Your brand publishes a rigorous study, your competitor writes a loosely argued opinion piece, and the assistant cites… a third player. That is the headache conversational search engines are creating for communications departments. When an answer replaces a results page, visibility no longer means simply ranking well in search: you need to become a source the machine can find, understand and select. By September 2026, this battle could weigh heavily on reputation strategies. But nobody has a “cite me” button.
From blue links to composed answers
The shift began to take shape in 2024. Google launched its AI Overviews in the United States, incorporating generated summaries into some searches. OpenAI unveiled SearchGPT, then launched search in ChatGPT. Perplexity had already popularised an interface combining written answers with references. These documented developments are the starting point for this forward-looking analysis for September 2026, not a verified inventory of the features available at that date.
The user journey is changing in practical terms. A buyer no longer simply searches for “supplier management software”. They might ask: “Which providers are suitable for a French manufacturing SME, with European hosting and rapid integration?” The tool reformulates the need, brings information together and proposes a shortlist. The brand may appear in the text, feature among the sources or disappear entirely.
Communications professionals therefore need to distinguish between three objectives: being mentioned, being cited as a source and being recommended. They are not interchangeable. An assistant may cite a company to explain a controversy without endorsing its products. It may also recommend an offering based on a third-party comparison, without linking to the company’s website.
Visibility that cannot be commanded
The terms “generative engine optimization” and “answer engine optimization” sometimes promise a new frontier to conquer. They describe useful practices, but do not constitute an established science. Systems differ in their models, accessible sources, search mechanisms and interfaces. The same question can produce different answers depending on the context, date or wording.
A distinction must also be made between knowledge acquired during model training and documents consulted when answering. Publishing a page today does not mean it will become part of the system’s knowledge tomorrow. Nor does its accessibility to a search engine guarantee that it will be consulted or cited. There is no universal procedure for forcing your way into an answer.
An academic study on GEO, made public in 2023 and subsequently presented at KDD in 2024, explored the effect of editorial changes on visibility in generated answers. In particular, it examined the addition of references and data. These experimental results suggest possible approaches; they do not prove that any formula will work consistently across all commercial services.
The raw material: facts that are easy to verify
The first task looks less like a creative campaign than a documentation overhaul. What exactly does the company do? Where does it operate? Which commitments have already been met, and which remain goals? An elegant but vague corporate page offers little help to a system tasked with answering a specific question. It is just as unhelpful to a journalist in a hurry.
Take a manufacturer claiming to offer “more responsible” packaging. The phrase leaves essential questions unanswered: more responsible than what, within what scope, and according to which method? A dated fact sheet specifying the composition, recycling requirements and limitations of the assessment provides more usable information. While it does not guarantee a citation, it reduces ambiguity.
- Identify sources: author, role, publication date and relevant contact details.
- Document claims: methodology, scope, references and limitations.
- Keep information consistent: names, figures and descriptions aligned across materials.
- Make access easier: readable pages, clear headings and essential information available as text.
The technical foundations remain important: permitted crawling, indexing, internal links, performance and appropriate structured data. But markup is not a passport to inclusion in a generated answer. Decisions concerning bots must also distinguish between search and training, where services allow it. They have implications for editorial strategy as well as content protection.
Media relations regain strategic value
A brand speaks about itself; a newsroom, a laboratory or a public authority lends information a different standing. Assistants may draw on these sources to put an offering into context or corroborate a claim. This strengthens the case for a presence in relevant publications, without allowing anyone to claim that securing an article will automatically generate a mention.
The right question is therefore not simply “how much coverage?”, but “what reliable information about us is circulating?” A detailed industry investigation, a technical interview or an independent test may do more to build understanding of a company than a string of identical reproductions of a press release. A third party’s independence is precisely what makes it valuable: it rules out complete control over what is said.
Conversely, churning out fake comparisons or supposedly independent content creates reputational risk. Even if manipulation temporarily fools a system, it undermines trust when discovered. Communications teams are better served by producing accessible evidence, not an artificial veneer of consensus.
Measuring without manufacturing misleading rankings
The instinct will be to ask for a dashboard: “What is our share of voice in AI?” The metric can be useful, provided its limitations are made clear. Testing ten questions selected because they favour the brand does not measure its actual visibility. It mainly measures the quality of the scenario prepared for the meeting.
A more robust approach is to assemble a panel of questions drawn from real-world interactions: sales enquiries, journalists’ questions, customer objections and sensitive issues. Tests need to be repeated, dates, services and query conditions recorded, and mentions, citations, tone and factual errors assessed separately. The results describe an observable sample, not the full range of possible answers.
Visits from these tools and any resulting conversions complement the analysis, without explaining everything. Someone may discover a brand in a summary and then contact it directly. Conversely, a prominent citation may generate no useful traffic. Appearing in an answer remains a means to an end, not a business outcome in itself.
A cross-functional undertaking, not another gimmick
The communications department cannot handle this transformation alone. SEO specialists contribute technical expertise; product teams verify specifications; legal teams review commitments; customer service exposes misunderstandings. Together, they can correct contradictions in publicly available information and establish monitoring for recurring errors, without promising to control every answer.
What next? Between now and September 2026, and beyond, the challenge may be less about “pleasing AI” than making the company understandable in an environment of automated mediation. Starting with a few strategic topics, publishing verifiable evidence and monitoring answers offers a reasonable way forward. Citations will remain uncertain. The quality of the information, however, remains a very human responsibility.
