Robin Allenson's Linkedin Analytics

I did an A/B test on increasing internal links, specifically to pages ranking between positions 4 → 15 in search results and here’s what I discovered: ➟ The approach: I used Similar.ai's InternalLinking.io’s Boost Internal Links recipe to systematically increase links to target pages while maintaining a control group for comparison. ➟ The data sources: Botify’s Logfile Analyzer (via API) + Google Search Console. ➟ The methodology: A/B testing with a "difference of differences" approach to isolate the impact.And as seen in the graph below, we divided pages into three groups: 🔶 Boosted (Pink): Received 2x more internal links than usual. 🔶 Linked (Blue): Maintained standard internal linking. 🔶 Unlinked (Green): No additional links from our tool (though these pages links may still exist). There’s no doubt – the results were significant! The performance differences between boosted, linked, and unlinked pages illustrate the considerable benefits achievable by strategically increasing internal links. Internal linking isn’t just about navigation, it’s a powerful ranking tool. And when applied strategically, it can drive measurable improvements in crawlability, rankings, and traffic.

Sam Altman recently said that people typing “please” and “thank you” to ChatGPT costs OpenAI millions of dollars in compute, water and energy. And it's led to an increased number of people who use “AI is bad for the planet” as an anti-AI argument. I find this peculiar. We don’t agonise over the energy cost of bingeing Netflix or scrolling on TikTok, even though those services burn through data centres too. We’ve just normalised that waste because it feels invisible. Are we holding AI to higher moral standards than other tech? I spoke with someone at a climate conference recently that asked me if AI is just a terrible waste of energy and resources, and part of my answer was: Perhaps yes, assuming we don’t solve any climate problems with AI. And I don’t mean that we’re necessarily actively working on solving a particular climate problem, but I do think AI is going to break through barriers in fundamental science as it improves.

I have built countless AI tools in my life. Some successful, some not. A few years ago, we built a generative AI tool for fashion designers with the guidance of Amy Jo Kim. This was before generative AI worked well. We used GANs (generative adversarial networks), and the results were… janky. So we killed it. As founders, it can be hard to know when to walk away and when to pivot. But I knew it was the right decision because the tech simply wasn’t ready then. Now, a lot of founders are thinking about forming an AI-first company with the hope that AI development will lead them to a good enough product. So a lot of products are crap at the moment, but are in the works under the assumption that the next model will be a lot better. But… Although the models are getting better on the eval, for many founders they're not ACTUALLY getting better because founders only really care about the thing they’re using AI for. Unfortunately, a rising tide doesn’t necessarily lift all boats – as founders are iterating their product around a specific job to be done for their users, they may find that what they need AI to do for them isn’t developing as fast as they need it to.

“Do you want a billion-user consumer business or an API company?” “Both.” A few months ago, I posted about how most startups die from indigestion, not starvation. But in a recent podcast episode of Stratechery, Sam Altman made it clear that he wasn’t willing to sacrifice his B2C or B2B. On the one hand, the consumer side (ChatGPT) gives mass adoption and rapid feedback. On the other, the API business fuels enterprise use cases and funds R&D. They’re symbiotic: the API informs the product, and vice versa. But competing on both fronts against deep-pocketed giants (Google, Meta, Anthropic) sounds like a hell of a challenge. You’re innovating AND keeping up with the Joneses, who happen to have (pretty much) infinite capital. It doesn't seem likely it will end well. Then again, I’m not Sam Altman.

Looking back at where AI was a year ago, it’s quite baffling to think about how much has changed. But as they say… Plus ça change, plus c'est la même chose.  The more things change, the more they stay the same. It reminds me of the early days of mobile. Every year, there was a new iPhone with clearer screens, faster processors and better cameras. But at its core, the ‘smartphone’ aspect of it was still the same. And now it’s happening with AI. Every few months, there’s a new model: faster, smarter, better at some benchmark. Between OpenAI, Anthropic, Google, Mistral, DeepSeek and others – with a bunch of new names: Claude, Claude 3, Claude 3.5, Claude 4 Mini, GPT-4, GPT-4o, Gemini 1.5, Gemini 2… It’s hard to keep track of everything that’s changing. But what’s interesting to me is also what’s staying the same: Hallucinations. I think hallucinations are going to be the new ‘no signal in this area’ (at least, for those of you who don’t live in advanced countries like the Netherlands). As in, they will be a permanent fixture that becomes a norm and a non-issue. Am I making sense?

The debate around AGI often treats it like a binary milestone – it’s either we have it or we don’t. But AI’s impact is already here, it’s just unevenly distributed. Some tasks are automated better than humans, while others still need a human touch. And, in many cases, we get which is which mixed up. Machines have been beating humans at specific tasks for decades. Calculations, chess, manufacturing. But because they’re no longer “magic,” we dismiss them as just… tools.