AI-Assisted Screening
AI-assisted screening surfaces a suggestion - include or exclude - for each study during the screening phase. These suggestions are based on the semantic similarity between the current study and studies you have already screened, helping you make faster and more consistent decisions across large study sets.
How it works
ActiveSLR generates a vector embedding for each study record using the study's title, abstract, and (if available) parsed PDF text. As you make screening decisions, the system learns from your pattern:
- Embedding generation - Each study is converted into a high-dimensional vector representing its semantic content.
- Similarity search - When you open a study, ActiveSLR searches for the most similar studies you have already screened.
- Suggestion derivation - The AI derives a suggestion (include or exclude) based on how your previous decisions align with similar studies.
- Confidence display - The suggestion is shown alongside a confidence level, so you know how strongly the AI pattern supports the recommendation.
Using AI suggestions during screening
AI suggestions appear in the study card during both title/abstract and full-text screening. Each suggestion shows:
- Recommended action - Include or Exclude
- Confidence level - A percentage indicating how consistent the suggestion is with your screening history
- Similar studies - A list of previously screened studies that influenced the suggestion, so you can verify the reasoning
You are not required to follow the suggestion. Click Include or Exclude as you normally would - the suggestion is advisory only.
AI suggestions become more accurate as you screen more studies. The first 20–50 decisions train the model. After that, suggestions are typically more reliable for high-confidence cases.
Suggestions and consistency
One of the main benefits of AI-assisted screening is consistency. When you have screened hundreds of studies over several sessions, it is easy to apply slightly different criteria over time. AI suggestions flag when a study is very similar to one you previously included or excluded, helping you notice potential inconsistencies before they accumulate.
AI suggestions do not automatically make decisions. You must explicitly click Include or Exclude. This preserves full researcher control and ensures your decisions are defensible.
Bulk screening with AI suggestions
In the bulk screening view, AI suggestions are shown for each study in the list. You can filter the list to show only high-confidence suggestions, review them quickly, and accept or override each one. This is particularly useful for accelerating the tail end of screening when most remaining studies are clearly off-topic.
When AI suggestions are not available
Suggestions are not shown for studies that:
- Have not had their embeddings generated yet (embeddings are generated asynchronously after upload)
- Are in projects where you have fewer than a threshold number of completed decisions (not enough training signal)
In these cases, the suggestion column shows a "Not available" state. Your screening decisions still contribute to future suggestions once the threshold is reached.