GEO (Generative Engine Optimisation) is the practice of structuring content so AI search engines cite it. Here’s the citation playbook — what works, what doesn’t, and how to measure it.
The three citation drivers
1. Structured definitive answers. LLMs cite sources that answer a question directly and authoritatively in the first paragraph. Burying the answer below 500 words of context kills your citation rate.
2. Entity authority. LLMs prefer sources where your brand is clearly identified as the authority on the topic. This means consistent entity references across your domain — your About page, your author bios, your structured data.
3. Schema markup. FAQ schema, HowTo schema, Article schema, Organization schema — all increase the probability of citation. Most B2B sites in the UK don’t ship these at scale.
The seven content moves that work
1. Answer the question in the first sentence. Then expand. Don’t bury the lede.
2. Use FAQ structured data on every important page.
3. Cite primary sources clearly (data, research, named clients).
4. Write definitive titles. “The 8 SaaS churn signals” outperforms “Some thoughts on churn”.
5. Use specific numbers. “+22% repeat purchase rate” outperforms “significant improvement”.
6. Maintain consistent entity references for your brand name and key terms.
7. Update content regularly. LLMs prefer recent sources for time-sensitive queries.
How to measure GEO performance
GEO measurement is early-stage. The honest tools available in 2026: manual sampling (asking ChatGPT/Claude/Perplexity questions about your category and tracking citations), Perplexity’s source visibility report, and emerging GEO tracking tools like Profound and Otterly.
We sample weekly across our active GEO clients and track citation appearances per category keyword. Citation lift is typically visible 60–90 days after structural content changes ship.
What doesn’t work
Stuffing keywords. AI models are largely keyword-agnostic at the citation layer. They care about authority and answer quality, not density.
Generic AI-generated content. LLMs are increasingly trained to detect and downweight obviously AI-written content. Brand-voice trained content reads as human; lazy generic output doesn’t.
Hiding your authority signals. If your About page and author bios are thin, the model has nothing to anchor your authority to.
Where to start
Run a GEO audit: ask ChatGPT, Claude, and Perplexity 10 questions about your category. See who they cite. That’s your competitive baseline. The work is moving up that list.