Your brand can rank highly on Google yet disappear from an AI answer. Conversely, a clear technical specification, an original investigation or an independent comparison can bring it into a conversation where nobody had mentioned it. For communications professionals, this is a major shift: being found is no longer enough; the brand must become usable as a source. That is the promise of GEO, or Generative Engine Optimization. Looking ahead to September 2026, this development deserves more than another miracle formula: it calls for a thorough rethink of how companies document their activities.
From search result to assembled answer
Traditional SEO seeks to make a page visible in search engine results. GEO aims to increase the chances that information will be reproduced, cited or used by a generative system. That distinction changes the work involved: users no longer receive just a list of links, but a synthesis that selects, connects and rephrases multiple sources.
The transition rests on developments that are already observable. Perplexity has made answers accompanied by sources a central feature of its product. Google launched its AI Overviews in the United States in May 2024, while Gemini forms part of its generative assistant strategy. OpenAI unveiled the SearchGPT prototype in July 2024, then integrated search into ChatGPT that autumn. These services do not all work in the same way, and their interfaces continue to evolve.
Above all, it is important to distinguish what a model learned during training from what a service retrieves from the web when a question is asked. A newly published page can inform a web-connected answer without changing the model’s internal knowledge. Conversely, allowing a training crawler access does not guarantee any citation. This distinction should precede any commercial discussion about GEO.
Being cited is not the same as being recommended
Imagine a company that sells heat pumps. Asked “what solution would suit an older house in a cold climate?”, an assistant might consult a public agency, an installer, a manufacturer and a specialist publication. The brand might appear among the sources without featuring in the final recommendation. It could also be mentioned without a link, or accompanied by a caveat about cost.
This is why generative visibility cannot be reduced to a ranking. Mentions, citations, accurate descriptions and recommendations must be distinguished from one another. An answer that cites your documentation to explain an incompatibility makes you visible, but does not necessarily benefit you. In communications, accuracy of representation matters as much as presence.
The real task: producing usable information
Much of the work resembles basic editorial housekeeping that has long been neglected. Essential specifications must be accessible on readable pages, not just in an image or a brochure from which information is difficult to extract. Dates, units, conditions of use and limitations must be explicit. A claim such as “responsible solution” offers little; an explained calculation method, with its scope clearly defined, provides verifiable material.
The technical foundations remain useful: crawlable pages, consistent internal links, stable URLs and relevant structured data. But no markup provides a universal passport into AI answers. Crawler access conditions and platform policies vary. Decisions about indexing, search and training therefore need to be treated separately, rather than allowing everything in the hope of gaining visibility.
Four priorities for communications teams
- Document the facts: publish primary sources, methodologies and results in context.
- Remove ambiguity: specify product names, relevant markets and validity dates.
- Make content maintainable: assign an owner and correct outdated information.
- Connect the evidence: make it easy for readers to trace a claim back to the document that supports it.
This approach does not require turning every page into a contrived FAQ. Instead, it encourages answers to real questions: who is the product suitable for, under what circumstances does it fail, and what alternatives exist? Clearly scoped, honest information is more useful than text saturated with superlatives.
Reputation is also built elsewhere
Brands do not control every document that might be consulted. News articles, scientific publications, reviews, forums and public databases can all contribute to an answer. Their weight depends on the service, the question and the available sources. GEO therefore intersects with media relations and reputation management, without being synonymous with buying exposure.
For communications professionals, this means offering journalists original data, making experts available and facilitating independent verification. Multiplying identical reproductions of a press release does not automatically create multiple pieces of evidence. Fake reviews and supposedly independent websites, meanwhile, add a risk of deception to a strategy whose effectiveness remains uncertain.
Measuring without creating false certainty
The ideal dashboard does not yet exist as a universal standard. The same question can produce different answers depending on the model, its version, the language, the conversation context or access to the web. Testing a query once and then announcing an “AI share of voice” provides a flimsy snapshot.
A more rigorous method involves compiling a set of questions representative of the customer journey, then repeating the observations. The date, service used, answer and displayed sources must be recorded. It then becomes possible to track the frequency of mentions, their accuracy, the competitors cited and the visits actually received. But a citation is neither a click nor a sale: attributing business results remains difficult.
Published research on GEO, including a study presented at the KDD conference in 2024, has explored the effects of changes such as adding references, quotations or statistics. These studies offer experimental avenues, not a law of the market. A result obtained using a particular corpus and setup does not prove that a technique will work consistently over time on ChatGPT, Gemini or Perplexity.
A new intermediary, new dependencies
The strategic risk extends beyond lost traffic. If the assistant becomes the first point of contact, it also chooses the criteria for comparison. A brand may have to explain why an overlooked metric matters more than a listed price. Even an impeccable source can be poorly summarized. It is therefore useful to establish a procedure for documenting errors, correcting the brand’s own pages and using the available reporting mechanisms.
What now? For September 2026 and beyond, the most plausible scenario is not the disappearance of SEO, but its integration with documentation quality, media relations and the measurement of generated answers. This is a projection, not a guarantee: platforms may still change their rules and interfaces. Nevertheless, brands have an immediate decision to make: invest in useful, verifiable information rather than opaque tricks. They will not control the AI’s answer; they can improve what it draws on.