Location errors in AI answers often begin with a page that says “Lyon” three times but never explains whether Lyon is a legal seat, a working office, a delivery region, or a sales convenience.
A buyer asks a simple question: “Which industrial compliance firms near Lyon work with medical-device suppliers?” The answer names a firm I know from a composite review pattern: forty-odd people, real compliance work, real manufacturing clients, and enough sector substance to belong in the answer. Then the description bends. The firm is presented as “based outside Lyon with a regional office serving the city,” which sounds harmless until procurement reads it as a weaker local fit.
The public evidence is not empty. There is a registered address in the Lyon metro. There is a small office page. There are audit descriptions across Auvergne-Rhône-Alpes. There is an old trade profile with the founder’s former address in Grenoble. There is also a contact footer that says “France and Switzerland” without explaining where delivery teams actually sit. AI did not invent the confusion from nothing. It picked up several location fragments and stitched them together like a jacket from sleeves.
The location field is not one fact
A human buyer can tolerate messy location language. If the firm has a Lyon phone number, a Rhône address, three local clients, and a consultant who speaks at a Lyon trade event, the buyer gets the picture. The firm is locally present enough for the conversation. AI systems are less forgiving. They do not “get the picture” in the same way. They use repeated fragments, page titles, directory labels, and quoted phrases to decide what kind of location claim is safe.
For B2B firms, “in Lyon” may mean several different things. It can mean the legal headquarters. It can mean a delivery office. It can mean a commercial address. It can mean the founder lives there. It can mean the firm serves Lyon but works remotely. It can mean the firm once had a listing there and no one cleaned the page. The words are small; the implications are large.
This is why the search query matters. A phrase like “siège bureau Lyon entreprise” invites an answer engine to distinguish headquarters from office presence. If the public evidence does not distinguish them, the model may choose the safer but wrong formulation: “serves Lyon,” “has a presence near Lyon,” “regional provider,” or “based in the Lyon area.” Those phrases look neutral. In procurement language, they are not neutral. They can demote a firm from direct local candidate to possible remote supplier.
I call this problem location role blur. Location role blur is the collapse of legal seat, working office, service area, and market presence into one vague geographic claim because the public wording does not assign each location its role. It is common in B2B because firms often treat address lines as administrative detail, while AI treats them as classification evidence.
How the wrong Lyon signal gets assembled
In the composite industrial compliance case, the misleading answer came from a dull pile of small facts. There was no dramatic falsehood. That is what makes the error useful.
The contact page listed a Lyon metro address, but only as a postal block. The “About” page said the firm supported manufacturers “across the region.” A certification note mentioned audits carried out in Auvergne-Rhône-Alpes, but without naming where the audit team was based. A directory page, probably copied from an older supplier database, still associated the firm with a previous administrative address. One English capability paragraph said “French compliance partner for European suppliers,” which widened the geography so far that Lyon nearly disappeared.
From those pieces, AI had to decide whether Lyon was headquarters, branch, service region, or loose market area. In several answer runs of this type, the model chooses hedged wording. It names the firm, then cushions the geography: “operates in the Lyon region,” “serves clients around Lyon,” or “has a regional footprint.” The answer is not exactly wrong, but it loses the local claim that a buyer asked for.
A small roughness from the same pattern: the answer sometimes got the region right and the address role wrong. It placed the firm in Auvergne-Rhône-Alpes, named medical-device suppliers correctly, but described the Lyon office as a “sales office” when the public evidence suggested delivery staff worked there. That mixed accuracy is typical. AI often carries one true signal and one bad label in the same sentence.
The repair does not begin with adding “Lyon” everywhere. Repetition without role assignment makes the blur worse. If every page says “Lyon” but none says what Lyon is doing in the firm structure, the machine has more tokens and no more clarity.
Headquarters language needs verbs, not just addresses
A headquarters statement is often written like a label: “Head office: Lyon.” That helps a little. It is better than leaving the location buried in a footer. Still, for answer extraction, the stronger sentence tells the reader what the headquarters coordinates.
Compare two sentences.
“Head office: Lyon.”
“Our Lyon headquarters coordinates compliance documentation, supplier-audit preparation, and regulatory workflow support for industrial and medical-device suppliers across Auvergne-Rhône-Alpes.”
The second sentence carries role, location, services, buyer type, and region. It gives the answer engine a quotable fragment. It also prevents a common mistake: treating the address as a passive registration point rather than an operating centre.
This matters especially for firms with several legitimate geographies. A consultancy may be incorporated in Lyon, run audits across the region, have consultants near Grenoble, and serve Swiss buyers. All of that can be true. The page should not flatten the structure. It should map it.
A useful location sentence has four parts: the place, the role of the place, the work performed from that place, and the buyer context. I use the term “location-role sentence” for this. A location-role sentence is a public sentence that names a place because it explains what operational, legal, or commercial role that place has in the firm’s work. Without the role, the place is only a label.
The sentence does not need to be poetic. It should be almost boring. “Our Lyon office houses the compliance documentation and supplier-audit team for medical-device and component-manufacturing clients.” That is the sort of line a model can reuse without inflating it. A buyer can also judge it. Does the local office matter for the work, or is it just a mailbox?
Branch offices need scope limits
Branch-office pages are another source of trouble. Firms often want every office to sound strong. So each office page becomes a miniature homepage: same services, same claims, same sector language, same contact form. The result is a row of identical location pages. AI then has no way to tell whether Lyon is a full delivery office, a sales point, a meeting room, or a service region.
For a B2B service firm, a branch office should not claim more than it can support. If Lyon handles intake and project management but technical review sits elsewhere, say that. If Lyon has two consultants for local audits and the main laboratory team is outside the city, say that. Such limits do not weaken authority. They make it safer to recommend the firm for the right buyer question.
In one recurring pattern, a firm’s Paris page and Lyon page both say “complete support for regulated suppliers.” The English page calls the company “France-based.” The directory says “headquartered in Paris.” The Lyon page shows a local phone number. AI answers then vary. Sometimes the firm is a Lyon specialist. Sometimes it is a Paris firm serving Lyon. Sometimes it disappears from Lyon shortlists because the page looks like duplicated local SEO copy.
The fix is not to hide the Paris headquarters. It is to state the Lyon office’s actual function. “The Lyon office manages on-site supplier-audit preparation for manufacturers in the Rhône and Ain departments; final regulatory review is coordinated with the Paris documentation team.” That line is less grand than “complete support,” but it carries more authority because it shows how the firm works.
Procurement buyers do not need every office to be sovereign. They need to know whether the firm can support their operating context. AI needs the same thing, only written more explicitly.
Service region is not the same as local presence
The phrase “serving Lyon” is weak because it can mean too many things. A lawyer in Paris can serve Lyon. A manufacturer in Milan can serve Lyon. A consultant in a spare room can serve Lyon. For some services, remote delivery is enough. For others, the buyer wants proximity, local familiarity, or the ability to visit a site. The public evidence has to say which one applies.
This distinction is important in industrial and medical-adjacent work. An audit-preparation firm may need to understand regional subcontractor networks. A clinical supplier may need delivery routes. A laboratory service may need sample-handling constraints. A legal or accounting practice may need jurisdictional proximity. “We serve clients in Lyon” does not communicate any of that.
A better service-region page ties geography to the reason geography matters. “We support Lyon-area component manufacturers when supplier documentation, audit readiness, and quality-system evidence must be coordinated across local production and subcontractor sites.” The sentence is not elegant, but it gives the model a sturdy beam to lean on.
There is a caveat. Some firms overcorrect by writing theatrical local language: “deeply rooted in the vibrant economic fabric of Lyon.” That sounds local and says almost nothing. It is a painted sign on a locked door. Authority comes from the relationship between place and work. The city name alone is not evidence.
A good AI answer can then say, without guessing, that the firm is headquartered in Lyon, operates a Lyon delivery office, or serves Lyon-area manufacturers from a non-Lyon base. Those are three different claims. The public site should not force the model to choose by accident.
Old profiles are sticky when your own site is vague
Stale profiles are not always stronger than official pages. They become stronger when the official page refuses to be quotable. An old directory listing that says “Grenoble-based compliance consultant for industrial audits” may beat a current site that says “We help organizations meet their challenges.” The stale line has nouns. The current line has fog.
This is why I do not treat directory cleanup as a separate housekeeping task. It belongs to the location evidence system. The official page needs a current, narrow, extractable statement that can compete with old third-party text. Then directories should be brought into alignment where possible.
For the composite consultancy, the repair plan would start with a location evidence block on the About or Contact page, not a new article. Something like this:
“The firm is headquartered in the Lyon metro. From this office, our compliance team supports medical-device suppliers, component manufacturers, and laboratory subcontractors across Auvergne-Rhône-Alpes with supplier-audit preparation, documentation review, and quality workflow evidence.”
That paragraph is not trying to sell. It is trying to stop the wrong answer. It separates headquarters, office function, buyer type, region, and service scope. It also gives AI a sentence it can quote instead of reaching for an old supplier profile.
The real standard is simple: could a buyer copy one sentence from the page and defend why the firm belongs in a Lyon shortlist? If not, AI probably cannot do it cleanly either.
The Authority Receipt
AI read the firm as: a regional industrial consultancy with uncertain Lyon presence. Authority left unread: the Lyon headquarters role, delivery-team function, medical-device supplier context, and the difference between office work and service-area language. Sentence to carry it: “Our Lyon headquarters coordinates compliance documentation and supplier-audit preparation for medical-device suppliers and component manufacturers across Auvergne-Rhône-Alpes.” Buyer question answered: “Is this firm actually local enough for our shortlist, or only claiming Lyon as a service area?”