On stage, the same words keep coming up: revolution, acceleration, transformation. In the corridors, the questions are less lyrical: how much does it cost, who checks the answers, and what actually works? To stand out around the RAISE Summit, the Paris event dedicated to artificial intelligence, a brand must move beyond the competition to make the biggest promises. As business picks up again in September 2026, the communications challenge can be summed up as follows: make an experience concrete enough for someone to understand it, discuss it and, above all, verify it.
The challenge is no longer introducing people to AI
Since ChatGPT’s public launch in late 2022, generative AI has become a fixture in sales presentations, office software suites and corporate strategies. Announcements of multimodal models, capable of processing several types of content, followed by the growing focus on agents, have expanded the possibilities further. As a result, saying that a company “uses AI” now conveys very little information.
This ubiquity creates a trap for exhibitors, speakers and summit partners. Everyone wants to appear visionary; many end up sounding interchangeable. The same vocabulary, the same conversational interfaces, the same spectacular demonstrations. Attendees may leave feeling they have seen twenty different solutions tell a single story: tomorrow, everything will be faster.
The observations that follow draw on already documented developments in generative AI, rather than a report on the 2026 event. The perspective offered for September 2026 is forward-looking: the more similar the offerings become, the more the quality of the evidence becomes a differentiating factor. This is particularly true for a company that has neither the name recognition nor the resources of the major model providers.
Replace the promise with a workplace scenario
“We are reinventing customer relations” says almost nothing. Describing how an adviser finds a clause in documentation, drafts a response with sources, then validates it before sending already says much more. We can see a person, a task, a tool and a responsibility. The benefit becomes clear without asking the audience to believe in a historic breakthrough.
For a speaking engagement at the RAISE Summit, the right starting point is therefore a clearly defined workplace scenario. What happened before? What does the system do today? What remains in human hands? A convincing demonstration does not conceal these boundaries: it makes them visible, right down to the less spectacular steps of checking and correcting.
Evidence must stand up to scrutiny
A robust customer case study specifies the scope, the observation period and the comparison method. A claim of time savings remains a weak argument if it does not identify the tasks involved, the expected quality level or the time spent reviewing the output. An average can also conceal very different situations, from simple requests to complex cases.
- The baseline: which process serves as the reference?
- The result: are you measuring speed, quality or the volume processed?
- The oversight: who assesses errors, and against which criteria?
- The scope: which users and data are involved?
If not all the details can be published, the reasons must be explained. Commercial confidentiality and data protection may justify anonymisation. They should not serve as a smokescreen for an unverifiable claim. An honestly described methodology is worth more than a customer logo accompanied by a percentage without context.
Discuss the full cost, not just the model
One blind spot in the AI narrative concerns its day-to-day economics. The price of a subscription or a model call represents only part of the expense. Data must also be prepared, tools connected, access managed, outputs evaluated, teams trained and the system maintained as models or requirements evolve.
Communications therefore benefit from distinguishing between demonstration costs, deployment costs and operating costs. A prototype can produce a striking effect with a few selected documents. A service used every day must handle outdated information, ambiguous requests and incidents. This transition changes the nature of the project, and sometimes its economic value.
At an exhibition stand, the useful question is not just “how much does your solution cost?” but “how much does a correctly completed task cost?” This framing includes review, rework and cases handed over to a human. It also requires comparing the tool with a credible alternative: a search engine, conventional automation or improvements to the existing process.
Showing a limitation can strengthen trust
The factual errors of generative models, their sensitivity to instructions and data-related risks are documented. Retrieving documents to support an answer can improve relevance, but does not guarantee accuracy. Presenting these techniques as eliminating risk sets the stage for rapid disappointment.
A demonstration with real editorial substance should include a difficult case: a missing document, contradictory sources or an out-of-scope request. What does the product do then? Does it refrain from answering? Does it flag its uncertainty? Does it hand over to someone else? How a system behaves when it fails often reveals more about its maturity than its best answer does.
This choice takes courage from sales teams. Yet it allows the conversation to shift: instead of selling a supposedly infallible machine, they are offering a system whose risks are managed. In professions where an error can be costly, this distinction is a core selling point, not a caveat to be relegated to the small print.
Make transparency an editorial principle
The European Union’s AI Act, which entered into force in 2024 with phased implementation, has established a lasting framework for the debate. Without turning every presentation into a legal briefing, a company must be able to explain its role, its uses of AI and its oversight measures. The phrase “AI Act-compliant” is no substitute for a precise analysis of the applicable obligations.
The same standard applies to words such as “sovereign”, “responsible” or “secure”. Where is the data hosted? Who can access it? Is it used to train a model? What technical dependencies remain? These questions call for separate answers. Collapsing them into a reassuring adjective fosters confusion rather than trust.
For communications professionals, this means preparing presentations alongside product, security, legal and operational teams. The best spokesperson is not necessarily the one who predicts the future with the greatest confidence. It is the person who can connect an ambition to concrete decisions, acknowledge what remains experimental and provide access to people capable of answering questions beyond the pitch.
What next? As the next round of AI events approaches, the advantage could go to companies that turn their presentations into evidence dossiers: a demonstration in context, an evaluation method, full costs and explicit limitations. This is no guarantee of commercial success, but it is a credible way to cut through the noise. At the RAISE Summit, as elsewhere, the next truly distinctive message may be the simplest: here is what works, here are the conditions under which it works, and here is what we do not yet know how to do.
