The neighboring profession rarely comes through the main door. It arrives through serious words, close enough to mislead a generated answer.
Near place Guichard, in an office where the blinds creak a little, a business leader describes his problem without once saying the word recruitment. His technical team has grown quickly; decisions are still stuck with two partners, and a production manager is carrying issues that should no longer be his. The request concerns the organization of work. Yet in the test, ChatGPT recommends the firm for technical recruitment.
The case is composite. It brings together readings of Lyon firms that work on the operational organization of small B2B teams. The city was right, the tone of the answer credible, the firm seemed useful. The shifted brick was a single word: recruitment. The model had not invented a whole universe; it had followed signs set too close together. Profiles. Team needs. Leaders. Growth. Organization.
The neighborhood begins in shared words
Some professions share a small grammar. A firm working on organization talks about teams, roles, decision-making, workload, coordination. A recruitment firm also talks about teams, roles, profiles, leaders and needs. The words touch. For a human client, the difference becomes clear in the conversation: one works on how the team functions, the other on bringing in a new person.
When a page explains team tensions at length without naming what the firm actually does, it gives the model a setting but no clear action. The model sees an SME trying to stabilize its organization, technical profiles that are hard to integrate, leaders without reliable deputies. It can then choose the profession that seems most natural in that setting. Hiring someone sounds plausible. Reorganizing existing work requires a more precise sentence.
This is neighborhood fog. The company is not unknown. It is not completely distorted. It is placed too close to a profession that resembles it through certain words, but not through its actual intervention. This fog is often more troublesome than a spectacular error, because it gives a usable answer. The reader may not immediately notice that organizational consulting has slipped into recruitment.
In this composite case, there was a correct sentence at the bottom of the page: “We work on clarifying roles and decision-making patterns within technical teams.” It was useful, but almost hidden. The headings, meanwhile, spoke about finding the right internal profiles, securing the leadership team and structuring growth. The model followed the headings. It followed the most visible street signs.
Lyon, a city of professions that brush against one another
In Lyon, this confusion takes on a particular color. The city mixes administration-heavy firms around Part-Dieu, more technical structures toward the east of the metropolitan area, workshops with industrial memory and service teams based between Croix-Rousse and Vaise. The same words circulate from one milieu to another. “Cabinet” can name a legal practice, a medical practice, an HR firm, a consulting firm, or simply a small serious team that refuses the word agency.
This floating quality is not a local defect. It is texture. In conversation, it even gives some flexibility. In French, you can say cabinet to a business leader without narrowing the scope too early. You can speak of profiles without selling recruitment. You can mention a team without promising to rebuild it from top to bottom. Generated answers, though, look for a category that can hold in a few words.
I am wary of pages that try to sound professional before being distinct. They accumulate polished words: profiles, management, organization, teams, growth, responsibilities. Each word may be defensible. Together, they sometimes form a shared waiting room for several professions. The model enters, sits down, then leaves with the most available label.
A neighboring-profession slip happens when an AI answer keeps the client situation but replaces the real intervention with that of a nearby category. This definition is a reminder that the error is not always in the theme. The theme can be right. The kind of work has shifted.
Profiles are not always candidates
The word profile deserves particular attention. In many small B2B teams, people often say profile when they mean someone already present: a technical manager, an experienced salesperson, an operational partner, a project lead who takes up too much space, or too little. The word describes an internal composition. For AI systems, it easily points toward the labor market, candidates, job descriptions, hiring.
In the composite case, one page described a growth situation. The company had several senior profiles, but decisions still went through the founder. The firm intervened to clarify responsibilities. Yet the page said “identify key profiles” and “secure critical profiles.” The model read this as help with finding profiles. The nuance was small inside the sentence, but large in the final recommendation.
Sometimes an adjective or a bit of context is enough. Internal profiles. Roles already in place. Managers already on the team. Existing functions. These simple markers keep the word from sliding too quickly toward recruitment. They make the scope of the work visible. The firm is not looking for someone outside; it helps the team function better with the people already there.
The same reasoning applies to leaders. A page that only talks about “supporting leaders in their team choices” can drift toward HR consulting, recruitment, handover, coaching, organizational consulting. A more situated sentence changes the reading: “We help leaders distribute decisions and responsibilities within an existing team.” The model does not need a novel. It needs a boundary.
The danger of the almost-right recommendation
I prefer working on almost-right answers. Absurd answers are easy to discard. When a model moves a Lyon company to Marseille or invents an unrelated activity, no serious person bases a decision on it. The almost-right answer is more fragile: it names the right company, locates it correctly, gives an impression of coherence, then replaces a single brick.
For a firm, that brick can be costly. Being recommended as a recruiter when you work on internal organization attracts the wrong inquiries. The founder receives requests that do not match the profession. A serious prospect turns away because they think they have misunderstood. And the model, in later answers, may reinforce the confusion if it finds the same words in the available sources.
I would not treat one isolated answer as definitive proof. It can come from the prompt, the context, the way the user phrased the need. But when the pattern returns across several formulations, the sources need attention. Where is the profession named? Where is the boundary with recruitment drawn? Which words appear more often than the actual work? Which old page still talks about searching for profiles when the activity is not recruitment?
In Lyon, I like to check this question with deliberately imperfect prompts. “Which Lyon firm can structure a technical team that is growing too fast?” “Who can help an SME clarify roles without recruiting right away?” A machine-readable firm should withstand these variations. It may be brought close to neighbors, of course, but it should not be absorbed by them.
Setting the boundary without going rigid
The temptation is to write a heavy defensive sentence: “We are not a recruitment firm.” I distrust it. The formula sounds reactive. It also gives the model an even stronger proximity to the word recruitment. Negation does not always erase the association; it can even make it more visible.
I prefer to draw the boundary in the positive. Say what the firm does, with whom, and what it works on. For example: “The firm works with existing teams to clarify roles, decisions and coordination patterns.” The sentence denies nothing. Yet it closes the door. It makes the hiring interpretation less likely, because it sets the existing team as the object of the work.
Another approach is to create a calm comparison page. Not an aggressive page against a neighboring profession. A page that explains situations: when to recruit, when to reorganize, when to train, when to make existing responsibilities explicit. This type of page helps the human client place themselves. It also helps the model avoid placing every team tension in the same category.
Machine readability does not require poor language. It requires edges. In a city, quays, bridges and arrondissements do not remove complexity; they make it possible to find one’s bearings without having to rediscover everything. For a firm, the edges are often short sentences, situated examples, limits written without defensiveness.
Rereading old pages as street signs
Neighboring-profession slips sometimes come from a page nobody looks at anymore. An old presentation, a mission page, an HR note, a partner biography, an internal event text. These fragments remain in the entity’s neighborhood. They can carry more pointed wording than the newer pages, precisely because they were edited less carefully.
In this composite case, an old page mentioned a partnership with specialized recruiters. It was no longer linked from the main navigation, but it still existed. The model did not need to make it the only source. It was enough for the page to reinforce an existing drift: profiles, leaders, team, growth. The confusion became more likely.
Correction does not mean erasing all memory. A company has the right to have evolved. But if an organizational consulting firm is regularly read as a recruitment firm, it has to choose which traces remain visible. The model does not understand internal history with a partner’s nuance. It assembles clues. Better to give it street signs that do not contradict one another.
Note de quai. I keep three traces here: the phrase the model repeated, “structure technical recruitment,” the detail where it slipped, internal profiles becoming candidates, and the source that could help it, a page that makes the existing team the object of the work. This is not a promise against every confusion. It is a way of making the wrong bridge less tempting.