
Introducing Custom Vocabularies: Accurate Expert Interview Transcripts
New feature dramatically improves transcription accuracy for technical terms, company names, and industry jargon in expert interviews.
InsightAgent Team
February 10, 2026
We've shipped a major improvement to transcription accuracy for expert interviews. Technical terms, company names, and industry jargon now come through correctly.
If you've experienced the frustration of reading "Jane and Tech" instead of "Genentech" in a transcript, this update is for you.
Smart by Default
Before diving into the new feature, it's worth noting: InsightAgent's transcription is already significantly more accurate than standard speech recognition for expert interviews.
Our proprietary AI-powered transcription process automatically analyzes each interview's context—the expert's background, their company history, the questions being asked—and optimizes for relevant terminology. When you interview an oncology researcher from Genentech, the system already knows to expect pharmaceutical terms, biotech company names, and drug nomenclature.
This intelligent processing handles the majority of cases without any manual setup. Most users will see accurate transcripts out of the box.
Custom Vocabularies builds on this foundation, giving you additional control for edge cases, highly specialized terminology, or niche coverage areas where you have domain knowledge the system doesn't.
The Problem We're Solving Further
Expert interviews are filled with specialized terminology that standard speech recognition doesn't understand. Drug names, company names, financial instruments, technical acronyms—the exact words that carry the most insight are the ones most likely to be wrong.
This isn't a minor inconvenience. When the most important words in a conversation are transcribed incorrectly:
- Search fails — Finding mentions of a company becomes impossible when it's spelled differently each time
- Analysis inherits errors — AI summarization tools perpetuate mistakes
- Credibility suffers — Sharing error-filled transcripts undermines confidence in your research
We set out to fix this systematically.
Real Results
Here's what we observed testing with an oncology research expert:
| Before | After |
|---|---|
| "Jane and Tech" | Genentech |
| "nevalumab" | Nivolumab |
| "boracetamutations" | BRCA mutations |
| "Janentech" | Genentech |
| "atesolizumab" | Atezolizumab |
Same interview. Same audio. Dramatically better output.
The difference is especially pronounced in industries with dense specialized vocabulary: pharmaceuticals, biotechnology, financial services, enterprise technology, and healthcare.
How Custom Vocabularies Extends This
While our AI handles most terminology automatically, Custom Vocabularies gives you explicit control through three additional layers.
Account-Level Vocabulary
Build a library of terms relevant to your industry and coverage area. This becomes your foundation—terms that should be recognized correctly across all interviews.
Organize terms into categories that make sense for your workflow:
- Companies you track
- Products and brand names
- Technical terminology
- Industry acronyms
- Key people
Terms you add here apply to every interview automatically.
Interview-Specific Tuning
Each interview can have vocabulary adjustments based on the specific expert and topic. Preparing for a call with a semiconductor analyst? Add relevant chip architectures and company names. Interviewing a healthcare executive? Include relevant drug names and medical terminology.
These interview-level additions supplement your account vocabulary without cluttering it with one-time terms.
Intelligent Suggestions
The system analyzes your interview context—the expert's background, their company, the questions you've prepared—and suggests relevant vocabulary you might have missed.
This catches terms that are obvious in hindsight but easy to overlook during preparation.
Getting Started
You don't need to do anything to benefit from our intelligent transcription—it works automatically. But if you want maximum control, here's how to set up Custom Vocabularies.
Step 1: Build Your Account Vocabulary
Navigate to Settings > Vocabulary and create categories for your most common terminology.
Start with the terms that cause the most problems:
- Company names with unusual spellings or pronunciations
- Product names specific to your coverage area
- Acronyms that expand to industry-specific meanings
- People whose names are frequently misspelled
You can add terms manually, import from CSV for bulk additions, or use AI suggestions to generate relevant terms from your existing interviews.
Step 2: Review Before Important Interviews
For high-stakes interviews, check the vocabulary section on the interview detail page. The system shows which terms will be active and lets you add interview-specific vocabulary.
This takes thirty seconds and can significantly improve transcript quality.
Step 3: Build Over Time
Each time you notice a transcription error, add the correct term to your vocabulary. Over time, your library becomes increasingly comprehensive, and accuracy improves across all interviews.
Use Cases by Industry
Pharmaceuticals and Biotechnology
Drug names are essentially random syllables to general speech recognition. Generic names like nivolumab, pembrolizumab, and atezolizumab follow naming conventions unfamiliar to general models.
Build vocabulary categories for:
- Approved drugs in your coverage area
- Pipeline compounds
- Mechanisms of action (PD-1, CTLA-4, CAR-T)
- Biomarkers and diagnostic terms
- Pharmaceutical companies
Financial Services
Ticker symbols, fund names, and financial instruments create transcription challenges.
Build vocabulary categories for:
- Companies in your coverage universe
- Financial products and instruments
- Regulatory terms and agencies
- Key executives and fund managers
Technology
Product names, technical standards, and acronym-heavy discussion cause problems.
Build vocabulary categories for:
- Enterprise software products
- Cloud services and platforms
- Technical standards and protocols
- Startup names in your coverage
Healthcare
Medical terminology, procedures, and clinical language require accuracy.
Build vocabulary categories for:
- Medical procedures and treatments
- Diagnostic terminology
- Healthcare systems and payers
- Regulatory and compliance terms
Best Practices
Start Focused
Begin with the terms that cause the most frequent and most problematic errors. Company names and drug names typically deliver the highest impact.
Use Categories Strategically
Organize vocabulary by how you work. If you have analysts covering different sectors, create categories that align with their coverage areas.
Review Suggestions
The AI suggestion feature analyzes interview context to recommend relevant terms. Review these suggestions before important interviews—they often catch terms you'd otherwise miss.
Iterate Based on Results
After interviews, note any remaining transcription errors. Add those terms to your vocabulary. Accuracy compounds over time as your library grows.
Technical Details
For those interested in the implementation:
Vocabulary Limits
- Up to 1,000 terms total (optimized for speech recognition performance)
- Maximum 6 words per term (longer phrases should be split)
- No limit on categories for organization
Processing
Vocabulary is applied in real-time during transcription. There's no delay or post-processing wait—you see accurate transcription as the interview happens.
Privacy
Your vocabulary remains private to your account. Terms are used only for your interviews and are never shared or used for training.
What's Next
Custom Vocabularies is the foundation for continued transcription accuracy improvements. We're actively working on:
- Automatic vocabulary learning from corrections
- Industry-specific vocabulary packs
- Enhanced suggestions based on expert profiles
- Cross-interview terminology analysis
Your feedback shapes our roadmap. Let us know what would make the biggest difference for your workflow.
Get Started
Our intelligent transcription works automatically for all InsightAgent accounts—no setup required. Your next interview will already benefit from context-aware processing.
For teams wanting additional control, Custom Vocabularies is available now:
- Go to Settings > Vocabulary
- Create your first category
- Add terms specific to your coverage
- See even better accuracy for your niche terminology
Questions? Reach out to support@insightagent.io or schedule a call to see our transcription accuracy in action.
InsightAgent provides AI-powered expert interviews with transcription designed for research teams. Learn more.
Related Articles
Introducing Live Observation: Watch Your AI Interviews Unfold in Real Time
Live Observation gives you real-time visibility into every AI-conducted interview. See the conversation as it happens, build trust in your AI agents, and never wonder what was said on the call.
ProductCustom Variables: Personalize Your AI Interviews at Scale
Learn how custom variables let you tailor AI-powered expert interviews for different research projects, companies, and industries—without rebuilding your agent prompts each time.
AIWhat the OpenClaw Moment Tells Us About the Future of Research
OpenAI just acqui-hired the creator of OpenClaw, the viral open-source AI agent. Here's what the rise of autonomous agents means for research teams in finance, pharma, and consulting.
Expert NetworksWhy Expert Vetting Is Still Stuck in 2015 — And How AI Fixes It
Expert networks live or die on expert quality, but the vetting process hasn't evolved in a decade. Here's how AI-moderated screening removes the bottlenecks.
Expert NetworksThe End of the Expert Network Broker
AI is automating every function of the traditional expert network broker — sourcing, matching, scheduling, compliance, and moderation. What does that mean for a $4B industry built on human intermediaries?
Ready to transform your expert interviews?
See how InsightAgent can help your team capture better insights with less effort.
Learn More