RAG System Quality & Grounding Assessment (Developer Survey)

Welcome! Thank you for participating in this survey about your experience with Retrieval-Augmented Generation (RAG) systems. This survey takes approximately 6–8 minutes. Your responses are completely anonymous, reported only in aggregate, and used for internal research purposes. There are no right or wrong answers — we are interested in your honest opinions and experiences. Participation is voluntary, and you may stop at any time.

Have you built or maintained a RAG system in the last 6 months?

• Yes
• No
• Not sure

Which role best describes your day-to-day work?

• ML engineer
• Backend engineer
• Data scientist
• MLOps/Platform
• Product engineer
• Researcher
• Architect/Tech lead
• Other

Which content sources feed your retriever today? Select all that apply.

• Proprietary documents
• Code repositories
• Product knowledge base
• Web crawl
• Vendor API docs
• Slack/Chat logs
• Support tickets
• Wiki/Confluence
• Database/warehouse
• Not applicable
• Other

How are model answers grounded or cited in your RAG system? Select all that apply.

• Inline citations with URLs
• Inline citations with document IDs
• Evidence block after the answer
• Tool outputs included verbatim
• Structured JSON evidence list
• No grounding/citations
• Other

Which evaluation tools or libraries do you use for RAG? Select all that apply.

• Ragas
• TruLens
• DeepEval
• Promptfoo
• Custom harness
• LlamaIndex evals
• None
• Other

What is the primary programming language you use for RAG development?

• Python
• JavaScript/TypeScript
• Java
• Go
• C
• Rust
• Other

How critical is retrieval quality to the overall success of your RAG system?

How many years of professional experience do you have in software, data, or ML?

• 0–1
• 2–4
• 5–9
• 10–14
• 15+
• Prefer not to say

Thank you for completing this survey! Your input is valuable and will help improve RAG systems and developer tooling. All results will be reported in aggregate only.

In the last 30 days, how well did retrieved context meet your task requirements?

Over the last 30 days, how much do you trust the correctness of cited evidence in your RAG system's outputs?

Which metrics best reflect your RAG quality today? Select all that apply.

• Precision@k
• Recall@k
• MRR
• nDCG
• Answer faithfulness
• Context precision/recall
• Groundedness score
• Human ratings
• Production usage signals
• Custom internal metrics

What is your primary vector store or retriever backend?

• Pinecone
• Weaviate
• Milvus
• FAISS
• Elasticsearch/OpenSearch
• pgvector
• Chroma
• Vespa
• Not applicable
• Other

How satisfied are you with your RAG system overall today?

In which region do you primarily work?

• Africa
• Asia
• Europe
• North America
• Oceania
• South America
• Prefer not to say

How do you set or tune top-k and related retrieval parameters?

• Manual experimentation
• Grid/Random search
• Bayesian optimization
• Vendor auto-tuning
• Learned retrieval policy
• Not tuned
• Other

Rank your top 3 preferred citation/grounding display styles.

1. Inline per sentence
2. Numbered endnotes
3. Collapsible evidence panel
4. Top-k sources with scores
5. Link to full passages
6. Show only on demand

If you use a custom metric, please briefly describe it and how you compute it.

Which embedding model is your primary choice?

• OpenAI text-embedding-3
• OpenAI small embedding
• Cohere Embed
• VoyageAI
• Jina embeddings
• E5 family
• Instructor
• BGE family
• Local model
• Other

Rank your top 3 priorities for improving your RAG system in the next 3 months.

1. Improve retrieval precision/recall
2. Better grounding/citations
3. Reduce latency
4. Lower cost per query
5. Scale to more data sources
6. Harden evaluation pipeline
7. Security/compliance
8. Developer ergonomics

What is the approximate size of your organization (number of employees)?

• 1–10
• 11–50
• 51–200
• 201–1,000
• 1,001–5,000
• 5,001+
• Prefer not to say

In the last 30 days, how often did you encounter irrelevant or off-topic passages in retrieval results?

In the last 30 days, how frequently did you observe hallucinations in your RAG outputs despite grounding?

How automated is your evaluation workflow?

• None (manual only)
• Some scripts
• CI-integrated checks
• Continuous eval in production

Do you use a reranker after initial retrieval?

• Yes
• No
• Experimenting

Based on your responses in this survey, please share any additional thoughts or experiences about your RAG retrieval or grounding challenges.

What is the primary industry or domain for your RAG work?

• Technology
• Finance
• Healthcare/Life sciences
• Retail/CPG
• Education
• Government/Public sector
• Manufacturing
• Media/Entertainment
• Other
• Prefer not to say

In the last 30 days, how often did you encounter missing context (key information not retrieved)?

Please describe a recent grounding failure you encountered and its impact on your work.

How often do you run RAG benchmarks?

• Before each release
• Weekly
• Biweekly
• Monthly
• Quarterly
• Ad hoc only

Which reranker do you use most often?

• Cohere Rerank
• Voyage Rerank
• Jina Reranker
• Cross-encoder (e.g., MS MARCO)
• Self-hosted reranker
• Other

In the last 30 days, how often did you encounter stale or outdated content in retrieval results?

We'd like to understand more about your experience with grounding and citation quality. An AI moderator will ask you a couple of follow-up questions.

What is your end-to-end RAG latency target per query?

• < 200 ms
• 200–500 ms
• 500 ms–1 s
• 1–2 s
• 2–5 s
• > 5 s
• No specific target

In the last 30 days, how often did you encounter duplicate or near-duplicate chunks in retrieval results?