Live captioning for conference sessions had historically depended on human stenographers or captioners working in real time, a service that was accurate but expensive enough that many conferences offered it only for a limited set of flagship sessions rather than across an entire multi-track program. By 2024, AI-driven live captioning had matured enough in accuracy and cost that a much broader range of conferences began offering it as a standard accessibility feature across most or all sessions, rather than a limited, budget-constrained exception.

Why Human Captioning Had Limited Coverage for So Long

Professional human captioners capable of accurate real-time captioning are a genuinely scarce, specialized skill, and their services were priced accordingly — a cost that made comprehensive captioning across a large multi-track conference program prohibitively expensive for most organizers, who instead concentrated their captioning budget on the highest-visibility keynote sessions and accepted that most breakout sessions would go uncaptioned. This wasn’t a matter of organizers undervaluing accessibility; it reflected a genuine cost constraint that limited coverage regardless of intent.

What Changed Technically

AI speech-recognition accuracy for live, unscripted speech improved substantially over the preceding several years, reaching a quality level by 2024 that made AI-generated live captions genuinely usable for general audience needs, even though accuracy still trailed dedicated human captioners for specialized technical vocabulary, strong accents, or multiple simultaneous speakers talking over each other. Combined with a cost structure dramatically lower than human captioning — since AI captioning scaled to additional simultaneous sessions without requiring additional scarce specialized staff — this shift made comprehensive, conference-wide captioning coverage financially realistic for organizers who could never have afforded equivalent human-captioner coverage.

Accuracy Gaps Remained for Specialized Content

AI captioning’s accuracy limitations were most visible in exactly the kind of dense technical content common at industry-specific conferences — specialized jargon, acronyms, and product names that a general-purpose speech-recognition model hadn’t been specifically trained to recognize reliably. Some conferences addressed this by providing session-specific vocabulary lists to their AI captioning vendor in advance, improving recognition accuracy for known technical terms, while others accepted a higher error rate for specialized vocabulary as a reasonable tradeoff against the alternative of no captioning at all for the majority of sessions.

Multiple simultaneous speakers, common during panel discussions and audience Q&A, remained a persistent accuracy challenge for AI captioning systems, which generally performed markedly better on single-speaker presentations than on fast-moving, overlapping panel conversation.

Deaf and Hard-of-Hearing Attendees Had Mixed Reactions

Feedback from deaf and hard-of-hearing attendees, the population most directly served by captioning, was genuinely mixed. Many valued the dramatically expanded coverage — being able to follow far more sessions than the previous human-captioner-only model had ever made accessible — even while acknowledging the accuracy tradeoff for specialized content. Some attendees who’d relied on the more limited but more accurate human-captioned flagship sessions in prior years expressed a preference for that higher accuracy where it was available, suggesting the ideal approach for some organizers was a hybrid model — human captioning retained for the highest-stakes flagship sessions, AI captioning extended to the much broader remaining program that would otherwise have had no captioning coverage at all.

Beyond voluntary accessibility commitment, evolving legal interpretation of accessibility requirements for public accommodations and events pushed some organizers toward broader captioning coverage as a compliance consideration, not solely a values-driven choice. This legal dimension varied by jurisdiction and specific applicable law, but it added a further practical incentive, beyond cost reduction alone, for organizers to expand captioning coverage using the newly affordable AI option rather than continuing to limit coverage to a small number of flagship sessions.

In the United States, the framework most often cited was the Americans with Disabilities Act’s requirement that covered entities provide “effective communication” through appropriate auxiliary aids — a standard the U.S. Department of Justice outlines in its guidance on effective communication. Notably, that standard is about the outcome, not the method: it asks whether communication is genuinely effective for the individual, which is exactly where AI captioning’s accuracy limitations became legally relevant rather than merely a quality concern. A caption stream riddled with errors on the very technical vocabulary an attendee came to hear arguably falls short of effective communication even though captions are nominally present, which is part of why the more cautious organizers retained human captioning or CART services for higher-stakes sessions and treated AI captioning as a coverage expansion rather than a wholesale substitute for accuracy where accuracy carried legal weight.

Frequently Asked Questions

Why couldn’t most conferences offer live captioning across their full program before 2024?

Professional human captioners are a scarce, specialized skill priced accordingly, making comprehensive coverage across a large multi-track program prohibitively expensive for most organizers, who instead concentrated limited captioning budgets on flagship keynote sessions only.

How accurate was AI live captioning compared to human captioners by 2024?

AI accuracy for general, single-speaker speech reached a genuinely usable quality level, though it still trailed human captioners for specialized technical vocabulary, strong accents, and multiple overlapping speakers during panel discussions.

Did organizers still use human captioners at all once AI captioning became available?

Some adopted a hybrid model, retaining human captioning for the highest-stakes flagship sessions where accuracy mattered most while extending AI captioning to the much broader remaining program that would otherwise have had no captioning coverage.

In some jurisdictions, evolving interpretation of accessibility requirements for public accommodations and events added a compliance incentive alongside cost reduction, pushing some organizers toward broader captioning coverage using the newly affordable AI option.

For related coverage, see our piece on what it actually took to livestream a keynote in 2016.