Technical & Structure · 9 min read · July 15, 2026
Image-heavy portfolio, invisible to AI: how to make your projects readable
Your portfolio is an image gallery: rendering, floor plan, detail photo, hardly any text. For people that is powerful, for AI almost invisible. ChatGPT, Perplexity and Google AI Overviews read language, not aesthetics. If a client asks who plans sustainable timber buildings in the region, you only appear if your projects are described in words and structured to be machine-readable.
Why your most beautiful project doesn't exist for AI
Architecture firms invest a lot in visualization: high-resolution renderings, elaborate photo series, elegant slideshows. That convinces people instantly. But generative AI systems like ChatGPT, Perplexity, Claude or Google AI Overviews work primarily with text. An image without descriptive text is an empty field for them. If your project page consists only of a rendering and the title Residence H., no AI knows that it's a passive-house new build in timber-frame construction with 180 square meters in Freiburg.
The result: a potential client asks their AI which architecture firm in Freiburg has experience with passive houses made of wood, and your perfectly fitting project isn't named. The information stands nowhere in text form. Your competitor with a factually described, perhaps visually weaker project gets recommended, because there the AI can read what it's about. Beauty wins with people, language wins with machines.
This is the central tension for your firm: exactly the visual strength that distinguishes you makes you blind to the new generation of search. GEO, that is Generative Engine Optimization, doesn't reverse this logic but complements it. You keep your images and additionally deliver the AI the words it needs to understand your work and recommend it onward.
What clients really type into the AI
People ask AI systems differently than classic search engines. Instead of Architect Munich they type whole situations: We want to energetically renovate and add a story to an existing building from the sixties, which firm in Upper Bavaria has done that? Or: Who plans barrier-free multi-generation houses with KfW funding? These questions contain the building task, region, building type and constraints. Your website must contain exactly these terms in natural language, otherwise the AI finds no connection.
Collect the real questions from your initial consultations. Clients ask about costs per square meter, about construction time, about the sequence of the service phases, about experience with the heritage authority, about dealing with hillside locations. Each of these recurring questions is a topic that should stand on your website as a clearly phrased text block. The AI draws its answers from such concrete passages and then names the source.
Think also of the language of the uses: daycare, medical center, winery, terraced-house quarter, commercial hall, attic conversion. If your references only run under Project 2023, you give away every one of these anchors. Instead, name what it is. A project isn't called New Build W., but Construction of a two-group daycare center in timber construction in Regensburg. That is the sentence an AI can quote.
Every image needs its words: alt texts and captions
The fastest lever lies right with your images. Every photo and every rendering has an alt attribute in the source code that is almost always empty or reads DSC_4821. This is exactly where a precise description belongs: Street facade of the renovated Gründerzeit house with new clinker brick layer and floor-to-ceiling windows, project in Leipzig. This text is intended for the visually impaired, but is read and understood just as much by AI and search engines.
Additionally add visible captions under your project photos. They help people and machines equally. Don't just write Interior, but Open living and dining area with exposed-concrete ceiling and oak parquet, ceiling height 2.80 meters. Such captions turn a mute gallery into a narrated tour through your project that the AI can take in paragraph by paragraph.
What matters is honesty instead of keyword stuffing. Really describe what can be seen in the image, with the technical terms you use anyway. Material specifications, construction method, location and use are entirely enough. A string of search terms is recognized by the AI as spam and tends to devalue the page. Concrete, true description is always strong.
The project fact sheet: facts AI loves
Give every project a structured fact sheet in text form, visible directly on the page. It contains the hard facts that get asked about: location, building task, building type, gross floor area, construction method, year of construction, service phases, client (as far as permitted), special features like funding, energy standard or certification. This list is gold for an AI, because it can answer every question directly without having to guess.
An example: location Ravensburg, building task new construction of a residential and commercial building, six residential units, 620 square meters gross floor area, solid timber construction, KfW Efficiency House 40, service phases 1 to 9, completion 2024. From this block an AI can serve several different questions: about timber construction, about Efficiency House 40, about mixed use, about the region. A single well-structured project thus covers a whole spectrum of questions.
Keep the format consistent across all projects. If every reference has the same fact-sheet structure, the AI recognizes a pattern and can compare and classify your projects cleanly. This repetition appears unobtrusive to the human viewer, but for the machine it is a strong signal of reliability and completeness.
Structured data: the language machines understand instantly
Beyond the visible text there is Schema.org markup, an invisible data scaffold in the source code of your website. For architects, ProfessionalService or LocalBusiness as well as Project and ImageObject markups are especially relevant. With them you tell the machine explicitly: this here is a firm that works in this city, here are its projects, here the services. Your web developer can install that in a few hours.
Particularly effective is structured markup for your contact and location data. If name, address, catchment area and fields of activity are stored machine-readable, the AI reliably assigns you to a region and a specialty. That is exactly what decides whether you appear for the question about a firm near me or not. Without this data your localization stays vague for the machine.
Add FAQ markup for your most frequent questions. If you store the questions from your initial consultations as an FAQ with schema markup, AI systems especially like to pick up these question-answer pairs and quote them directly. That is one of the most reliable ways to be mentioned in a generated answer at all.
From picture book to technical text: the project report
Supplement every important reference with a real running text of two to four paragraphs. Describe the starting situation, the client's task, your design answer and the result. Explain why you decided on wood instead of concrete, how you dealt with the tight plot, which funding you unlocked. This narrative is exactly the material AI answers are built from.
Such reports do double duty: they convince the human reader who wants to know whether you understand their building task, and they deliver the machine context and technical terms. Write in your own technical language, but understandably. A sentence like The story was added in lightweight construction so as not to overload the existing structure is valuable and quotable for both target groups.
Deliberately compare the difference: a pure image series tells the AI nothing about your way of thinking. A project report makes your methodology, your values and your experience explicit. That is exactly what clients look for when they select a firm for a demanding undertaking. The text is the bridge between your ability and the AI that is supposed to recommend it onward.
Reputation outside your own website
AI systems base their recommendations not only on your page, but on the entire web. Mentions in trade media, architect databases, the chamber list, in competition documentations and construction blogs strengthen your profile enormously. If your project is described in a professional article and named to your firm, the AI gains trust and names you more often. You can actively build these external signals.
Maintain your entries in Baunetz, architect chamber directories and relevant portals with the same clear descriptions as on your website. Make sure that firm name, location and specialties are written identically everywhere. Contradictory information - sometimes Munich, sometimes Munich-Schwabing, sometimes a different firm name - confuses the machine and weakens your assignment. Consistency across all sources is decisive.
Awards and competitions are worth it twice over too. A prize is often documented in several media and thus creates many agreeing mentions of your name in connection with a building task. For AI systems this agreement is a strong quality signal that brings you forward in recommendations.
Concrete first steps for your firm
Start small and targeted instead of overturning everything at once. Choose your five strongest reference projects and fully upgrade them: precise alt texts for all images, visible captions, a structured fact sheet and a short project report. These five well-prepared projects bring more AI visibility than fifty mute galleries. Quality of description beats quantity.
Then test yourself. Ask ChatGPT, Perplexity and Google AI Overviews exactly what a client would ask, for instance about a firm for timber construction in your region. Are you named? Is a competitor named and not you? Then look at how their page is described differently. These tests show you in black and white where your gaps lie.
Anchor the new standard in your process. From now on, every new project gets a fact sheet, image descriptions and a short report from the start, before it goes online. That way your AI visibility grows automatically with every commission, instead of remaining a one-off cleanup. Step by step, the image-heavy portfolio becomes a body of work that machines can read and people can recommend onward.
Common questions
Do I now have to replace my beautiful renderings with text?
No, absolutely not. Your images remain the emotional core of your portfolio and convince the human clients. You merely supplement them with words: alt texts, captions, project fact sheets and short reports. Image and text work together. The image wins the person, the text makes you findable and quotable for AI systems in the first place.
For confidentiality reasons I'm not allowed to name many project details. What can I do?
That's normal, especially with private clients. You don't have to name names or addresses. Instead, describe the design-relevant facts: building type, region roughly, construction method, area, energy standard, building task and your solution. It's exactly this factual information the AI needs, not the identity of the client. An anonymized but professionally precise fact sheet is fully effective and stays confidential.
How quickly do I see results when I rebuild my portfolio?
Classic search engines need a few weeks to re-index changes. AI systems partly access updated content faster, partly only with the next training or index cycle. Realistically reckon with a few weeks to a few months. What matters is continuity: whoever describes every new project cleanly and builds external mentions will over time be noticeably recommended more often by AI.
Read on