A procurement buyer does not need a firm to sound impressive in the abstract. The buyer needs enough hard edges to compare it: who it serves, what it handles, where it works, and where the scope stops.
The page I am looking at says “personalized support for complex needs.” There are two locations in the Lyon metro, a bilingual site, and several specialist practitioner biographies that contain the real evidence. The public-facing service page, though, is soft. In a composite scenario based on clinic-group visibility work, an AI answer recommends three better-known names and skips this sixteen-person professional clinic group entirely. It is not absent because it lacks competence. It is absent because its competence is scattered.
The buyer question is not a consumer question. It is closer to procurement, referral, partnership, or specialist selection: which Lyon provider can handle a regulated service for a defined patient or organizational context? When the AI system looks for shortlist material, it finds staff credentials here, location details there, a stale directory elsewhere, and an English page that says more than the French one. No single paragraph behaves like a capability statement. So the model chooses firms whose public pages are easier to compare.
A shortlist needs edges
A capability statement is a compact public description of what a firm can do, for which buyer or user, under what constraints, and with what proof. It is not a slogan, because a slogan asks to be admired; a capability statement asks to be checked.
That difference is larger than it sounds. A slogan says “expert care for every need.” A capability statement says the clinic group provides regulated specialist consultations across two Lyon metro locations, with named practitioner qualifications, bilingual intake, and defined referral pathways for a particular treatment area. One sentence opens the drawer. The other tells the buyer there may be a drawer somewhere.
AI systems, especially in search-assisted answers, do not simply list the most deserving firms. They assemble candidates from evidence that can be compared. If one firm offers clear scope and another offers atmosphere, the clear firm often wins the shortlist. This is irritating when the atmospheric firm is actually excellent. It is also repairable.
The first edge is buyer type. Who is the page written for? A patient, a referring clinician, a corporate buyer, a laboratory partner, a regulated supplier, a procurement officer? Many professional-service pages try to speak to everyone. That widens the voice and weakens the category.
The second edge is operating context. Is the service delivered in clinic, on-site, remotely, across multiple locations, in French and English, under a regulated protocol, inside an industrial production chain, or before a formal audit? This context is often where AI decides whether the firm fits the prompt.
The third edge is proof. Certifications, practitioner biographies, case studies, association memberships, and named service limits all count. But they count more when the page explains what they prove.
The missing sentence makes the firm hard to compare
In the clinic-group scenario, the biographies were stronger than the service page. One practitioner profile named a regulated treatment area. Another mentioned hospital experience in careful terms. The English patient page explained intake better than the French page. The map listings gave two addresses. An old directory used a broader wellness label that was no longer accurate.
A human reader with patience could build the case. A machine did something thinner. It described the group as a general clinic option in Lyon, then preferred other names for specialist queries because their pages stated the scope more plainly. In one run, the model got the number of locations right and still softened the service category. That is the kind of imperfect answer I pay attention to. It shows the entity was recognized, but not trusted enough for the shortlist.
This often happens when the key sentence is missing from the public page. Not the key idea. The key sentence. AI extraction is sentence-hungry. It likes claims that can be lifted, compressed, and attached to the firm without too much interpretation.
A useful sentence might look like this: “Our Lyon metro clinic group provides regulated specialist consultations across two locations, with bilingual intake and practitioner-led assessment for patients referred by clinicians or seeking a defined treatment pathway.” That sentence is still only a teaching example. A real clinic would need exact, supportable wording. But it shows the structure: place, provider type, regulated scope, language surface, referral or buyer context.
The sentence does not try to win affection. It gives the shortlist something to hold.
The four-line capability test
When I review a page for shortlist use, I do not begin by asking whether it is persuasive. I ask whether it can survive a comparison table. AI systems often behave as if they are building a rough comparison table in the dark. A firm with clean fields has an advantage.
I use a simple internal test called the four-line capability test. The page should be able to answer four lines without forcing the reader to gather clues from five places: buyer or user, service scope, operating constraint, and proof. If those four lines cannot be written from the page, the firm is asking AI to infer too much.
Buyer or user is the group the firm is actually fit for. Service scope is the thing it does and the thing it does not claim to do. Operating constraint is the condition that makes the work specific: regulated treatment, industrial documentation, bilingual intake, occupied-site work, export support, or supplier audit readiness. Proof is the public evidence that makes the scope credible.
A Lyon B2B firm can be locally strong and still fail this test. It may have a page full of polite paragraphs, testimonials from unnamed clients, a certification badge with no explanation, and a service menu written in internal language. The buyer senses competence. The machine sees loose surfaces.
There is a danger here. Some firms respond by overloading the page with every possible credential. That creates a different problem. The page becomes a cupboard with everything thrown forward. AI may extract the wrong item. The better repair is not more evidence everywhere. It is cleaner evidence in the sentence where the buyer question arrives.
Procurement language is not ugly if it is accurate
Many firms avoid procurement-readable wording because it feels dry. I understand the instinct. Nobody wants a website that sounds like a tender appendix. But there is a middle register that serious buyers appreciate: precise, calm, and bounded.
For a consultancy, that might mean naming the sector and engagement type. For an industrial supplier, it might mean naming production context, certification relevance, and delivery constraints. For a clinic group, it might mean naming the regulated service area, practitioner qualification surface, location structure, and intake conditions without making medical claims the public evidence cannot support.
The point is not to imitate procurement documents. It is to borrow their clarity. A procurement buyer using AI to shortlist vendors is not asking for a poem about excellence. The buyer is trying to reduce uncertainty. A capability statement reduces uncertainty before the sales conversation begins.
The statement should also carry limits. If the clinic group does not provide a certain treatment, the page should not imply it. If the consultancy works with medical-device suppliers but not pharmaceutical manufacturers, that distinction may matter. AI systems make mistakes when limits are absent. They stretch the nearest category until it covers the question.
A bounded sentence can actually increase trust. It tells the model what not to infer.
The bilingual problem can split the shortlist
In Lyon, many B2B and professional firms have bilingual surfaces that do not quite match. The English page may be written for international buyers and therefore names sectors, client types, or formal service contexts. The French page may sound more familiar, more local, and less specific. Or the reverse happens: the French page carries the real professional vocabulary, while the English page flattens it into general consulting or care language.
For AI shortlists, this matters because the prompt language changes the evidence path. A French query may lean on French pages and local directories. An English query may lean on international-facing descriptions. If those two surfaces tell slightly different stories, the firm can appear more specialized in one language than the other.
In the clinic-group scenario, the English page carried stronger intake language, while the French page assumed the reader already knew the service category. That assumption may be safe for a local human. It is not safe for answer extraction. AI does not always import the missing specificity across languages, and when it does, it may borrow too much.
A capability statement should exist in both languages, not as a stiff translation but as matched evidence. The French line and the English line should carry the same buyer type, scope, constraints, and proof. Style can change. Meaning should not leak.
The same principle applies to industrial suppliers and consultancies. If the English page says “medical-device suppliers” and the French page says “entreprises industrielles,” the AI answer in French may lose the sector that gets the firm shortlisted.
A page can invite selection without begging for it
The best capability statements do not shout. They behave more like a good label on a laboratory sample: enough information to prevent the wrong handling. That is the tone I look for. Clear, specific, calm. No inflated promise. No fake breadth. No hidden proof.
A useful page for AI shortlisting usually has a top paragraph that carries the firm’s category, buyer type, location, and constraint. It has a second layer that explains proof: certifications, practitioner qualifications, association memberships, case contexts, or delivery examples. It has service descriptions that match the category instead of drifting into different language on every page. It gives the answer engine a stable object to cite.
This is not only for machines. Human buyers benefit from the same structure. The difference is that humans can forgive missing links. They can call, ask, infer, and lean on reputation. AI systems answer before that conversation begins. Their shortlist becomes the first gate.
So the capability statement has a small but serious job. It tells the machine the firm is selectable for a particular buyer question. It gives the buyer a reason to click. And it keeps the firm from being passed over merely because its authority was written as atmosphere.
The Authority Receipt
AI read the firm as: a general Lyon professional provider with two locations but no clear specialist shortlist role. Authority left unread: regulated service scope, practitioner-led assessment, bilingual intake, and referral context. Sentence to carry it: “Our Lyon metro clinic group provides regulated specialist consultations across two locations, with bilingual intake and practitioner-led assessment for defined treatment pathways.” Buyer question answered: “Can this provider be compared seriously, or is it only another local name?”