Related

AI could do for SMMEs what the internet never did
Martin Butler 19 hours


PR award recognises PFG CEO
17 Jun 2026

AI shopping booms in SA, but trust remains central
Irene Auma 12 Jun 2026


Top stories


![(Image source: @ Kinsta https://kinsta.com/ Kinsta]])](https://biz-file.com/c/2606/814361-700x364.jpg?2)
Text AI search can read, verify, and quote. But AI has trust issues. Rightly so. It's the substance behind the transcript that decides whether your content earns authority.
When a host meanders into verbal diary territory, there's nothing of substance to cling to. No insights. No clear point of view.
The transcript becomes word salad - for both the reader and for AI. A rambling, unstructured conversation produces a rambling, unstructured transcript.
The reader moves on. So does AI.
Imagine this: two interviews on the same topic. The first - a guest makes a sweeping claim and the host nods along, moving straight to the next question.
Fact: a nod does not make it into the transcript.
In the second, the host pushes back - asks for an example, a number, a counterargument.
The first produces a transcript abundant in assertions with nothing to verify.
The second produces a transcript abundant in claims that have been tested in real time.
One of those is citable. The other is noise with better audio quality.
AI search did not invent a separate definition of authority. It learned ours.
These models are trained on human behaviour - which sources people have cited, trusted, and quoted over time.
Say something once, and it's an opinion. Say it consistently, episode after episode, and it becomes a pattern AI can verify.
When a system decides a transcript is credible, it's running a version of the same instinct a sceptical listener runs in their head, just at scale.
Specificity reads as expertise. Structure reads as clarity. Repetition reads as reliability.
The same data points that make a stranger's voice trustworthy to us are the criteria that make a transcript trustworthy in a model's training data.
The risk here is that once brands or hosts learn AI rewards structure and specificity, they overcorrect.
Every answer gets engineered like a featured snippet. Every host starts talking in bullet points.
The conversation stops sounding human. Which defeats the entire mechanism I just spent two paragraphs explaining - because AI is trained on human behaviour, and humans don't trust a voice that sounds like it's reading from a script any more than AI search does.
A solid AI search strategy is simpler than it sounds, but it does require intent.
Publish the transcript, not just the audio. Text for the machine, audio for the listener.
Structure it so a system can locate the claim, not just follow the conversation.
Choose guests and topics specific enough that each episode produces something worth quoting, not just something worth listening to.
None of this happens by accident, and none of it happens in one episode.
It's built, deliberately, the same way authority is built anywhere else - one credible claim at a time. Again and again.
