A portfolio can win every award in the room and still read to an AI system as an empty page
Many architecture practice sites are built as one long, heavily animated, client-rendered page: full-bleed imagery, minimal text, project names with little further detail. That is a defensible design choice for a person scrolling deliberately. Most AI crawlers read differently, and cannot execute the JavaScript that page depends on to show anything at all.
Short answer
An AI-readable portfolio for an architect states each project's typology, scale, and location in real text and structured data that loads without JavaScript, so an AI system can verify what the practice actually builds instead of finding an empty shell where the gallery should be.
Why a client-rendered site is a real risk here, not a stylistic detail
A portfolio built almost entirely as a single-page, script-driven experience, animated transitions between full-bleed project images, everything assembled in the browser, is a common and often striking choice for an architecture practice. The risk is what happens when the script does not run at all, which is closer to the normal case for an AI crawler than an edge case.
Vercel's study of AI crawler traffic found that these bots generally do not execute JavaScript and cannot read client-side rendered content, ClaudeBot and ChatGPT's crawler together account for a large share of that traffic. A project description that only appears after a script finishes running may simply never reach whichever system is deciding whether to name the practice for a given kind of building.
This rarely shows up as an obvious problem, because a human visitor experiences the finished animation, not the empty document underneath it. A practice can have a genuinely striking site by every visual measure and still hand an AI crawler nothing but a blank shell and a handful of script tags, which reads to that system as a project with no name, no typology, and no location at all.
What actually needs to be readable
Each project needs its own entry with real text present in the page's own markup, not only in an image: typology (residential, cultural, hospitality, institutional, adaptive reuse), rough scale, city, and a plain description of the design approach the practice actually took. "Award winning design" says nothing an AI system can act on; "a low-rise cultural centre built around an existing courtyard" does.
Structured data should mirror that same text as a CreativeWork or Project entry connected to the practice's own Organization schema, naming the architect or firm, the project, and where it sits. The one rule that matters most is that the markup never claims more than the visible page actually says: schema is only useful to an AI system when it can be checked against real content on the page.
A short written narrative for each project, two or three sentences on the brief, the site, and the response, does more here than another rendered angle of the same building. It is also the one piece of the portfolio a script cannot substitute for: no amount of visual polish gives an AI system a typology, a client type, or a design decision to cite unless a sentence somewhere states it plainly.
A gallery that only appears after a script runs is, to most AI crawlers, a page that never loaded.
23.84%
ClaudeBot share of AI crawler requests measured across Vercel-hosted sites, against 11.50% for ChatGPT. None of these crawlers execute JavaScript.
Vercel, AI crawler studyWhere to start
Check whether a project's name, typology, and description exist as text in the page's HTML before any script runs, not only as an image with no accompanying caption. Server rendering or static generation at build time solves this without giving up the animated presentation for human visitors.
Next, audit the practice's structured data against what the page actually states. Where the two disagree, an AI system is right to distrust both, so the fix is always to bring the markup back in line with the visible text rather than the other way round.
A practical way to test this without special tools: disable JavaScript in a browser and reload the site's own project pages. Whatever text and links are still visible with the script turned off is a reasonable approximation of what an AI crawler actually sees. If the project titles and descriptions vanish along with the animation, that is the exact gap worth fixing first.
Revisit this whenever a new project ships or an older one is removed from the site, since a portfolio that stops updating its descriptions quietly drifts out of date even while the gallery itself keeps looking current. The same discipline applies here as for any structured data: it is only trustworthy while it is current.
AI-readable websites
The hub this use case sits under.
AI-readable websites for law firms
The same verification problem applied to a different profession.
AI-readable websites for founder-led businesses
The same entity-clarity argument in general terms.
How to check if a website is AI readable
A broader diagnostic that applies before any portfolio-specific fix.
Find out whether your practice's portfolio gives AI systems any content to actually read.