Since the worldwide rollout of Google AI Overviews in the spring of 2025, and the parallel rise of ChatGPT Search and Perplexity as serious gateways to information, one shift has become undeniable to any business leader watching their analytics: organic traffic is changing in nature. Impressions stay flat or even grow, but clicks drop the moment a synthesised answer appears above the classic results. The real question is no longer whether you rank, it is whether you are cited as a source inside the generated answer.
This shift already has a name in specialist publications: Generative Engine Optimization, or GEO. Behind the acronym lies a simple reality. The language models powering AI Overviews, ChatGPT and Claude do not rank pages the way a traditional search engine does. They extract passages, recompose them into an answer and credit a handful of sources. Being chosen as one of those sources does not follow exactly the same rules as ranking on the first results page.
The first criterion that stands out when you analyse which passages actually get cited is factual clarity. A paragraph that states a precise data point, dated and attributable, is mechanically more likely to be extracted than the same information buried in commercial prose. Models are looking for self contained sentences that can be inserted into an answer without heavy rewriting. Writing for generative AI means writing in blocks of meaning that can stand on their own.
The second criterion is documentary freshness. Generative engines favour content whose publication or update date is explicit, ideally both in structured data and visible in the body of the page. A case study dated 2023 on a fast moving topic will be systematically discarded in favour of an article published in the same year as the query. Maintaining a rigorous editorial calendar, and republishing strategic content with a genuine update rather than a cosmetic date change, becomes a direct investment in generative visibility.
The third criterion is semantic structure. Models rely heavily on HTML markup to understand a document. An article segmented with explicit subheadings, ordered lists where the logic calls for them, a meta description aligned with the first answer provided in the body, and correctly deployed JSON-LD structured data offers the model a reading plan that continuous prose simply does not. That work is invisible to the naked eye, but it weighs heavily on extraction decisions.
The fourth criterion is source credibility. AI Overviews avoid, as far as they reasonably can, citing pages with no identifiable author, no clear legal information, no visible editorial history. A professional blog that displays the author, their role, their contact details and their previous publications is treated very differently from anonymous content hosted on a shallow domain. The E-E-A-T signals Google formalised for classic ranking become, in the generative context, the primary selection criteria.
The fifth criterion, often underestimated, is brand presence in the training and retrieval corpora of the models. A site referenced by recognised sector media, mentioned on Wikipedia when legitimately relevant, present in serious professional databases, carries a footprint that models pick up when they generate their answers. That footprint is built slowly, through patient press relations, cross publications and genuine participation in the sector ecosystem.
A practical consequence emerges for SMEs and professional firms that invest in content. The logic of short, high volume publishing, long promoted by some agencies to occupy the long tail, is rapidly losing relevance. Generative engines reward depth, precision and a unique point of view. One documented, quantified, opinionated article on a narrow topic is now worth more than ten generic pieces that recycle information available everywhere.
Another consequence concerns transactional pages. A service page that stacks flattering adjectives has no chance of being cited, even on queries directly related to the activity. Rewriting these pages as genuinely informative documents that explain how the service works, its limits, its conditions and its expected outcomes, serves both the human reader and the extraction logic of the models. That is probably the highest short term return work for businesses selling custom professional services.
It is worth remembering, finally, that being cited in an AI Overview does not guarantee a click. A growing share of users consumes the generated answer without ever visiting the source. The strategy has to move beyond click metrics to include the brand equity generated by the citation itself. Seeing your brand associated, in an AI answer read by thousands of users, with a specific area of expertise, builds authority even without an immediate visit. This new visibility regime changes how the return on investment of a content strategy is measured.
At AKREA DIGITAL, this generative lens is now built into every editorial engagement. Our work in SEO and digital performance combines classic ranking work with generative engine optimisation, backed by structured data, rigorous semantic markup and content written to be quoted. Our digital audit now includes live tests across the main generative engines to measure your actual brand presence, and our content strategy team helps restructure existing blogs to maximise their extraction potential.