If you’ve spent any time on LinkedIn recently, you will have seen various acronyms being thrown around for what is fast becoming a necessary extension of SEO – AEO, GEO, LLMO, AIO, WTFOIT (OK, I made that last one up – clue: the last 3 letters are ‘optimisation is that!?’, I’m sure you can guess the rest …)
They essentially mean all the same thing – optimisation with the aim of becoming more visible in AI-generated answers, whether that be directly in search in the form of AI Overviews or AI Mode, or in conversational generative AI platforms such as ChatGPT, Gemini, or Perplexity. We happen to have adopted GEO at Red Stag Media, for various reasons, but call it what you will.
But what does it mean for companies that are looking to remain visible to farmers online as they research products and services for their farming operations?
The reality is that search habits are changing. In everyday life, the phrase ‘just Google it’ is being replaced with ‘just ask ChatGPT’. This presents a number of challenges and opportunities for ag businesses, with some of the main ones being outlined below.
So how do you actually go about improving your visibility in LLMs? What are the key things you need to know?
First, we need to cover a few concepts you may or may not be aware of. Some of this can get quite technical, so I’ll explain it as plainly as I can.
Worry not – this sounds more complicated than it actually is. Generative engines build responses from two main sources – in-model data (i.e. the data that the model was trained on, for example websites, forums, academic papers, books, etc …) and grounding (live data feeds, search engine indexes, social media, etc …) Grounding is a process that ensures that the generated answers are ‘grounded in reality’ by supplementing the training data with fresh information. Not all queries require grounding, but many do.
RAG is the method in which the answers are ‘grounded’ – Retrieval Augmented Generation retrieves documents which are relevant to the query and feeds them into the model, which then generates an answer using the additional context.
Another technical-sounding word, but a concept we all use in our daily lives when trying to arrive at a sensible answer.
If a farmer fired up ChatGPT and asked it, “What are the best spreader settings for my granular nitrogen fertiliser?”, just like if the farmer asked you the same question, the LLM would need to know some additional information, which it does by using fan-out queries.
Fan-out queries are simply additional queries which are used to help break a complicated question into smaller pieces so that it can answer the main question more accurately.
Example fan-out queries for this question might be:
You can see why the LLM knowing the answers to the above questions will help provide a more accurate and comprehensive answer to the farmer’s original question.
As much as we like to think that AIs are just digital versions of a human brain, this isn’t exactly the case. They’re still machines, and machines rely on algorithms.
Vector embeddings are mathematical representations of meaning. Words and phrases are vectorised, which essentially means that they’re turned into a set of numbers. Similar things sit close together, and different things sit far apart.
Carrying on with the granular fert example, ‘granular nitrogen fertiliser’ sits near ‘urea application rates’ but far away from ‘Texas style BBQ’.
It’s a way for machines to understand meaning, not just words.
Turning this into something arguably more actionable, Cosine Similarity is how LLMs decide which pieces of content are most relevant to a query by comparing how close their meanings are in an embedding space.
You can assess the relevance of your content (or a specific ‘chunk’ of your content – see below) by comparing the cosine similarity to the query you want to be returned for.
LLMs, instead of retrieving whole pages, break content into passages, or ‘chunks’ and retrieve the most relevant segments.
A simple way of defining a chunk is a section of content designed to satisfy a user’s intent. That could be answering a specific question, for example, and might be 2 – 3 sentences in length.
Because of the way that LLMs use chunks, it’s important that each one within your content provides enough context to be used as a standalone snippet. Each snippet should be very focussed on a single concept, without any ‘fluff’ in the copy.
For example, on a page about Red Stag Media, the following chunk doesn’t provide enough standalone context:
“We deliver high-performing campaigns for our clients with a laser focus on ROI. Our senior team combine their skills to ensure the most strategic, holistic activation possible.”
A much better version of this would be as follows:
“Red Stag Media is a UK-based PR and digital marketing agency that delivers ROI-focussed campaigns to business in the agriculture sector. Our team includes former senior journalists, editors, and highly experienced digital marketers who have worked with some of the country’s biggest brands. We specialise in combining PR and digital strategy to drive measurable results.”
The improved version includes:
A page which contains many chunks, written in this way, will be infinitely more useful to an LLM compared to a page comprising chunks written like the first example.
Now you know some of the key concepts which relate to content, you should be able to create new pieces and optimise existing ones more effectively.
For new content creation, we have put together the following graphic, which outlines the process step-by-step.
When followed, using the guidelines outlined in this article, you should end up with content you can publish safe in the knowledge that LLMs will be able to extract it well, with all the information and context it needs.
So far, we have focussed mainly on content creation and optimisation. However, if an LLM can’t crawl your site properly and extract information from it, the rest is irrelevant.
The technical side of things is a huge topic, and for the average marketer working in an agribusiness, not very exciting!
That said, it’s worth quickly covering some of the basics.
LLM bots do not execute JavaScript as Google’s crawler does. Although there’s a two-step process employed by Googlebot, it can still discover your content, even if it’s served by JavaScript. LLMs either don’t execute it at all or have limited capabilities to do so.
A lot of websites use JavaScript to different degrees, so it’s important for it to be rendered in the correct way to work for LLMs.
In a nutshell, avoid client side rendering, instead use server side rendering. Simply put, server side rendering ensures that any content served by JavaScript will be visible in the HTML code and therefore visible to LLM bots for extraction.
Some web developers will review server logs and see that a website is being hammered by bots and in response, will block them.
If some of those bots happen to be LLM bots, those LLMs will not be able to discover or extract your content.
A quick way to check is to view your robots.txt file. These can be found by going to your website’s home page and pasting /robots.txt next to your URL. For example, https://redstagmedia.co.uk/robots.txt
If you see something like the below, it means your robots.txt file is blocking ChatGPT’s crawler from accessing your content.
User-agent: GPTBot
Disallow: /
Structured data is code that helps crawlers understand more about the content on a website. It’s essentially a labelling system which provides information in a machine-readable format and is used by search engines and LLM crawlers to better understand content.
Schema.org is probably the best-known and most widely supported structured data markup system, so it probably makes sense to start there. There is a lot of information online about it, and your web developer will certainly be able to talk you through the options for implementation.
The final leg of the GEO optimisation ‘tripod’ is trust signals.
LLMs want to return accurate answers, so trust in the sources they use is critical.
In traditional SEO, authority was largely driven by high-quality external links pointing at a website, whereas in GEO, citations are king.
The good news is that citations happen as part of a well-rounded marketing programme, which most brands have in place.
Creating content in relevant places such as social media plaforms, Reddit, YouTube, etc … is a good starting point. Having consistent PR coverage in industry publications and local news outlets (and of course mainstream national titles if possible!) shows LLMs that your brand is credible and relevant.
User-generated content is also well-received by LLMs, so influencer marketing can be useful.
The main aim is to get your brand mentioned in different places alongside competitors, key product category terms, and industry-relevant phrases.
GEO is becoming a necessary extension of SEO. SEO is not dead, and although click-through rates are dropping in the search results, traditional content strategies are still viable and can deliver great commercial value if executed well.
With LLMs using grounding for many queries, search engines like Google become a source for information, and therefore, it’s still massively important to be visible in them through a solid SEO programme.
That said, there’s no getting away from the fact that the starting point of online content discovery for many people is now an LLM, not Google search. Data suggests that they’re using LLMs in addition to search as opposed to instead of it, but being visible in both is the future of search.
If you need any help assessing your current visibility in LLM platforms and/or implementing a holistic, future-proofed organic search strategy, please get in touch.
