AI SEO – What Is It and How Will It Impact Your Business?

The rise of AI tools – specifically generative AI (GenAI), which creates content, mostly text but not limited to that — including the capabilities large language models offer in information retrieval, has given birth to concepts known under various acronyms: AI SEO, AISO, GEO, SAE, LLM SEO… 

Regardless of the acronym used, it boils down to ranking up in AI tools like ChatGPT, Perplexity, Gemini (formerly Bard), DeepSeek, Claude, and others, though many of these have less significance due to lower user interest.

From the perspective of business owners, for whom visibility on Google has so far been a key channel for reaching and acquiring customers, involving significant time and money, this creates a serious dilemma. Is it still worth investing in “classic SEO”? Is AI SEO the future, and should we shift efforts and budgets there? If yes, then what should be done to appear in ChatGPT search results?

You may also read: Will Artificial Intelligence Replace Google?

Making the right decisions is complicated by marketing hype around AI products, as well as a flood of training courses, tools, and services portraying AI as a complete revolution and paradigm shift in all marketing activities.

It may seem that AI-powered search engines (more precisely, large language models, LLMs) are a completely new way of consuming resources accumulated on the internet. However, it’s more of a technological evolution than a revolution. Contrary to popular belief, Google was the leader and pioneer of this evolution.

Early search engines, including Google, were lexical search engines. They relied on the presence of specific words, or, more precisely, character strings, in the documents they scanned. Because of this, they struggled with grammatical variations, which led to unnatural keyword phrases in SEO texts like “seo company New York” or “men barber London”.

The growing number of texts written in an unnatural way was problematic for Google. Another issue was ambiguous user queries — without knowing exactly what the user searched for, it was impossible to serve satisfactory search results. Google’s business model relies on user satisfaction: if users return regularly, Google collects data about their interests, which it can then monetize through Google Ads.

In 2011, Google took its first steps beyond classic search with Knowledge Graph – a structured knowledge base enabling direct answers to at least some queries. 

Google implemented its first AI algorithm in 2015 – seven years before ChatGPT’s launch. This was RankBrain, a machine learning system designed to understand the intent behind search queries to better match results. 

In 2018, Google introduced BERT (Bidirectional Encoder Representations from Transformers), an algorithm allowing the understanding of query context. Note that the “T” in BERT stands for “Transformer,” the same as in OpenAI’s models. GPT means Generative Pre-Trained Transformer. 

Transformer architecture is a topic for a separate article, but it’s important to note that the technological foundation for OpenAI was developed and consistently implemented by Google, albeit in different ways over many years. Only in 2023, after ChatGPT’s surprising success, did Google start adopting OpenAI’s solutions, initially through a separate chatbot called Bard (later renamed Gemini), and through search result summaries initially named SGE, now known as AI Overviews (launched first in the USA in May 2024 and in Poland on March 2, 2025).

This shows that the dichotomy between Google and AI search engines is false. Google is increasingly becoming an AI search engine. 

How does this impact SEO? 

A clear example is keyword optimization. While 15 years ago, unnatural keyword phrases like “best pizza delivery New York” or “plumber Chicago services” had to be stuffed into pages, today’s search algorithms understand that “pizza delivery” and “food delivery” or “plumber” and “plumbing services” are semantically related. Optimization now involves not only exact keywords but also semantically related terms. 

You may also read: How to Use AI for SEO? 5 Best Use Cases

Of course, there are differences between Google and chatbots like ChatGPT or Perplexity. GenAI search generally works in one of two ways:

Classic AI chatbots like ChatGPT (without Search enabled) or Claude use training data. In a nutshell, they use a text corpus that the model was trained on, including internet data (CommonCrawl). 

Importantly, this is not live data. New information published on the Internet must first be scraped (downloaded) and then used to train the model, which also takes time. As a result, the “knowledge” of such a ChatGPT is always at least several months behind – for the latest models from OpenAI, it goes back to the autumn of 2024. So we will not get up-to-date information from the training data models on what happened after that date.

It’s different for engines like Perplexity or ChatGPT if the Search module is enabled, or for AI Overviews. These tools are based either on a self-built index or on an index created by traditional search engines; for example, ChatGPT is based on Bing’s index. So when a user types a query into SearchGPT, the system queries Bing’s index for that phrase, collects the best-matched subpages, and generates a response based on those.

Note: The RAG technique can also be used to build internal tools or website chatbots that answer user questions based on your documents, blog posts, or other content attached as knowledge bases. 

Importantly, because this will determine which actions can provide greater exposure in GenAI search engines – language models operate based on transformer architecture. Simply put, blocks of text are first tokenized – split into small fragments called tokens. A token can consist of anywhere from 2 up to about 5 characters.

Next, the model analyzes which groups of tokens appear together. The statistics of these occurrences help establish relationships between them. Words (token sequences) that frequently appear next to each other are likely related (e.g., “SEO” and “services”), while words that often occur in similar contexts tend to be synonyms (e.g., “SEO” and “search engine optimization”).

Based on this, words are converted into numerical vectors (embeddings), allowing calculation of cosine similarity between words or larger text sequences. Next, attention mechanisms detect relationships between words in larger groups. For example, given these sentences:

The model can recognize that although the word “bank” is the same, in the first sentence it means the edge of a river, while in the second it means a financial institution, based on the surrounding words and context.  

Statistical frequency and similarity help predict what words should come next in generated text; this is how chatbots create answers. When using RAG, the model’s default statistics are modified by embeddings extracted from documents identified as most relevant to the query. 

Effective SEO in GenAI search engines should therefore be based on several principles.

You may also read: How to Build Local Business Visibility in AI-Powered Search

The AI SEO principles described do not differ significantly from current Google SEO. Content is still fundamental: bots must first crawl, index, and display it. Content should be as useful and helpful as possible, specifically answering user intent.

External validation like backlinks is still important, although LLMs do not consider backlinks directly – but…

Brand mentions are becoming “new backlinks.” Google patented the treatment of brand mentions as links back in 2014. Google now accurately recognizes entities like company or person names and can associate them with specific domains.

The more mentions your brand has as a good choice for X or Y (where X and Y are your target keywords), the more sources LLMs learn from that your brand is a correct answer for those queries. More such mentions can indirectly improve your Google rankings even without direct backlinks. 

SEO for Google ≠ SEO for ChatGPT, but Google requires more work and factors, including behavioral data gathered via Chrome browser usage (greatly simplified). LLMs do not have this, nor need it, since users typically remain on chatbot pages.

For example, Google takes into account behavioral factors, i.e., data about users’ behavior on the site collected using the Chrome browser (I’m simplifying heavily here, but this text is already long anyway). From LLM’s perspective, it’s not quite there, but there’s no need for it either – the user will mostly stay on the chatbot’s site anyway. Thus, factors like page speed or mobile layout are less critical for LLMs, provided their bots can correctly retrieve your content.  

Here we move to translating AI SEO activities into measurable marketing and business KPIs. The picture is mixed: Google traffic will fragment across other search engines, including GenAI-powered and social media platforms like Instagram and TikTok, which are also introducing AI search features. In fact, no internet search engine today can be said to not rely at least partially on AI. 

We can measure some GenAI traffic with Google Analytics. For example, Looker Studio dashboards can clearly show this, but page links appear in only about 25% of answers. No link means no click and no measurement. Our AI traffic report shows that artificial intelligence is a tiny fraction of total website traffic, growing dynamically but unlikely to exceed 5% by year-end.

Traffic alone is not the point. A good analogy is social media profiles: a post on Facebook with a link may or may not bring users to your site. However, mere exposure in a feed increases brand awareness.

Similarly, being cited by ChatGPT or in Google AI Overviews gives your brand an additional touchpoint. To leverage this, highlight your brand effectively.  

This changes marketing KPIs: 

Dedicated tools for monitoring AI chat visibility are emerging.

The evolution of search engines requires changes in marketing strategies, especially your marketing funnel at all stages. Users may look in ChatGPT for general ideas (e.g., “how to increase online store sales”) or for specific suppliers (“best marketing agencies in Poland”).

The shift will mostly affect content. On the one hand, you need content that AI can learn from and cite as a source (brand exposure). This requires building a topic base defined by parameters like SiteRadius and SiteFocus, then writing concise yet comprehensive content.

On the other hand, if your goal is to attract users to your site for conversions (like newsletter sign-ups) – which chat windows can’t do – your content must have something unique. It could be substantive expertise, but edutainment (education + entertainment) works better. Humor evokes emotions, and emotions improve memorability. AI can’t yet craft good jokes well, and it won’t gain this skill overnight. 

Online activities outside your website will also need some love. AI search engines don’t rely on links but on brand mentions. ChatGPT will use many sources – your site, social media, business directories (used heavily in SEO years ago), review sites, media mentions, Reddit, Quora, etc. Ensure existing sources have updated info and that new ones are regularly added. 

The rise of social media forced additional activity on those platforms. The same applies to AI search engines. ChatGPT won’t eliminate the need for a website or replace Google, but will coexist alongside them. Your online marketing must incorporate this new channel. The key to business success is not to rush to appear first in AI search results, but to leverage synergy across all channels to reach your customers.