Gaslighting at work—being told you said something you didn't, that you agreed to something you disputed, or that you're misremembering a decision—can be spotted by searching your email and messages for what you actually said at the time. AI semantic search finds the relevant conversation even if you don't remember exact words, creating a record of reality that contradicts the false version.
Semantic search is a technique where a computer understands meaning, not just keyword matches. Traditional search finds "upset" because you searched "upset." Semantic search finds "frustrated," "exasperated," and "exhausted" because it understands they're semantically similar to your query. In workplace gaslighting documentation, semantic search helps you identify patterns that might not use consistent language.
Gaslighting relies on inconsistency and deniability. A toxic manager might say contradictory things about your performance across different conversations, deny things they obviously said, or reframe your concerns as paranoia. These patterns aren't always obvious when incidents are spread across months of emails, Slack messages, and meeting notes. Semantic search helps you pull together all instances of a pattern, even when the exact wording varies.
Under the hood, semantic search converts text into embeddings—mathematical representations of meaning. "My manager said I was wrong" and "My manager contradicted me" have different words but similar semantic meaning, so they have similar embeddings. A semantic search engine compares your query embedding to document embeddings and returns matches based on conceptual similarity, not keyword overlap.
For workplace documentation, this means you can search for: "instances where manager contradicted previous statements about my work" and find all relevant incidents, even if some say "contradicted," others say "changed the story," and others describe the contradiction indirectly. This is much more powerful than searching for the specific phrase "contradicted."
You don't need to build a semantic search system from scratch. Several tools now offer semantic search capabilities:
Gaslighting often involves your manager denying they said something, then saying it again differently. Semantic search helps you find the pattern: Search for "moments my manager denied responsibility" or "instances where my manager reframed my concern as my problem." Pull the results. Now you have a collection of incidents demonstrating the pattern that might not be obvious if you're reading chronologically.
Another pattern: your manager making contradictory statements about expectations. Search for "communication about project scope or deadlines" and you'll get all relevant incidents. Review them chronologically. A judge or HR investigator can see the evolution and inconsistency clearly.
Semantic search is only as good as your source material. If you haven't documented incidents carefully, semantic search can't help—garbage in, garbage out. Also, semantic search works best with full documents or substantial passages. Single-sentence Slack messages are harder to embed meaningfully.
There's also a risk of over-interpretation. Semantic search might return incidents that are thematically related but not actually evidence of a pattern. A manager saying "we need to revisit this" in two contexts might be flagged as inconsistency by semantic search, but it might just be standard management language. You still need human judgment to verify that semantic search results constitute a real pattern.
The most powerful approach: Use semantic search to retrieve all incidents matching a pattern theme, then feed those incidents to Claude or ChatGPT with instructions to analyze them for actual gaslighting indicators. Ask: "Do these incidents show a manager contradicting themselves about my responsibilities? What's the evidence?" The AI can then synthesize across incidents and note whether the contradictions are about substance (genuine gaslighting) or just variation in phrasing (not meaningful).
Try this: If you're using Notion for incident documentation, try Notion's AI semantic search feature. In your incident database, ask questions like "Find all times my manager said something contradictory about my performance expectations" or "Find all moments my manager denied making a commitment." Review the results. Does a pattern emerge? If yes, document that pattern explicitly in a summary, citing the specific incidents. This searchable, synthesized record is far more useful for any HR investigation or attorney consultation than scattered incident notes.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.