Five minutes with Rokas Tarasovas

What are clients coming to you for at the moment when it comes to AI?

Visibility is the big one, whereas previously it was about Google rankings and optimising your website for search. Increasingly, people are using AI agents for their initial research. The difference is that LLMs don’t simply rank sources according to SEO best practice.

Specialist publications can suddenly become much more influential. An LLM might pick up a quote from a very niche title because it considers that publication authoritative on a particular subject, even if it only has a few thousand monthly visitors.

You also mentioned reputation. Is that becoming harder to manage because of these systems?

Yes. Reputation can have a very long tail, because an AI agent can resurface a story from five years ago that is buried beyond the first few pages of Google. If you ask specifically about a brand’s controversies, that history can come back.

The answer isn’t to rewrite the past, but it is to make sure the original information is accurate and, where appropriate, provide context about what happened afterwards: what measures were taken and what results were achieved.

There’s been a lot of talk about Reddit becoming more important. Are you actually seeing brands respond to that, or is some of it still hype?

Reddit is one of the sources that was previously somewhat overlooked, and now it is being quite heavily favoured by AI agents, so attention is shifting there.

But it depends on the AI agent. Gemini tends to favour YouTube, while Grok focuses more heavily on X. The media landscape was already becoming fragmented, but now communications teams have to take all of these different sources into account.

How do you actually measure how a brand is being represented when there are so many different AI systems and ways of asking a question?

You need to test systematically. We create prompts based on different audiences – a consumer, an investor, a journalist or someone considering acquiring the company – and analyse the responses.

Location matters too, because the same question can produce very different results in the UK and Norway, for example. We can run hundreds or even more than a thousand prompts around a particular topic, looking at different audiences and markets. That gives a much broader picture than relying on whichever chat agent a client happens to use.

How different is GEO really from the SEO approach that communications teams are already familiar with?

We’re definitely increasing our focus on GEO, but there is still a lot of education to do with clients and account teams.

Clients are interested, but they are approaching it carefully. They want to understand what is genuinely impactful for their brand rather than simply responding to the latest hype. Once they understand it, I think it is just as important as having your website properly optimised for SEO.

And what do you think isn’t getting enough attention yet?

Synthetic audiences are one example. Conventional message testing has relied on focus groups and surveys for a long time. I’m not saying AI will replace them – human responses can be more nuanced – but synthetic audiences can be a useful first step.

More broadly, I encourage my team to use AI as a critical assistant rather than simply an end-product generator. Stress-test an idea, ask it to be critical, find the weaknesses and refine it. Then bring it to colleagues for a second pair of eyes.