An AI answer may keep the company’s name and lose the shape of its work. The shift is quiet: a design assignment becomes general technical support.
In a composite case, the question starts with an industrial workshop between Bron and Vénissieux, on that eastern edge of Lyon where design offices, loading docks and small business parks face one another without quite belonging to the same world. The need is precise: find a team able to design automation logic and prepare the supervisory layer before an integrator steps in. ChatGPT names a small Lyon design office, mentions industrial projects, and places the company correctly in the region. Then it sums up the service as “general technical support”, as if it were a general-purpose technical provider. The firm mainly offers a design assignment. The model kept the city, but put the work on the wrong shelf.
The scene resembles certain shopfronts on the Presqu’île, where you only understand the business after stepping inside. Outside, a broad word: workshop, studio, office, consultancy. Inside, a precise practice. In generated answers, many service companies stay on the threshold. The model reads the sign, adds two pieces of context, and builds a plausible category. It holds together, but not well. A client looking for automation design does not need a technical provider in general. They need to know where the work begins, where it stops, and who takes charge of what.
Over-broad verbs shift the work
A service assignment often changes category before the model even enters the scene. It happens in the company’s own texts. To avoid closing the sales conversation too soon, service pages use open-ended verbs: support, lead, intervene, assist, help. These verbs are not false. They are roomy. They let several trades step in at once.
In the composite case of the design office, the commercial page did speak about automation, but also about scoping, support for teams and project follow-up. An old profile described the company as a “technical partner for industrial workshops”. A post mentioned a project carried out with an integrator, without specifying that the integration belonged to another party. None of this was untrue. Yet together, the pieces made the category porous.
Design needs firmer words. To design, specify, size, formalise, document, prepare the supervisory layer, describe automation logic: these verbs give the work a shape. They say that the company is not merely standing beside the client. It produces an architecture, a decision scheme, a base for implementation. Even if integration comes later with another party, the design assignment has its own substance.
I notice that many service pages avoid this substance. They prefer to speak about value, fluidity, coordination, time saved. The human reader may understand, especially after already speaking with the team. A response model, though, classifies. When the page does not provide precise professional gestures, it stores the company in the nearest and most available category.
The fog of neighbourhood
The fog of neighbourhood appears when the model recognises the company, but pulls it too close to an adjacent trade when describing its service.
In this family of errors, the model does not erase the company. It shifts it one notch. A design office becomes an integrator. A scoping assignment becomes technical support. Preparation for the supervisory layer becomes a general service around a workshop. The neighbouring word is not entirely wrong; that is why it settles so easily. It gives an answer that is acceptable at first reading and imprecise at the second.
This fog is common in B2B services, because the borders are often clarified in conversation. A client calls with a poorly named request. The team reformulates. One discovers that the need is not “installing a system”, but understanding how the system should be designed. Or the reverse. This clarification happens in human conversation. On the site, it sometimes remains implicit.
The model does not like the implicit. It can fill the gap, but it fills it with nearby categories. If the page says “we support industrial companies in their automation projects”, the answer may imagine a provider who advises, integrates, follows up, repairs, coordinates. If the page says more clearly “we design automation principles and supervisory materials before implementation”, the neighbourhood remains visible, but the border begins to show.
The correct city is not enough
Many teams feel reassured when an AI answer places the company in Lyon. I understand the relief. In imperfect local maps, the city name is often the first check. Yet an exact city can hide a false category. The model knows where to put the company, but not always which professional category it belongs to.
In Lyon, this phenomenon takes on a particular cast because professional neighbourhoods are mixed together. Around Gerland, Villeurbanne and the eastern part of the metropolitan area, you find design offices, workshops, integrators, technical consultancies, suppliers and maintenance teams. The vocabulary overlaps. Everyone talks about industrial projects, shop-floor constraints and operational teams. For a local client, the nuance can be clarified with two questions. For a model, it may remain in the mist.
In the composite case, the answer used Lyon as a seal of plausibility. In substance, it said: Lyon-based provider, industrial projects, technical support. The triptych sounds credible. The heart of the assignment is missing. A recommendation built this way can steer a prospect toward a conversation that is too broad. The company will then have to correct the record: “We can scope and design, but we are not the same kind of provider as a generalist integrator.”
That correction in conversation costs little once. It costs more when it keeps returning in generated answers. It shows that the public sources do not carry the difference clearly enough. The city is stable; the category moves.
A design assignment needs objects
I often ask teams to show me the objects of the work. Not only the benefits. The objects. A diagram, a specification, an operating logic, a supervisory scenario, a scoping note, a constraint matrix, a document handed over before integration. These objects cannot always be published in detail. They can be described with care. Their presence changes the page.
A service becomes more readable when it lets the reader see what it produces. “Technical support” describes a posture. “Design of an automation architecture before deployment” describes an object. “Shop-floor support” opens several trades. “Formalising supervisory needs for industrial teams” tightens the category. The text does not need to become heavy. It needs to give the model fragments that are less interchangeable.
In industrial engineering, caution is normal. Companies do not want to expose client details, processes or sensitive diagrams. But confidentiality does not always justify blur. One can describe the type of assignment, its limits, the general deliverables, the points of intervention. A well-written proof page does not reveal a workshop’s secrets. It tells the reader, and the model, where in the trade the company sits.
The old vocabulary may remain in the background. A company may have been an integrator at the start, a partner to an integrator, or a support provider around a solution. These traces belong to its history. They become troublesome when they still seem to define the main activity. Here again, the problem is not the trace. It is the trace’s status on the page.
Limits make the category more stable
Commercial texts often fear limits. They want to show that the team can adapt. I understand that. A small firm also depends on atypical cases, conversations where the need changes, assignments that begin with a diagnosis and end somewhere else. But an AI answer cannot handle this flexibility like a founder on the phone. It flattens it.
A well-worded limit does not reduce the company. It prevents the company from resembling three neighbours at once. For a design office, saying that the work mainly covers scoping, design and formalisation before implementation may be more useful than a broad sentence about “support for industrial projects”. If the company works with integrators without presenting itself as an integrator, that nuance deserves a sentence.
I also look at implicit comparisons. When the site mentions distributors, installers, design offices or consultants, it must avoid mixing their roles. A model easily picks up neighbouring terms. It sees the words together, then brings the categories closer. A page that explains the neighbourhood can help: the situations for which the office is suitable, when a client would be better off looking for another type of provider, how the interventions fit together.
This clarity does not need to be aggressive. In a city like Lyon, technical trades cross paths constantly. The right phrase resembles a calm separation on a map: here the quay begins, there the ramp descends, farther on the bridge leads to another use. The reader understands where to walk.
Return to the sources that carry the classification
When an answer changes the category of a service, I return to the sources that carry the classification: service page, About page, old profiles, short presentation texts. I am looking less for a brilliant phrase than for a stable phrase. It must be possible to repeat it without adding a neighbouring profession.
In the case of the Lyon design office, the reference page had to make three things hold together. First, the moment of intervention: before implementation, when the architecture and supervisory layer are taking shape. Then, the objects delivered: specification, operating logic, scoping note, supervisory materials. Finally, the limit with integration: the team may work with an integrator without becoming an integrator in the answer.
This phrase does not prevent every slip. No page fully controls a generated answer. But it gives the model a more honest grip. It no longer needs to store the company under “general technical support” to complete what the source has not said.
The machine readability of a company is its ability to be recognised, located and distinguished in answers generated by models. In this case, the name and the city already hold. The trade distinction, though, calls for a less porous page.
Quayside note. I keep here the correct name, the correct city, and the work placed on the wrong shelf. The model has not forgotten the company; it has brought it too close to a convenient neighbour. A more precise page can give other anchor points: design verbs, delivered objects, a limit with integration. The fog does not clear all at once. Sometimes it simply stops pushing in the wrong direction.