How architecture firms get recommended by ChatGPT and AI search

Last week I typed a question into ChatGPT that a prospective client might type: best social media agency for architects. It did not return a page of links. It returned a shortlist of three companies, described what each one does, and linked its sources.

That is the shift, and it is worth sitting with. A search used to produce ten results and leave the choosing to the human. An assistant produces a shortlist and does most of the choosing for them. If you are not on that list, you were not rejected. You were never considered.

The same thing is already happening with the questions your clients ask. Who should design a timber office building in Berlin. Which practices specialise in listed-building conversions. Find me an architect for a small hotel. Those are shortlist questions, and they are increasingly answered by a machine reading the web on the client's behalf.

Why a typical architecture website is invisible to these systems

Architecture websites are built for humans with good taste. Large images, restrained text, a portfolio that unfolds as you scroll. That is exactly what makes them hard for an assistant to use.

The recurring problems we see:

1. The important information is inside pictures. The project name, the city, the building type and the client are on the image, not in text. A model reading the page sees very little.
2. Nothing states plainly what the practice does. A page that opens with a mood image and the word "Studio" tells a machine nothing it can repeat to a client.
3. There are no answers, only presentation. Assistants quote sources that answer questions. Most practice sites never phrase a question, let alone answer one.
4. The entity is inconsistent. The name is written three ways across the site, the directories and LinkedIn, and the city appears nowhere in text.
5. Almost nobody else mentions the firm. An assistant weights what other sites say about you far higher than what you say about yourself.

None of that is a design failure. A portfolio site is supposed to look like that. It simply is not doing the second job that now exists.

What assistants actually read

Different systems work differently, and none of them publish their rules. What is consistent across them is roughly this: they read pages they can parse as text, they lean on structured data when it exists, they rely heavily on third-party sources, and they favour content that answers a question directly rather than approaching it slowly.

So the work splits into two halves. Make yourself readable, and make yourself corroborated.

What to change, in order

1. Put the plain facts in text. Who you are, what you design, where you work, for whom. One paragraph in real text on a page that is not blocked to crawlers. This sounds trivially obvious and most practice websites do not have it.
2. Make the entity consistent. The same firm name, city, discipline and description everywhere: your site, LinkedIn, directories, chamber listings, event pages. Link the profiles from your site so they resolve to one identity rather than five loosely related ones.
3. Add structured data. Organization and Article markup at minimum, so the machine does not have to guess what the page is. It is a small amount of code and it removes ambiguity.
4. Write answers, not brochures. Take the questions clients actually ask you on a first call and answer them on the site, with the question as a heading and the answer immediately under it. This is the single highest-return piece of work, and it is the same discipline that makes a page quotable.
5. Earn third-party mentions. Press, publications, partner sites, event listings, university and association pages. This is the part practices skip and it is the part that carries the most weight, because it is the only evidence that is not self-reported.
6. Check that you are readable at all. Text that only exists after JavaScript runs, or only inside a PDF, may as well not exist. Neither may a site that blocks the crawlers these systems use.

Most of this is not new SEO wearing a new name. Classic search rewards the same things. The difference is that classic search forgives a beautiful site with thin text, because a human will click into it anyway. An assistant will not.

Where we stand, stated honestly

When I ran that first query, ChatGPT named us first, described our focus on architecture and engineering firms accurately, and linked to our site as a source.

I am not going to dress that up as proof of a method. It was one question, from one machine, from one network, on one day. What I can say is what we did that plausibly earned it, and all of it is in this article: the Journal answers real questions in plain text rather than presenting a portfolio, every article carries structured data and a question-format section, our entity is consistent across the site and our profiles, and the events we run produce mentions of us on other people's sites.

We are now measuring it properly across regions rather than trusting the first pleasant result, because the difference between a signal and a measurement is exactly what we would want a client to hold us to.

Common questions

Is AI search optimisation different from SEO? Mostly it is the same foundation with a different emphasis. Crawlability, clear text, structured data and third-party mentions serve both. The emphasis shifts toward answering questions directly and toward being corroborated elsewhere, because an assistant cites rather than lists.

Do architecture firms actually get clients this way yet? It is early, and anyone who tells you otherwise is guessing. What is already true is that prospective clients use assistants to build shortlists and to check names they have been given, which is the same moment described in referrals for architects.

Will AI kill Google search for architects? Not soon, and it does not matter much for the work. The things that make you visible to an assistant also make you visible in classic search.

How long does this take to show up? The readability work lands within weeks. The corroboration half is slower, because press, listings and partner mentions arrive on other people's schedules.

Can we do this ourselves? The first three steps yes, with a developer afternoon and someone willing to write plainly. The mention side is the hard part, and it is the one worth outside help.

The shortlist is the new first page

Being invisible to an assistant does not feel like anything. There is no drop in a dashboard, no lost ranking to investigate. A client simply asks a question, receives three names, and none of them are yours.

Fixing it is unglamorous work: say clearly what you do, be the same firm everywhere, answer real questions in real text, and get other people to mention you. If you want that done properly for your practice, tell us about your firm.