AI is changing how media monitoring companies track, process, and analyse coverage. I do not believe the choice is between technology and people. AI can take on more of the work involved in collecting and processing media, but there are still points where human judgement determines whether what reaches the client is accurate and useful.
Used well, AI can help monitoring teams work through more coverage without taking away the judgement clients still rely on to understand what they are seeing. Broadcast monitoring makes that particularly clear. In South Africa, the challenge is heightened by the range of languages and accents heard across radio and television, along with names and local expressions that automated systems do not always interpret correctly.
A recent 2026 AfriVox benchmark tested speech models across 20 African languages and more than 100 African English accents. Performance varied substantially, including among languages the models were intended to support, while unscripted speech proved particularly difficult in some tests.
Why broadcast creates a harder problem
Broadcast does not arrive in neat blocks of searchable text. A presenter may introduce an interview before an ad break, return to the guest several minutes later, and continue for an unpredictable length of time.
An automated system may identify the correct brand or spokesperson but still get the start and end points of the clip wrong. If it begins too late, an important part of the exchange may be missing. If it runs too long, unrelated discussion can be attached and change the context.
That is when a technical monitoring error becomes a business problem. A communications team may brief an executive from an incomplete record or misunderstand how an interview landed. Finding the mention is not enough if the information being used to make a decision is inaccurate.
Why human judgement still matters
Someone listening to the coverage can often recognise when part of an interview is missing or when the surrounding discussion changes what was said. That is much harder to judge from an automated transcription alone.
A broadcast clip may help a communications team decide whether an issue needs attention or shape how an executive is briefed on a developing story. If the clip is incomplete or misleading, the team is working from the wrong picture.
This is why I am cautious about treating maximum automation as the goal. I am strongly in favour of using AI wherever it improves the work, and I expect its role in media monitoring to keep growing. But the level of automation means very little if the end result becomes less reliable.
Where experience still matters
At Novus, our broadcast team has a combined 116 years of experience. That knowledge has been built in a South African media environment where understanding language, pronunciation, and context can make the difference between identifying a mention and understanding what was actually broadcast.
That experience now sits alongside AI in the monitoring workflow. The technology can help teams work through large volumes of media more efficiently, while specialists remain involved where a clip needs to be checked or interpreted before it reaches the client.
Clients will ultimately judge a monitoring service by the quality of what they receive. We should automate where the technology improves that result, rather than assume that every part of the workflow needs to be handed over to AI.
