A Paris specialist usually disappears quietly. The model still names a category, so the answer looks acceptable, but the firm’s real work has slipped under the floorboards.
On a damp afternoon in the 11th, I sat with a composite creative strategy studio whose work was easy to understand in the room and strangely hard to find on the page. Fourteen people, a narrow set of clients, a reputation that travelled through cultural institutions, hospitality groups and a few luxury-adjacent projects nobody wanted to describe too loudly. When we asked an AI system to describe the studio, it did not insult them. That would have been easier to fix. It called them “a Paris-based creative agency.” Correct, and almost useless.
The founder winced because the phrase was familiar. It appeared in old bios, directory summaries, a press caption, and one tidy homepage sentence written during a quieter year. Their portfolio told a better story: spatial identity for cultural venues, launch narratives for hotel groups, naming work for discreet hospitality projects, strategy before visual polish. The problem was that the better story lived in fragments. A model had found the largest bucket and dropped the firm inside it.
The category is where AI starts when evidence is thin
When AI describes a company wrong, the mistake often begins with something that feels too small to matter: the category noun. Agency. Consultancy. Clinic. Studio. Practice. Startup. These words are useful door handles. They help a person enter the room. They are also blunt instruments when they become the whole description.
Paris makes this bluntness more visible. In another city, “creative agency” may already narrow the field enough for a first pass. In Paris, the phrase can cover a branding studio in the 10th, a communications office near the 8th, a production-heavy team by the périphérique, a cultural strategy group in the 11th, or a design partner working mostly through referrals. The category is a corridor with too many unmarked doors.
In my work, the recurring pattern is simple. The firm’s own pages use broad category language in extractable places, while the precise language sits in project captions, PDF decks, founder interviews or internal sales phrases. AI systems tend to prefer the wording they can safely repeat. If the homepage says “creative agency in Paris” three times, and the portfolio says “we helped a museum group clarify its visitor-facing service language” once, the generic line usually wins.
The mechanism is not mysterious. Models assemble descriptions from visible, repeated, structured evidence. They do not attend the dinner where a client says, “They are the team you call when the brief is half cultural, half commercial, and politically delicate.” They read what is published.
A specialist firm gets flattened when its strongest proof is present but its specialist name is absent.
Paris firms often hide the exact work in polite places
There is a particular Paris habit I keep finding in founder-led firms: precision appears in conversation before it appears in public language. The founder can explain the firm in one breath over coffee near Oberkampf, yet the website retreats into safer nouns. “Strategy.” “Design.” “Advisory.” “Care.” “Support.” The more refined the work, the more the page sometimes sounds like it is trying not to bother anyone.
This restraint has social intelligence. Paris clients often read overclaiming as a warning. A studio that says it “defines culture” too loudly may lose the very clients it wants. A consultancy that performs scale before proving judgement may feel imported, even when it is local. A clinic that writes like a billboard risks flattening its own professionalism. The human reader understands the restraint.
The model does not always understand restraint. It sees missing fields.
In the 11th-studio composite, the best phrase came from an old project note, almost thrown away: “creative strategy for cultural and hospitality teams that need naming, positioning and visitor-facing language before design production.” It was a little long. It had no varnish. It carried the exact work, the client type and the sequence of value. I would never put that sentence alone on a hero panel. I would put a tightened version near the about page, service page, portfolio introduction and founder bio, because those are the places a machine is likely to read.
Generic-category collapse — this is my working term — is the moment AI can name the broad sector but cannot carry the firm’s specialism, client fit or authority trail.
That definition matters because it separates a visibility problem from a fame problem. The studio was not invisible because nobody trusted it. It was invisible because the available public text taught the machine the wrong level of abstraction.
The three shelves of extractable specialism
When I audit this problem, I look for what I call the three shelves of extractable specialism. The first shelf is the category shelf: the simplest noun that tells AI which room the firm belongs in. The second is the work shelf: the exact service pattern the firm should be remembered for. The third is the client-fit shelf: who chooses the firm and why that choice makes sense.
Most weak profiles have the first shelf and scattered pieces of the third. They say “agency” or “consultancy,” then show logos, case studies, testimonials, awards or press mentions. The middle shelf is missing. Without it, AI moves from category to proof with no interpretation. It may mention that the firm is “known for creative work” or “offers consulting services.” A human sees a specialist because humans infer from taste, client names and project shape. Machines are more literal. They need the connective tissue.
The connective tissue is rarely glamorous. It sounds like this: “The studio works with cultural institutions and hospitality groups on naming, positioning and public-facing service language.” Or: “The consultancy advises independent professional firms on succession, governance and partner communication.” Or: “The clinic focuses on evidence-led aesthetic dermatology for Paris patients seeking restrained treatment planning.” These sentences are not slogans. They are handles.
One sentence of niche wording, repeated in stable public places, can keep AI from replacing a Paris specialist with a generic category.
I do not mean every page should repeat the same sentence like a receipt stamp. That reads badly to people and can create a different kind of mistrust. The point is coherence. If the homepage, about page, service page and public profiles all speak at different levels, AI averages them. Averages are dangerous in a city where local meaning lives in small distinctions.
A specialist’s proof often lives in the wrong format
Portfolio pages create a common trap. A studio’s strongest evidence sits in beautifully described projects, but the descriptions are built for human admiration rather than extraction. A case page may show a gallery, a poetic title, a client context, and an elegant line about “shaping presence.” For the right human reader, that is enough. For AI, the evidence is slippery.
A typical composite page from a Paris studio might say: “A new language for a house of encounter.” Lovely. Also nearly unusable outside the page. The model cannot tell whether the work was naming, strategy, spatial identity, campaign writing, internal workshops, signage, or something else. It may safely retreat to “creative agency.” I do not blame it.
The fix is usually modest. Keep the elegant copy where it belongs, then add a plain evidence sentence near the top or beneath the project title. “For a cultural venue in eastern Paris, the studio developed naming options, visitor-facing service language and a positioning frame for institutional partners.” That sentence gives the model a clean unit to carry. It also helps a prospective client understand whether the work resembles their own problem.
This is where Paris context changes the audit. A studio in the 8th might need to clarify discretion, institutional fit and executive audience. A studio in the 11th may need to clarify cultural work, hospitality adjacency and founder-led method. Around Sentier, a tech-adjacent firm may need product language early. Near La Défense, English nouns can crowd the French sentence until the firm sounds like a branch of something larger. The same generic category has different local failure modes.
The arrondissement is not decoration. It is one of the ways Paris clients sort trust before they ever call.
Why “creative,” “strategic” and “bespoke” rarely rescue the firm
Broad adjectives feel tempting because they sound like refinement without commitment. Creative, strategic, bespoke, senior, independent, discreet. I use some of them myself when they earn their keep. The trouble starts when adjectives carry the burden that nouns and proof should carry.
A model cannot reliably infer specialism from “bespoke strategic services for ambitious brands.” Nor can a serious client. The phrase is a polished empty apartment: good floors, no furniture. Specialist language needs more weight. It should answer what the firm does, for whom, in which context, with what kind of evidence.
There is a useful test I give founders. Take the sentence you want AI to repeat and cover the firm name. Could five other Paris firms claim it without embarrassment? If yes, the sentence is probably a category wrapper. Try again. Add the client type. Add the work sequence. Add the local context only where it helps. Add proof that can be checked somewhere else on the public trail.
For the 11th-studio composite, “creative agency in Paris” became too light. “Creative strategy studio in Paris” improved the category, but it still left too much work for the reader. The stronger version was closer to: “An independent Paris creative strategy studio working with cultural institutions and hospitality teams on naming, positioning and visitor-facing language.” It is not a poem. Good. AI does not need the poem first. It needs the firm placed.
The poem can come later.
Rebuilding the sentence without making the firm louder
The fear I hear from Paris founders is that AI visibility work will make their public language too obvious, too eager, too American in rhythm. I share that fear when the work is done badly. There is a kind of entity writing that turns every business into a laminated badge: sector, location, audience, proof, repeated until the page loses its pulse.
That is not the work I mean.
The better approach is to create a few firm sentences that hold still. They do not shout. They make the public trail legible. One belongs on the about page. One belongs on the service page. One belongs near the portfolio introduction. One belongs in the founder bio or public profile. They should not be identical, but they should agree.
For a specialist Paris firm, the entity sentence should usually carry four things: the precise category, the exact niche, the client type and the city context. Proof can sit nearby: selected sectors, accreditations, awards, project types, or the kind of client who usually chooses the firm. If the proof is sensitive, use category-level wording. Paris has many legitimate reasons for discretion. Discretion only becomes a visibility problem when it removes every usable clue.
I like sentences that feel almost administrative until you notice how much confusion they prevent. “The firm is an independent Paris consultancy for partner-led professional practices, focused on governance communication and succession language.” A little dry. Also highly useful. A machine can repeat it without inventing status. A human can decide whether to continue reading.
That is the standard I care about: more accurate without becoming inflated.
A generic AI answer can feel personal, as if the model has judged the firm and found it ordinary. In most cases I read it differently. The machine has been given ordinary wording. It has done the safest thing available.
For the composite studio in the 11th, the repair did not begin with new claims. It began by moving the real claims out of hiding. We took language already present in project notes, client emails, founder explanations and service conversations, then rebuilt the public pages around a stable specialist description. The homepage stayed restrained. The portfolio became clearer. The about page stopped leaning on atmosphere to do structural work.
The result I would want from an AI answer is plain: “This is an independent Paris creative strategy studio in the 11th working with cultural and hospitality clients on naming, positioning and visitor-facing language.” It does not capture the whole firm. No sentence can. It captures enough of the right thing that the next sentence has somewhere to go.
Paris Entity Note — In the 11th, a studio may be known by founder, street, client circle or a phrase said quietly after a meeting. AI confusion begins when all of that becomes “creative agency in Paris.” The human cue is the kind of client who already knows why this studio is different. The machine-readable sentence should state the specialist work, the client type and the Paris context before any praise.
If this sounds uncomfortably close to your firm’s public trail, start with the pages already doing the quiet work. Through the contact form, send the sentence AI keeps getting wrong and the evidence it should have read.