Your tech questions: How can our work appear in AI chats?

When working with organisations to explore what "AI" is and the effects the fast rollout of the technology could have in their work, I was surprised to hear this question multiple times: If people start using chatbots to get their information, how do we make sure our reports and outputs are included in the responses?

Even though chatbots are used regularly by only a fraction of the population, that number is growing. Google has also rolled out AI overviews for most searches, which means even if users don't want to use AI to find a piece of information, they will be forced to. The urgency of understanding how to appear in responses built by LLMs is clear, but what surprised me was the way a question I considered being part of only the marketing remit appeared when talking to people across different roles within organisations.

Isn't this SEO?

The practice of trying to understand changes to the algorithm that decides which website features first on Google searches and developing tactics to be seen on relevant queries is part of the search engine optimisation (SEO) field.

In SEO, there were never any rules. Google released a few general best practices, but real tactics were built based on millions of experiments, hours spent analysing dashboards and exchanging information with peers. All of that, together with any results you've ever achieved, could come crumbling down with a single update. These updates are mostly unannounced, with no transparency on what changed in the algorithm.

Maybe because they are used to look at a black box and try make sense of it, many SEO professionals (including me!) are now working with products based on large language models. With the introduction of chatbots built on LLMs, another race was born: who would be cited first in responses? This field has many names today, such as answer engine optimisation (AEO), large language model optimisation (LLMO), artificial intelligence optimisation (AIO or AI Overviews optimisation for a Google-focused approach), or the one I think makes more sense, generative engine optimisation (GEO).

Optimising for generative engines

If you are interested in learning more about GEO, this is a great article to get started.

If you're here for the headlines, the key thing to understand is that search engines based on generative AI usually have two main sources: web search and the data sets they were trained on. We can't control the training data set, as it is defined by the lab or company developing the model, but the web search optimisation is still relevant here.

Your website still needs to be fast to access, responsive to different screen sizes, and open to be processed by bots (crawlers), looking for information (here's a quick tool to check). Your content still needs to be high quality, showcasing expertise and authority in your field. You still need strong websites to link back to your pages, providing evidence that you are a reference to others.

What changes the most is how the information you are providing will be processed and displayed to users. Mark Edwards published an interesting reflection on how the user journey changes: “If a paragraph on one of our pages is lifted out of context, does it still make sense on its own? Does it say something that’s safe and accurate even when separated from the journey it belongs to?”

Bots will prioritise concise, direct content, but each paragraph should also stand alone, providing enough information without relying on what surrounds it. Avoid assuming every visitor knows what you mean or the context you are referring to, as well as not relying on visual explanations, which are usually ignored by crawlers.

The problem with PDFs

Most outputs produced by civil society organisations are reports, usually embedded into a page or available to download via PDF.

These are usually beautifully designed and packaged exactly how donors and peers are used to consuming them, but PDFs, interactive widgets and embedded content are not prioritised and sometimes even inaccessible to crawlers.

If there is one thing you can start doing today to appear in AI chat responses, it should be to stop publishing your reports and outputs as PDFs only, making them available as text on the page, or as HTML. This page should be formatted with SEO best practices such as structured headings (H1, H2, H3), conclusions at the end of each section, supporting details and references, as well as the famous schema markup that makes it easy for bots to identify different pieces of information.

The good news is that most of what makes a good website still applies. If your organisation has followed recent SEO guidance, you are already part of the way there. The additional step is ensuring your content can be read and cited by a generative engine without losing its meaning or credibility.

If your search traffic is declining or you are unsure how visible your outputs are in AI chatbot responses, it is worth seeking specialist support. Civil society organisations produce some of the most credible, evidence-based content available, but it might not be reaching the people asking relevant questions.