparisons would benefit from an expanded collection.

Citation

If you use this dataset in your research, please cite:

@dataset{ai_vs_human_art_instagram_2026,
  author    = {Lakshminarasimhan, Akshaya and Vuong, Cynthia and Yang, Grace and Phadke, Soham},
  title     = {Instagram AI Art vs Human Art - Engagement and Comment Dataset},
  year      = {2026},
  note      = {40 matched pairs of AI-disclosed and human art Instagram posts
               with engagement metrics and top-50 timestamped comments},
}

License

This dataset is released for non-commercial research purposes. All Instagram content remains the property of its respective creators. Users of this dataset are responsible for ensuring their use complies with Instagram's Terms of Service and applicable data protection regulations.