Users are actively seeking ways to close the feedback loop with AI tools by providing concrete examples and visual references, yet they still encounter 'AI slop' and imperfect outputs, indicating a gap between user intent and AI execution.
Users rely on specific, visual examples (screenshots, URLs) to guide AI tools, but the tools often fail to fully replicate the desired outcome, suggesting a need for better iterative feedback mechanisms.
Snapshot context
What makes this snapshot distinct
This snapshot is distinct because it combines reddit + g2 evidence within last 15 days, with 4 quote-backed signals around Close The Feedback Loop is a recurring product feedback theme. It also references related products such as close the feedback loop.
Most important finding
Users rely on specific, visual examples (screenshots, URLs) to guide AI tools, but the tools often fail to fully replicate the desired outcome, suggesting a need for better iterative feedback mechanisms.
Suggested focus
Monitor how users refine AI outputs over multiple iterations and whether tools that support visual references (like screenshots) reduce the 'AI slop' perception.
AI feedback clusters
Iterative refinement challenges
Users provide specific examples and visual references to guide AI, but the output often falls short, requiring multiple attempts and still yielding 'AI slop'.
βI like to give specific examples of what I want. Current project needs a library tree, and the default one is ugly. I googled library tree design, and found one I liked the look of. I screenshot it and gave it to claude and said do this. It got close, and theβ
βhttps://impeccable.style/ this one is good, and the unslop skill. Also give it an example, some people use mobbin or dribble and actual url of websites you like. That works for me. At the end, you will still see ai slop, most of the time it cannot one shotβ
Misunderstanding of data structure
Users debate whether LLMs are appropriate for structured data tasks, highlighting confusion about AI's role in handling metadata vs. unstructured search.
βMetadata is literally a table of structured data. That is a structured record. If you want to search actually unstructured data, the tool you need is a search engine. Not an LLM. The structure of a filesystem is to aid the user in storing files in a way thatβ
AI root-cause hypothesis
The root cause may be that AI models lack the ability to deeply understand and incorporate nuanced visual or structural feedback, leading to outputs that are 'close but not quite right'.
Product implications
Enhance feedback loops by allowing users to provide visual examples and iterate more effectively, reducing the 'one-shot' limitation.
Competitors that offer superior iterative refinement or better visual understanding could capture users frustrated with current AI output quality.
Startups can differentiate by focusing on closing the feedback loop with more robust example-based learning and user-guided corrections.
Source evidence supporting this signal
βMetadata is literally a table of structured data. That is a structured record. If you want to search actually unstructured data, the tool you need is a search engine. Not an LLM. The structure of a filesystem is to aid the user in storing files in a way thatβ
βI like to give specific examples of what I want. Current project needs a library tree, and the default one is ugly. I googled library tree design, and found one I liked the look of. I screenshot it and gave it to claude and said do this. It got close, and theβ
βhttps://impeccable.style/ this one is good, and the unslop skill. Also give it an example, some people use mobbin or dribble and actual url of websites you like. That works for me. At the end, you will still see ai slop, most of the time it cannot one shotβ
βWhat I like best about Notion is how it combines notes, documentation, project management, and collaboration in one flexible workspace. Itβs simple enough for everyday use while being powerful enough to organize complex projects and workflows. The only thingβclose the feedback loopΒ· g2