Instacart quietly removed AI-generated recipe photos users found impossible and unappetizing
Instacart deployed AI-generated images alongside recipe content on its platform that contained physically impossible food depictions such as conjoined chickens, hot dogs with tomato interiors, and lemons fused with lettuce. After users flagged the images on Reddit and press coverage ensued, Instacart quietly removed the offending AI images and replaced some with stock photography. The company stated it reviews AI-generated content and may remove it when it does not deliver a high-quality consumer experience.
An AI image generator with no grasp of food anatomy produced stomach-churning photos that slipped straight into production because no one checked them first.
Key facts
- What
- Instacart deployed AI-generated images alongside recipe content on its platform that contained physically impossible food depictions such as conjoined chickens, hot dogs with tomato interiors, and lemons fused with lettuce.
- Incident date
- Jan 1, 2024
- Who
- Instacart
- Failure mode
- Hallucination
- AI surface
- Search / RAG
- Severity
- Medium
What happened
In early January 2024, users on the Instacart subreddit began noticing that recipe pages featured AI-generated food images with glaring visual artifacts, including conjoined chickens, hot dogs with tomato interiors, and lemons fused with lettuce. Business Insider reported on the issue on January 28, 2024, cataloging multiple examples of the unappetizing and misleading images that also accompanied recipes with nonexistent ingredients. Within days of the report, Instacart quietly removed the highlighted AI-generated images and replaced some with stock photography, though some AI imagery remained on the site. An Instacart spokesperson confirmed the imagery was AI-generated and stated that when the company receives reports of AI content that does not deliver a high-quality experience, the team reviews and may remove it.
What broke inside the model
- 01 · TriggerA user asks for a fact, a citation, or a figure.
- 02 · Model stepThe model writes a fluent, confident answer.
- 03 · Control gapNothing ties the claim back to a real source.
- 04 · FailureA fabricated fact ships as if it were verified.
- 05 · ConsequenceThe false claim reaches a customer, a court, or the public.
Confidence holds, and even spikes, as the claim detaches from any source.
The AI image generation model produced food images with physically impossible features because it lacked any understanding of real-world food anatomy, creating outputs like conjoined poultry, lemons fused with lettuce, and hot dogs with tomato interiors. These hallucinated visuals were deployed to production without adequate human review or quality gating that could have caught the visual artifacts before they reached customers browsing recipes on the platform.
What it cost
Sources
- PressInstacart Is Now Using AI for Food Pics. It's Incredibly Unappetizing.businessinsider.com
- PressInstacart Quietly Deletes Its Unsettling AI-Generated Food Picsbusinessinsider.com
- PressInstacart's AI Recipes Look Literally Impossible404media.co
Cite this entry
https://failureindex.ai/failures/instacart-quietly-removed-ai-generated-recipeAI Failure Index. "Instacart quietly removed AI-generated recipe photos users found impossible and unappetizing" (FI-0092). Realm Labs. https://failureindex.ai/failures/instacart-quietly-removed-ai-generated-recipe (indexed Jun 4, 2026).Data fields CC-BY 4.0, prose citation permitted. Incident ID FI-0092. Full dataset at /data.
Note from Realm Labs, the Index steward
How Realm would have caught this
- Prism
- OmniGuard
- AI Detection & Response (AIDR)
A runtime layer that watches the model's internal state can flag the moment a model commits to a claim it has no support for, and hold or reroute the response before it reaches a user. Realm reads those signals in real time rather than grading the transcript after the fact.