GEO — Generative Engine Optimization

If your customers do research before they decide, GEO is worthwhile for you. Questions like 'which one should I choose', 'how do they compare', and 'who should I ask' are now being put to AI assistants as well. Consulting, healthcare, education, B2B services, and technical products are the fields where questions like these are concentrated.

It has a payoff for a local business too: when assistants recommend a service in a given area, they can draw on business profiles, reviews, and the information on your site. The more consistent, verifiable information there is about your brand online, the more likely you are to be mentioned accurately.

Also referred to as 'GEO SEO' or 'AI SEO', this work isn't separate from traditional SEO; it's a continuation of it. If your site's basic SEO is lacking, that has to be completed first, because AI search optimization is built on a site that can be crawled and is considered trustworthy.

The aim of the technical setup is simple: AI tools should be able to reach your site, read the content, and not misunderstand who you are. On some sites, crawlers have been blocked without anyone realizing; on others, content loads only through scripts that run in the browser, so crawlers see an empty page.

If we built your site on a server-rendered stack, most of these conditions are met from the start. For sites on other platforms, we first audit the current state and rank the gaps by priority. In the setup, we do the following:

• robots.txt access for search crawlers (OAI-SearchBot, Claude-SearchBot, PerplexityBot) and classic search engines; crawlers used only for model training (GPTBot, ClaudeBot) are a separate decision
• Bing registration (Bing Webmaster Tools, IndexNow): Microsoft Copilot builds its answers from the Bing index
• An llms.txt file that summarizes what the site is about and its key pages
• schema.org structured data: organization, service, FAQ, and content markup
• Verifying that the content can be read without running JavaScript
• Making brand information consistent across your site, business profile, and other sources

Each tool has its own crawler and rules for finding sites. A site that is accidentally closed to a crawler can't be cited in that tool's answers. The table below is based on the platforms' own documentation:

Sources:Google — AI features and your website ↗Google — Crawlers (Google-Extended) ↗OpenAI — Bots ↗Anthropic — Crawlers ↗Perplexity — Crawlers ↗

When AI assistants generate an answer, they draw on passages that are short, clear, and make sense on their own. A page that asks the question in the heading, gives the answer in the first sentence, and then goes into detail is easier to cite than one that buries the same information behind a long introduction.

So we restructure content in an answer-first format: definitions, comparisons, step-by-step explanations, and question-and-answer sections. We back claims with sources, and expertise with author and business information. That's also the essence of what people search for as 'ChatGPT SEO': offering verifiable information an assistant can pass along with confidence.

Original information matters too. There's no reason to cite a page that repeats general facts found everywhere; observations from your own experience, descriptions of your process, the factors that determine your pricing, and clearly stated limits, on the other hand, can't be found anywhere else.

Research points the same way: in the first comprehensive study of generative engines, citing sources and adding quotations and statistics stood out among the methods that improved visibility, while keyword stuffing didn't help. Google's own guide also says there's no need to break content into chunks or write in a special way for AI; original content that is useful to readers is what counts.

Sources:Aggarwal et al. — GEO: Generative Engine Optimization (KDD 2024) ↗Google — Succeeding in generative AI search ↗

This field doesn't yet have measurement tools as mature as those of traditional SEO, and we say so up front. For testing, we first draw up a list with you of the questions your customers would actually ask: ones that ask about your brand directly, ones that ask about your service in general, and ones that ask for a comparison. The same list is asked again each period, so change can be tracked. The indicators available to us are:

• Testing the agreed list of questions in ChatGPT, Gemini, and Perplexity at regular intervals and recording whether the brand is mentioned
• Whether your site is cited as a source in the answers
• AI crawler activity in the server logs
• Visits from AI tools in Analytics (chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com, claude.ai)
• Search Console's generative AI performance report (impressions in Google's AI features; rolling out gradually)
• AI performance data in Bing Webmaster Tools (appearances in Copilot answers)
• Audit results for the technical signals (crawler access, structured data, llms.txt)

No one can guarantee a place in an AI assistant's answer. Answers vary with the question, the user, and the model version; the same question can produce different results on two different days. An offer that promises the top spot in the answers isn't realistic.

Assistants may also give incomplete or incorrect information about your brand. We can't correct that directly; what we can do is make the right information clear, consistent, and easy to find on your site and in other sources.

llms.txt is no magic file either: Google states plainly in its own guide that Search doesn't use such files, and other tools offer no official support either. We add it as a low-cost, harmless extra, not as a source of visibility. We'd suggest looking at one-off packages sold as 'AI SEO' in that light. Our job is to put in place everything needed for your brand to be understood correctly and cited as a source, and to track the outcome with regular tests. Nor is this a setup you do once and walk away from: crawler names and behavior change and new tools appear, so we re-audit the setup at regular intervals.