By late 2024, the SEO and wider digital marketing industry were treated to an abundance of new tools, experiments and thought leadership relating to AI. Up to now it’s felt a little too early to make any comparisons between the likes of ChatGPT vs. Gemini, or any other AIO agents. But as 2025 has flown by and we realised it was already May, it seems like a good time to dive into some real data and investigate what insights there are to uncover from our own, real-world implementations of AI search reporting.
Towards the end of 2024, we began tracking AI traffic for our SEO clients more specifically through a custom report in GA4 (for our sins). Whilst a bit fiddly (and who knows quite how perfectly accurate), it was at least beginning to provide us with some evidence that these new competitors to traditional search were a) actually being used and b) our clients were optimised to appear in results.
On the surface, Gen AI tools are largely doing the same thing. They respond to queries in a similar way, even look similar in their interfaces. But when they start to specialise in certain tasks, things get really impressive. If you haven’t already tried it, ask Loveable.ai to build you something! In our case though, comparing ChatGPT against Gemini and Perplexity is a closer matchup, with the primary variable being search and website visibility.
Admittedly, the original purpose of setting up AI reporting wasn’t with the foresight of preparing any detailed analysis or public report. It wasn’t even to try and pit one AI tool against another. Most importantly, we wanted to ensure our clients’ websites were visible and prepared to receive consistent traffic from AI search. It was also a bit of a experiment to start monitoring how our ongoing strategy was performing for this channel of search, in some cases without any specific adjustments to ‘target’ AI.
As already mentioned, and at the time of writing this, tracking specific AI-referred traffic from the likes of ChatGPT and Gemini is not very intuitive from basic reporting. Instead, we looked to the ‘Explore’ function and built a custom data set that would track things like sessions and engagement from a defined list of page referral URLs. Again, no pre-existing template for this available in GA4, but helpfully stumbling across some Regex on LinkedIn, we were able to quite quickly collate what we wanted to see.
If you’re interested in setting something like this up yourself, here’s a quick Loom video which takes you through a few steps. The formula itself is quite simple to put together, or send us a message on LinkedIn and we’d be more than happy to share.
What were we looking for? The longer time went on and the more data that came through, it became clear there would be some insights from this given a little longer. Firstly, what will happen to everyone’s favourite vanity metric of ‘traffic’. Well actually, it’s a bit unfair to do this metric dirty when dealing with a new channel of search. We would probably expect fluctuations in sessions from AI Search in it’s infancy, but ideally a gradual growth in session data would show that our clients’ sites are being picked up more and more – this is not irrelevant data.
The next thing we wanted to understand was if there was a pattern in the types of pages being ranked. Our client base is largely B2B, so lead generation is a key measurement of successful campaigns and not as immediately-measurable as e-Commerce. This would suggest that zero-click is going to have an impact on how much data we can actually gather. But our assumption would be to find case studies and articles being the primary page types ranking at a glance.
For a bit more context, our findings are based on primarily B2B clients. We used a sample of 7 with data being taken from the start of January 2025.
For our tested clients who are primarily B2B, AI-driven traffic was largely concentrated on sector-specific and informational content. As we might of predicated, ChatGPT led the charts as the overall top referrer. Our retail dataset is probably not the most efficient example, but the 80% share of voice here was an interesting find, particularly considering the volume of sessions was well over 100 . Perplexity AI and Gemini had a more notable presence in property, recruitment and law. We often hear too much about Perplexity as a competitor to the others, but this was flagged as a key player very early in the analysis process.
Our IT sector saw the most traffic through Gemini, which generally, was a bit of a non-event. The most visited page types were again very aligned with our educated guesses – insights and industry news or specific articles (e.g., UK housing forecasts).
Recruitment and candidate resources (e.g., oil & gas sectors, relocation guidance), featured prevalently alongside case studies and locational information. Overall, sessions reflected strong interest in industry expertise, geographic capability, and sector trends.
From what we’ve been able to measure, there isn’t much of a comparison to be made here, yet. It’s frequently reported that AI search clicks make up on average about 1% of total traffic, a statistic recently backed by key speakers at Manchester’s SEO event HIVE MCR. Statistics from our own research here and beyond corroborate these findings and although we all expect the almost negligible percentage to grow, it’s rarely going to be an excuse for ‘losing’ traffic month-on-month. You have those hit-and-miss AI overviews to thank for that.
What do we know about targeting AI searchers? At the moment, the consensus is an evolution of what search professionals have been recommending for years, through every update. The more relevant and original your content is, the more authority you can build, and the better engagement you can get with those pages, the ranking signals will start to build. There’s also been a lot more attention recently paid to building out detailed Schema on pages where it’s possible to do so. The thinking here is simply building out the context you need in one of the most straightforward ways a machine can understand it, before crawling the page itself. Shortening the path to crawlers understanding the relevancy of your page vs. the search you’re targeting, the more chance you have to increase those impressions.
If we’re thinking about targeting specific AI tools, and wondering whether it’s better to rank for ChatGpt vs. Gemini, there isn’t enough data out there to make a firm stance on what’s best.
The best thing we can do as an SEO community is keep experimenting and sharing our findings. Optimising for AI is an evolution of traditional SEO best practice, with a refined approach to ensuring relevancy against your target search.
Digital Marketing
16th May, 2025
Call us on: +44 (0) 161 941 5330 or email us: info@firstinternet.co.uk
Get in touch today!
