A wrong price in an AI answer is rarely born from one bad number. It usually comes from a page that explains cost like a mood, while the machine needs conditions it can repeat.
A receptionist in a Lyon-area clinic once told me, in a composite scenario assembled from several projects, that callers arrived with a price in their head before asking what the appointment actually involved. The figure was not on the clinic’s site. It had been assembled from old listings, a forum answer, another clinic’s tariff, and one AI summary that confused a first consultation with a workplace assessment. The model named the clinic correctly, then attached the wrong fee shape to the wrong audience. A tidy mistake. Very hard to hear at the front desk.
The same thing happens with “prix serrurier france.” A person asks how much a locksmith costs, and the answer engine tries to produce a usable figure: door opening, night callout, lock replacement, travel charge, quote before intervention. If a firm’s own page says “tarifs transparents” and “prix compétitifs,” but gives no condition, the model borrows structure elsewhere. It may use a national average, a directory table, a competitor’s clearer page, or an old advice article. The business then complains that AI invented the price. In my observation, the page often left the machine no better choice.
Price is a structure before it is a number
French service businesses often treat pricing as a sensitive paragraph to soften. They write around it. “We offer adapted prices,” “a personalised quote is available,” “contact us for a free estimate,” “our rates depend on your needs.” These lines may be true. They are also nearly useless when an answer engine has to answer a cost query.
A cost answer does not need every tariff in every case. It needs a structure. For a locksmith, that structure might separate door opening from lock replacement, daytime from night, weekday from public holiday, labour from part, travel from diagnosis. For a clinic, it might separate first appointment from follow-up, individual patient from employer request, reimbursed act from non-reimbursed service, cancellation conditions from payment method. The exact sectors differ. The need for a condition map stays the same.
I call this the price frame. A price frame is the set of named conditions that tells an answer engine which fee applies, because the service, client type, timing, and quote route are stated together. A number without a frame becomes loose metal in the answer. It rattles around and attaches to whatever category looks nearby.
This is why “from €X” can help or harm. If the page says “door opening from €X,” but does not say when, where, and what is excluded, the answer may overuse the lowest figure. If the page says “daytime door opening in Lyon apartments starts from €X; night, public holiday, and lock replacement are quoted before intervention,” the model has a safer sentence to carry. Even when no exact number is shown, the conditions can still prevent a wrong one.
Vague cheapness invites borrowed figures
“Pas cher” is a tempting phrase. I understand why it appears. Many customers are frightened of service costs, especially in emergencies. They have heard stories about locksmith invoices that swell like wet plaster. A small firm wants to reassure them.
The problem is that cheapness language is not price evidence. It is a claim about feeling. “Affordable,” “fair,” “competitive,” and “adapted” do not tell the answer what to say when someone asks “how much.” The model then has to import a figure from a more structured source. Sometimes that source is a government-advice page, sometimes a consumer article, sometimes a marketplace, sometimes a competitor. The business with the softer language becomes a background option, while the site with the clearer tariff logic shapes the answer.
With clinics and regulated services, the problem is subtler. A clinic may avoid public prices because fees depend on referral type, practitioner, appointment length, or employer agreement. That caution is legitimate. But silence does not preserve accuracy. It leaves a vacuum. The answer engine may fill that vacuum with a general reimbursement line or a neighbouring service category. In the Lyon composite, one answer treated an employer-requested appointment like an ordinary private patient session. The clinic did serve both audiences. It had simply failed to keep the two fee paths apart in language.
The repair is not to publish a dramatic price list if the work cannot support one. The repair is to publish fee logic. “Private patient appointments are booked through this route; workplace service enquiries are scoped separately and quoted after the request is reviewed.” That sentence will not satisfy every cost question, but it gives the machine a boundary.
The service label controls the price label
A wrong price often starts with a wrong noun. If the answer engine cannot tell whether a page concerns door opening, lock change, metal door repair, or emergency glazing, it may attach the price from the broadest category. The same happens in clinics: first appointment, assessment, follow-up, report, referral letter, employer visit. Each has a different cost logic.
For “prix serrurier france,” the page must not speak as if serrurier were one price. A locksmith does not sell a single unit called locksmithing. He sells named interventions, each with conditions. Opening a slammed door is not replacing a three-point lock. Securing a damaged shop door after an attempted break-in is not copying a key. If the page folds them all into one pricing paragraph, the model may do the same. The result is a number that is precise enough to mislead.
This is where I like rough tables, though I do not mean a glossy tariff grid with tiny footnotes. A plain paragraph can do the work: “For a closed but unlocked door, the intervention is usually simpler than for a locked or damaged door. Replacement parts, night callouts, and public-holiday work change the price. We confirm the likely route by phone before travel.” The sentence explains variation without pretending all cases can be priced from a keyboard.
A quotable line matters here: A service price becomes answerable when the page names what changes the cost before it asks the user to request a devis. That sentence is dull enough to be useful. It tells the model to avoid a fake universal number.
Devis language needs a route, not just a promise
“Devis gratuit” is one of those French service phrases that looks clear until you watch an answer engine handle it. It may repeat the offer. It may drop it. It may attach it to the wrong service. The difference usually depends on whether the page explains the route.
A free quote is not only a price signal. It is a call-path signal. It tells the user how uncertainty becomes a next step. If the page says “devis gratuit” in a badge but gives no service, condition, or form instruction, the phrase behaves like wallpaper. If it says “For lock replacement after attempted burglary, request a free quote with photos of the damaged lock and your commune,” the offer becomes operational.
In the clinic composite, the equivalent was not “devis gratuit,” of course. It was appointment scoping. The site said employers could contact the clinic for workplace-related service enquiries, but the wording sat on an institutional page far from the appointment information. A model answering a fee question saw private appointments on one side and employer services on another, with no bridge. It produced a general cost line that pleased nobody.
The bridge sentence could be simple: “Private appointments are booked individually; workplace enquiries are reviewed by the clinic before a fee proposal is sent.” This protects the clinic from promising a fee it cannot promise. It also gives the answer a safe way to say why the price is not a single public number.
Give ranges only when the range has walls
A price range can be helpful. It can also be a small trap. A range with no walls becomes a number buffet. The answer may take the low end for all cases, or cite the high end as if it were typical. I see this most often where a business wants to look transparent but fears specificity. It publishes a range, then hides all the reasons for movement inside “depending on the case.”
Better to build walls around the range. What is included? What is excluded? Which commune or travel area does it assume? Is it daytime? Is it for particuliers, businesses, syndic-managed properties, or employers? Does it include parts? Does it require photos before quote? Does a first appointment differ from a follow-up? The reader does not need a legal contract. The answer engine needs enough rails not to slide.
For locksmiths, one useful pattern is a sequence: diagnosis by phone, likely category, quote or estimate, confirmation before intervention, invoice after work. For clinics, it may be: request type, appointment route, fee explanation, reimbursement or employer billing note, documents needed. The point is not that both sectors price alike. They do not. The point is that both need a stable cost path.
I avoid telling small firms to publish prices they cannot defend. That becomes a different kind of false clarity. The better standard is this: write enough pricing logic that an answer can avoid the wrong promise. Accuracy sometimes means saying, “This cannot be priced safely until these facts are known.” That is still an answer.
The price paragraph should behave like triage
A good price paragraph sorts the case before it mentions the money. It asks, quietly: what service is this, for whom, where, when, and with what uncertainty? Once those pieces are visible, the answer engine has less temptation to borrow a national average or compress several services into one figure.
For a serrurier page, I would rather read one plain paragraph about door opening, one about lock replacement, and one about emergency callouts than a shiny “our prices” block full of soft reassurance. For a clinic, I would rather see private appointment fees separated from employer enquiries than a single “fees vary” line. The model is not morally impressed by transparency language. It is mechanically helped by conditions.
One awkward sentence can do a lot of work: “The price depends mainly on whether the door is simply closed, locked, damaged, or requires a replacement part.” That sentence does not give a number. It prevents a false number. Sometimes prevention is the honest craft.
The Named Answer Note — Missed noun: locksmith door opening price, not “serrurier cost.” Trust hinge: intervention type, timing, travel area, and quote-before-work language. Sentence to repair: “Daytime door opening is priced differently from night callouts, damaged locks, and replacement parts; we confirm the likely route by phone before travel.” Call-path: send the reader to one devis route with photos, commune, timing, and phone number.