Picture this (fictional scenario, but stick with it): You run a small accountancy firm in Sheffield. You've done all the right things: decent website, good Google reviews, ranking on page one for "accountant Sheffield." Business is ticking along nicely.
Then your nephew mentions he's been using ChatGPT to find services lately. Doesn't Google anything anymore. Just asks the AI. Curious, you type in "best accountant in Sheffield." Your firm isn't mentioned. Not once.
That gap (between your Google presence and your AI presence) is exactly the problem GEO was designed to fix. And in late 2023, a group of researchers decided to run a proper experiment to find out whether there was anything you could actually do about it.
Spoiler: there is.
The research
Researchers from Princeton University, Georgia Tech, and IIT Delhi published a paper titled "GEO: Generative Engine Optimization" (arXiv:2311.09735). It was accepted to ACM KDD 2024, one of the most well-respected data science conferences in the world. This isn't blog speculation. It's peer-reviewed, published research.
Their question was simple: can you deliberately make your content appear more in AI-generated answers? And if so, how?
They built something called GEO-bench, a large test dataset covering questions across business, finance, law, health, and tech. Then they ran different content strategies against real generative AI systems, measuring how often each approach got the content cited, quoted, or drawn on in AI responses.
What they tested
Six strategies. Each one a different way of tweaking your content.
The headline finding
Using these strategies together boosted visibility in AI responses by up to 40%.
That's a real number. If you get leads through online search (and most UK small businesses do), a 40% improvement in how often you appear in AI answers is commercially significant.
There was a catch: not all industries responded the same way. Tech and software content was the most responsive to GEO. Legal and medical content was harder, probably because AI systems apply extra caution in those areas. But harder doesn't mean impossible, and it doesn't mean you shouldn't try.
"The efficacy of these strategies varies across domains, underscoring the need for domain-specific optimisation methods."
Aggarwal et al., GEO: Generative Engine Optimization, KDD 2024
What this means for your business
Three things worth knowing.
Your content now has two audiences. You're writing for human readers and for AI systems that will decide whether your content is worth citing. That changes what good writing looks like.
Having real data helps a lot. If your business generates any interesting numbers (case studies, client stats, survey results, anything), publish them. Specific, verifiable data points are exactly what gets quoted.
Permissions matter more than people realise. Here's something the KDD paper didn't fully explore: AI companies are under growing pressure to use content they have clear permission to use. A business that's explicitly said "yes, you can cite my content" via an llms.txt file is a more attractive source than one that's stayed silent.
The opportunity is still wide open
Most UK businesses haven't started thinking about this. They haven't updated their content for AI, they haven't created an llms.txt file, they haven't even checked whether they appear in AI responses.
That's the opportunity. The businesses that move in 2025 will build a head start that's going to be genuinely difficult for later movers to close. AI systems build up citation patterns over time, and being in there early matters.
It doesn't require a big budget. It doesn't require an agency. A lot of it is just writing better and setting up a small text file on your website.
Start with your llms.txt
geotool.co.uk generates a compliant, UK-law-aware llms.txt file for your business in 60 seconds. Free, no account required.
Get my free llms.txt →