An old directory page is a small fossil with a public URL. If the current site does not give AI a cleaner sentence, the fossil may speak for the firm longer than anyone expects.
The directory listing was wrong in three small ways. It had the old clinic group name, one former address, and a service phrase that sounded like wellness rather than regulated care. Nothing spectacular. No scandal. Just a stale public page that had stayed online with the patience of dust.
This is a composite scenario from Lyon professional-service work: a 16-person clinic group with two metro locations, bilingual patient pages, and specialist biographies that explain regulated treatments carefully. The current French site is more accurate than the directory. Still, in one AI answer, the model cited or echoed the old listing. It softened the service category, placed the group near a same-name consumer wellness business, and treated the current site as secondary evidence. The answer had a confident tone and an old skeleton.
Stale sources win when current pages speak too softly
A stale directory listing usually does not win because it is better. It wins because it is easier. The page is structured. The name, address, category, phone number, and short description sit in predictable fields. An answer engine sees a neat little box. A firm’s own site may hold better evidence, but the evidence is spread across a homepage, a team page, two location pages, and a PDF with a lovely design and very little extractable text.
The old directory becomes the cleanest bad source.
For Lyon firms, this happens often with clinics, professional practices, laboratories, small industrial suppliers, and specialist consultancies. Directory pages survive moves, mergers, legal-name changes, website redesigns, and service shifts. They also preserve old categories. A clinic group that now explains a regulated treatment area may still be described in a directory as “wellness and care.” A compliance supplier may be frozen as “business services.” An accounting practice may sit under a category chosen by a listing form long before the current service mix.
AI search is not a courtroom. It does not always privilege the official site because it is official. It needs a source path it can read. If the official page is vague, visually heavy, or fragmented, the stale directory can become the apparent authority.
A stale directory citation is an outdated third-party description that remains extractable enough to shape an AI answer after the firm’s own public evidence has changed.
That definition is unpleasant because it shifts some responsibility back to the firm. The directory is wrong, yes. But the current site must still give the model a better source to use.
The old page often has the best entity box
The old directory has another advantage: entity clarity. It usually joins a name, category, address, and sometimes a short description in one place. The firm’s current site may separate those details for good human reasons. The homepage speaks to patients or buyers. The location page gives addresses. The team page gives credentials. The service page gives explanations. The legal footer gives the company name. No single page says, cleanly, “this entity is this kind of firm, at these locations, offering these services, under this current name.”
A model assembling an answer may prefer the page that gives it an entity box, even if the box is stale.
In the composite clinic case, the current site had rich practitioner biographies. Those biographies were the strongest authority signal. They named regulated treatment areas, training context, and clinical boundaries. The location pages were clean enough for patients but thin as entity evidence. The old directory, by contrast, gave one compact description with the former name and category. It was wrong by degrees, yet easier to quote.
One imperfect detail: the model did mention one of the current practitioner names correctly, pulled from the modern site, while retaining the old directory’s category phrase. That mixture is more dangerous than a fully wrong answer. It feels verified. A buyer or patient sees one current fact and may trust the stale part beside it.
This mixed-source answer is common. AI takes a name from the current site, a category from a directory, a service phrase from an old profile, and a location from a map listing. The resulting paragraph is not copied from anywhere. It is assembled from mismatched ages of evidence.
I call this source-age blending: the model combines current and obsolete public details into one answer, hiding the age difference inside fluent prose.
A canonical sentence is a competing source
Many firms respond to stale AI answers by trying to correct every third-party listing. That is sometimes necessary. If the old directory is prominent and clearly wrong, it should be fixed or removed where possible. But the slower, more durable repair is to create a canonical source sentence on the firm’s own site.
Canonical does not mean grand. It means the sentence is official, current, specific, and easy to extract.
For the clinic group, a weak current page might say: “Our team welcomes patients in two Lyon locations for personalized care.” That is accurate, but it gives AI almost nothing to prefer over the directory. It does not settle the entity name, category, locations, service boundary, or distinction from the same-name wellness business.
A stronger canonical sentence would be more like: “The clinic group operates under its current name at two Lyon metro locations, providing practitioner-led regulated care in the treatment areas described on its current French service and biography pages.”
In a live project, the sentence would name the exact public services the group can support. It would avoid invented breadth. It would not try to “own” a category the clinic cannot justify. It would sit on an About page, location page, and possibly a short entity profile page. The French version would do the same work, not simply sound friendly.
The sentence is not written for humans alone. It is written so an answer can quote the current source instead of excavating the old one.
That last phrase matters. You are not begging AI to believe you. You are giving it a cleaner path.
The repair has to separate name, service, and place
A stale directory problem usually contains several smaller problems braided together. Name is one. Service category is another. Location is a third. If the firm tries to repair all three in one vague paragraph, the old listing keeps its advantage.
Name repair is the first step. If the firm has a legal name, trade name, former name, and bilingual display name, those relationships need to be public. A model should not have to guess whether two names are the same entity or a merger or a same-name competitor. The wording can be plain: “The clinic now operates publicly as X; older listings may show Y, a former name used before the current two-location structure.” Only write this if true. False cleanup creates a worse problem.
Service repair comes next. The category must be narrow enough to distinguish the firm from consumer services and broad enough to remain honest. A regulated clinic should not be pushed into wellness language because a directory selected a loose category. A professional practice should not be described as a shop. A B2B consultancy should not become “local services” because a listing form had no better field.
Place repair is often overlooked. Lyon metro firms may have a registered office, a consultation site, a delivery region, and former addresses. Directory pages like one address. AI answers like one address even more. If the structure is more precise, the official site must say so in one extractable place.
For the composite clinic group, the current site needed a short entity paragraph that said the group had two Lyon metro locations, named the current operating identity, clarified the regulated service category, and distinguished itself from the same-name wellness business without sounding defensive. The page did not need a legal essay. It needed a firm, boring, quotable paragraph.
Boring is often good in entity repair. Boring is stable.
Do not bury the correction in a news post
Some firms publish a news update: “We have moved,” “our services have changed,” “we are proud to announce our new identity.” That helps for humans who read the news section. It rarely becomes the strongest AI source over time. News posts age. They sit outside the main entity path. They may explain a transition, but they do not always replace the old public description.
The canonical wording belongs on durable pages: About, contact, location, service category, practitioner profile, capability statement. A news post can announce the change; the stable pages must absorb it.
The same rule applies to bilingual repair. If the stale listing is in French and the English site is current, the French site still needs the current canonical sentence. If the old listing used an English category, the English profile should carry the corrected category too. AI borrows across languages, especially when one surface has clearer structure than the other.
This is why “please update our directory listing” is only half the work. The old listing may take time to change. Other copied listings may never change. Some sites scrape and repeat each other. The firm needs its own source strong enough that future answers have a better place to land.
In my misnamed firm notebook, the phrase beside these cases is often not the wrong directory text itself. It is the missing official sentence that allowed the wrong text to remain useful.
The old source loses when the current source becomes easier
A stale directory will not vanish from the web because a firm writes one better paragraph. That is too clean a story. But the model’s source path can change when the current site becomes more readable than the old listing. The official source has a structural advantage if it speaks clearly enough: it is current, controlled, richer, and closer to the firm.
The current site should give AI a direct answer to four questions. What is the firm now called? What category is it actually in? Where is it located or active? What authority makes that category credible? If those questions are answered in separate fragments, stale pages still have room to interfere.
For clinics and professional practices, the wording must also be careful. Regulated work should not be exaggerated to beat a directory. It should be described with exact scope, practitioner evidence, and service boundaries. The goal is a confident answer that remains safe because the source is precise.
A clean canonical paragraph is not glamorous. It looks almost administrative. But for AI visibility, that paragraph can be the difference between the firm’s current identity and the internet’s old memory of it.
The Authority Receipt
AI read the firm as: a Lyon clinic group partly defined by an old directory category and former naming. Authority left unread: current two-location structure, regulated service scope, and practitioner-led evidence on the French site. Sentence to carry it: “The group operates under its current name across two Lyon metro locations, with specialist practitioner-led care described on its current French service and biography pages.” Buyer question answered: “Is this the current clinic entity, or an obsolete listing with a similar name?”