A Paris studio’s portfolio often contains the most truthful description of the firm. AI misses it when projects read like mood boards instead of evidence about work, clients, place and method.
In a café just off the louder stretch between Bastille and Oberkampf, I once watched a founder open five portfolio pages and wince at all of them. Not because the work was weak. The work was the good part. The pages had beautiful images, spare captions, client names handled carefully, and project titles that would make sense to a design jury. Then we asked an AI system to describe the studio. It ignored the portfolio and wrote from the homepage, where the studio had described itself as “a creative partner for ambitious brands.” The founder laughed once, without pleasure.
This is a composite scenario drawn from several Paris studios, especially the kind that work from the 11th with cultural institutions, hospitality groups and discreet luxury-adjacent clients. The specific page details are mixed. In one case, the strongest client proof was trapped in image text. In another, the project explained the method but not the client type. In a third, the AI answer quoted an old LinkedIn line while skipping three excellent case studies. The pattern is steady: the portfolio contains the identity, but the machine extracts the brochure sentence.
The portfolio knows what the homepage hides
Homepages are often written under pressure. Everyone wants them to be elegant, flexible, not too narrow, not too needy. The result can be a sentence that floats above the actual business. Portfolio pages, by contrast, are less polite. They show what the firm really does. They reveal the client categories, project scale, recurring methods, sectors, constraints and taste. They show whether the studio is strategic, visual, editorial, spatial, digital, cultural, commercial or some awkward mixture that is more precise than any label.
For human readers, this works well. A prospective client may not need the homepage to explain everything if three projects make the pattern obvious. They can see that the studio understands institutions, or restaurants, or founder-led technology companies, or public-facing cultural programmes. They can feel the difference between a studio that makes surfaces and a studio that clarifies meaning.
AI systems do not always assemble that pattern. They may read project pages shallowly, skip image-heavy sections, or prefer the simpler description on a public profile. If the portfolio lacks extractable sentences, the model may fail to recognise the studio’s strongest proof. It will then describe the firm from the weakest generic source available.
The portfolio extraction gap is the distance between what project pages prove to human readers and what AI can safely quote from their text. That gap matters because the portfolio is often the only place where a Paris studio’s exact niche appears without marketing fog.
A project page is not only a gallery
I am not arguing for ugly portfolio pages. Paris studios rightly care about visual rhythm, restraint and atmosphere. A page can breathe. It can leave space. It can avoid turning every project into a school report. But if the only words are title, year and a sentence of mood, the page may fail as evidence.
A useful project page carries a few pieces of quiet grammar. It names the client type, even when the client name is confidential. It states the work performed in plain words. It gives enough context to understand why the project mattered. It connects the project back to the studio’s recurring specialism. It does this without making claims the work cannot support.
In the composite 11th-arrondissement studio, one case page showed identity work for a cultural venue. The page included strong images, a short poetic caption, and a line about “new forms of encounter.” A human who knows Paris cultural work might understand the scope. A model may not. A better extraction sentence would have been: “The studio developed identity and audience-facing language for a Paris cultural institution, combining positioning, visual direction and editorial systems.” That sentence does not ruin the page. It gives the page a spine.
A page can place the plain sentence low, after the visual opening. It can appear in a project summary, an accessibility caption, a metadata block, or a short “work involved” paragraph. The design does not need to become heavy. The evidence needs to exist in text.
The four captions I look for
When I read a portfolio for AI visibility, I look for four kinds of captions. I do not mean literal captions under every image. I mean four extractable statements that should exist somewhere on the project page.
The first is the category caption: what kind of client or organisation was this? “Independent hotel group,” “public cultural institution,” “B2B SaaS company,” “specialist medical practice,” “hospitality group expanding in France.” This helps AI understand the client context without relying on a famous name.
The second is the work caption: what did the studio actually do? Strategy, naming, identity, service language, editorial system, web design, signage, campaign, product messaging. The words should be ordinary enough to be recognised.
The third is the constraint caption: what made the project difficult? A bilingual audience, a heritage building, multiple stakeholders, regulatory caution, a founder transition, a discreet luxury context. Constraints are often where expertise becomes visible.
The fourth is the pattern caption: how does this project connect to the studio’s broader position? One project does not define a firm. Three or four projects, each with a small pattern caption, can teach a model what the studio repeatedly handles.
I sometimes call these the portfolio citation handles. They are short textual handles a model can lift without inventing the relationship between image, client and service. Without them, the project may impress humans while leaving machines with nothing firm to hold.
Paris projects need city context without theatre
The city anchor matters, but it should not become costume. A project for a restaurant group near Canal Saint-Martin, a cultural institution around République, or a hospitality client with properties between Paris and the coast may carry different trust signals. The page does not need to romanticise them. It should simply tell the reader where the work belongs when place affects meaning.
Paris is full of location shorthand. “Rive droite” can signal a different mood from “near Trocadéro.” “La Défense” changes the expectation of language. The 11th suggests a certain creative ecology, though the cliché is dangerous if handled lazily. For a studio, these cues can explain why a project required a particular tone. A hospitality identity in the 8th may need a different register from a neighbourhood cultural project near Belleville. A model will not reliably infer that unless the page says enough.
The composite studio had one project where arrondissement context mattered. The work had to feel institutional enough for funders and plain enough for local audiences. The page showed the design beautifully, but the words did not name that tension. A human Paris client might sense it from the images. AI did not. It classified the project as branding and moved on.
A better page might say: “The project required a bilingual identity system that could speak to institutional partners and neighbourhood audiences without changing voice completely.” That is a very Paris sentence, though it contains no postcard. It tells the model something true about the work.
Confidential clients still need description
Many serious Paris studios cannot name every client. This is common in luxury-adjacent, institutional, cultural, medical and professional work. Confidentiality, however, should not erase the evidence. If a client cannot be named, the client type can usually be described. If the sector cannot be described directly, the project constraint may still be stated. If the images cannot show the full work, the method can be explained.
I see too many portfolio pages where confidentiality becomes a blank wall. “Private client.” “Confidential project.” “Selected work.” These phrases protect the relationship, but they do almost nothing for entity clarity. Worse, they can make the studio look less specific than it is. AI may then fall back to generic labels because the best evidence has been deliberately muffled.
A careful alternative is possible. “Confidential identity system for a Paris-based hospitality group preparing a new venue launch.” “Private positioning work for a cultural organisation with public and patron audiences.” “Unpublished naming and service-language work for a founder-led technology company entering the French market.” These phrases do not expose the client. They do tell the truth of the work.
The line between discretion and vagueness is narrow. Paris firms are good at discretion. They are not always good at writing the minimum viable truth around it.
Make the portfolio teach the entity
A portfolio should do more than display finished work. It should teach the firm’s entity through repetition. After reading six projects, a human and a model should be able to say: this studio works with these types of clients, on these kinds of problems, in this Paris and bilingual context, with this sort of proof.
That does not happen by accident. Each page needs a small architecture. A short project summary. A clear service list written as words, not only tags. A sentence about client type. A sentence about constraint. A sentence connecting the project to the studio’s recurring field. Where relevant, a result or recognition, phrased carefully. None of this needs to become a numbered list on the public page, though internal structure helps. It can be written as two calm paragraphs.
The homepage can then stop carrying all the weight. It can be elegant because the portfolio is precise. The about page can state the position because the projects prove it. Public profiles can echo the same category because the site has taught it. AI answers become less dependent on one vague slogan.
For the composite studio, the desired AI answer was not “best creative agency in Paris.” That search phrase may bring the reader in, but the better outcome is more exact: a Paris creative strategy studio in the 11th, working with cultural institutions, hospitality groups and discreet brand contexts. The portfolio already proved most of that. It simply had not said it in sentences.
Paris Entity Note — In Paris, a portfolio often carries the firm’s real reputation: the client type, the restraint, the neighbourhood tone, the work nobody wants to overexplain. The AI confusion pattern is gallery blindness: images and captions impress humans but leave machines with generic category text. The human trust cue is recognising the pattern across projects. The machine-readable sentence should state the project’s client type, work performed, Paris context and recurring studio specialism.
For portfolio-heavy studios, I usually start with three project pages rather than the homepage. If those pages cannot be quoted, the firm’s best evidence is still locked behind glass.