Skip navigation

How ChatGPT Recommends Architects:
What We Found

November 12, 2026

Almost half of consumers now ask AI tools to recommend local businesses. In BrightLocal's 2026 consumer survey, 45% were using AI to find a local business recommendation.

When someone asks ChatGPT to recommend an architect, it tends to suggest the firms it can find described clearly and consistently across the web: on their own websites, in online listings, and in reviews from past clients.
That can sound kind of abstract, so we did a little experiment...

We asked three AI assistants, ChatGPT, Google's Gemini and Anthropic's Claude, the same simple question a homeowner might type: "looking for a residential architect in Seattle." We asked each one twice.

The answers differed more than we expected. Across six answers, the assistants named 18 different firms, and there was only one overlap across all six. We also looked at where the information came from. How does AI know what it knows?

In this post, part of our ongoing series on architect website design, we'll walk through what we found, and then look at what your firm can do to show up.

What happened when we asked ChatGPT, Gemini and Claude for a Seattle architect?

When we asked ChatGPT, Gemini and Claude for a residential architect in Seattle, each one gave us a confident list of six or seven firms, with a short description for each.

But there was very little overlap.

Most firms appeared in only one assistant's answers. Only one firm, Studio Zerbey, appeared in all six result sets.

 

Here's a look at every firm that appeared in more than one assistant's answers:

Firm

ChatGPT

Gemini

Claude

Studio Zerbey

Named

Named

Named

Board & Vellum

Named

Named

—

CAST Architecture

Named

Named

—

Shuler Architecture

Named

—

Named

AOME Architects

Named

—

Named

Source: Bizango test, November 2026. Each assistant was asked twice. 

 

Why would one firm come up everywhere? Studio Zerbey is a client of our architecture websites and marketing services, so we know their story well. We worked with the studio on their brand pillars and messaging, then carried that language consistently through their website, google business profile, and elsewhere online.

Using consistent language everywhere seemed to directly impact the AI results. Their homepage describes Studio Zerbey as creating "warm, natural modern custom homes, remodels, and additions."

When the assistants described Studio Zerbey, they used nearly the same words.

That's the main idea here: AI assistants don't have opinions about your architecture. They repeat what they can find about you. If your firm describes itself clearly, in the same words, everywhere it appears, and those claims are echoed in reviews and public listings, that description is what a prospective client will hear.

How do AI assistants like ChatGPT decide which architecture firms to recommend?

AI assistants like ChatGPT decide which architecture firms to recommend in two ways. Some of it comes from what the assistant already knows, absorbed from the web during their original training and refreshed every so often. The rest comes from what it looks up in the moment — through a hidden web search. Most answers draw on some mix of the two.

It's a bit like asking a well-connected colleague for a referral. Some names come to mind right away, because they've heard them for years. For others, they need to look something up before answering.

How often do assistants look things up? Less than you might think. When Ahrefs studied 66.6 million ChatGPT answers from free, signed-out users, only about one in seven involved a live web search. The rest came from what the model already knew.

Both elements showed up in our test. When we asked for a Seattle architect, ChatGPT ran a search and listed the websites it read. Claude ran a map search and pulled up business listings on a map — but not every recommendation came from those sites. DeForest Architects (also a Bizango client) came second in both of ChatGPT's answers, yet its website wasn't among the pages ChatGPT read in the live search. It recommended the firm anyway, based on what it already knew about the firm.

Practical takeaways

  • Build your reputation over time. What an assistant already "knows" about your firm comes from years of your website, press, awards, directory listings and reviews. Every project write-up, award entry and published mention adds to it.
  • Make your website easy to read quickly. When an assistant does search, it skims. A site that states plainly what you design, where you work and who you work with gives it something clear to repeat.

The next section looks at where that information comes from.

Where do ChatGPT, Gemini and Claude get their information about architecture firms?

We said earlier that the information AI agents already know about firms is an important element in which firms they recommend. Obvious, right? But where do they get that information from?

ChatGPT, Gemini and Claude get their information about architecture firms from a handful of familiar places: the firm's own website, its online listings, what clients say in reviews, and mentions of the firm elsewhere on the web.

None of these is new. Google and other search engines have been drawing on these same sources to understand your firm for years. What's new is that an AI assistant reads them all and turns them into a few sentences.

Here's what each source looks like in practice:

  • Your website. This is the source you control completely, and it's often the one assistants lean on most. The way you talk about your work is the most important thing that differentiates you.
  • Your listings. Google Business Profile, Yelp, Houzz, and directories like AIA Seattle's Find an Architect. These allow AI search agents to verify that a firm is real (anybody can't just launch a website and claim to be an architect!) and also supply information like your neighborhood, phone number, star rating, etc.
  • Your reviews. What past clients said about working with you, in their own words.
  • Mentions elsewhere. Articles, award lists, partner websites, design publications: anywhere your firm is named — with or without a link.

In our test, each assistant leaned on a different mix. ChatGPT relied almost entirely on firms' own websites, cross referenced with AIA Seattle. Gemini referenced each firm's site and cited Houzz. Claude ran a map search and described nearly every firm through its client reviews.

Larger studies confirm these findings. Yext Research looked at 155.5 million citations from AI assistants answering location-based questions, the "near me" kind people ask about all sorts of local businesses. Most of the citations pointed to a business's own website or listing.

 

How often AI assistants cite a business's own website or listing

AI assistant

Share of citations to the
business's own website or listing

Source

ChatGPT (OpenAI)

85%

Yext Research, Q1 2026

Gemini (Google)

81%

Yext Research, Q1 2026

Perplexity

80%

Yext Research, Q1 2026

Claude (Anthropic)

69%

Yext Research, Q1 2026

 

Claude's lower number has a simple explanation, and it matches what we saw. In the Yext data, Claude drew 19.5% of its citations from reviews and social media, compared with 2.6% for ChatGPT.

Mentions elsewhere matter too, especially over time. In a study of 75,000 brands, Ahrefs found that how often a brand is mentioned across the web was more closely linked to appearing in Google's AI answers than how many links pointed to its site. For larger firms, that's where awards, rankings and trade press demonstrate their value.

So which source matters most? Your website. It's where you set the tone: what you design, how you work and who you work with. Your listings, reviews and mentions then echo that story back. When they all say the same thing, the signal gets stronger, and an assistant can describe you with confidence. Getting a firm's website, listings and reviews to tell one consistent story is a big part of what we do for architecture firms at Bizango.

Practical takeaways

  • Start with your website. Settle how you describe your firm, then make the website the source every other profile copies from.
  • Check your listings against it. Google Business Profile, Houzz, Yelp and directories like AIA's should use the same name, services and language.
  • Ask for reviews that describe the work. When a client mentions the kind of project you did, their words reinforce your own.

See our post The Architecture Firms Thriving Online Aren’t Necessarily Who You Think

Why do AI assistants recommend different firms each time you ask?

AI assistants recommend different firms each time you ask because they write every answer fresh and tuned to your exact situation. They don't keep a stored ranking the way a search engine does. Some of the change comes from things like whether the assistant decides to search the web this time. But some of it is built into how these assistants write.

An AI assistant writes its answer one word at a time. At each step it weighs which words could plausibly come next, and it doesn't always pick the single most likely one. It deliberately chooses among several good options, with the likelier ones chosen more often, and a very slight element of randomness built into the algorithm as well. The companies that make these tools design them this way on purpose. An assistant that always picked the most predictable word would sound stiff and repetitive.

Those small choices add up. If one answer happens to open with "For a modern custom home…" and another opens with "Seattle has many excellent firms…," each one heads in its own direction, and different firms come to mind along the way. It's a bit like giving the same brief to the same designer on two different days. The two sketches will share a lot, but they won't be identical.

That's why AI recommendations and rankings are so hard to track over time. There's actually no fixed list sitting behind the answer, and no "position three" to hold onto. A firm's real standing is closer to a likelihood: out of many people asking a similar question, how often does your name come up? The firms that come up most reliably are the ones the assistant has the clearest, most consistent picture of.

In our test, the three assistants behaved quite differently:

  • Claude gave us the same seven firms both times.
  • Gemini kept five of its firms and swapped in a couple of new ones.
  • ChatGPT kept the same top two, but most of the rest of the list changed.

Larger studies show far more variation than we saw. Steady Demand, a local search agency, repeated the same local questions in Gemini across 50 U.S. cities. Gemini named the same top business only about 5% of the time. For comparison, Google's regular map listings, the ones at the top of a search for "architects near me," kept the same top business about 83% of the time.

The assistants also rarely agree with each other. In the same research, Gemini and ChatGPT named the same top business in only 4.2% of searches. That's what makes Studio Zerbey's result unusual: it came first in both assistants, in every run.

Practical takeaways

  • Don't judge from one answer. A colleague (or marketing agency) might send you a screenshot showing your firm, or not showing it. One answer is a snapshot. To get a real picture, ask several times and see what keeps coming back.
  • Consistency is what holds steady. When a list changes from one answer to the next, the firms that stay are usually the ones described most clearly and consistently across the web.

Is being recommended by AI the same as ranking on Google?

Being recommended by AI is not the same as ranking on Google, but Google is still the foundation it's built on. AI adds a new layer of subtlety and complexity on top.

When you search Google for "residential architect Seattle," you get a page of ranked options: map listings, websites, directories. You do the choosing. When you ask an AI assistant the same thing, it does the choosing for you. It reads, decides, and hands back a short list of firms, with a sentence or two about each, and why it might be a good fit for you.

That short list is also much harder to get onto. SOCi, a local marketing company, tracked about 350,000 business locations. On Google, those businesses appeared in the top three map results about 36% of the time. But when people asked ChatGPT for a local recommendation, it named only 1.2% of them. In other words, a business was about 30 times more likely to show up near the top of Google Maps than to be recommended by ChatGPT.

AI assistants visit your website before they decide whether to recommend you

This was the most striking thing in our test. When ChatGPT searched the web for our question, it went to the websites of 11 architecture firms. It read them. Then it recommended three of those firms and passed over the other eight.

Those eight firms did everything we usually think of as successful SEO. They were found, and their websites were visited. But when the AI read their sites, it didn't recommend them.

We can't see inside ChatGPT's decision. But in every case where it recommended a firm whose site it had read, its description came straight from things the website said plainly. AOME's site states the size of the projects it takes on, and ChatGPT repeated that range. Studio Zerbey's site says what the studio designs and how, and ChatGPT repeated that too.

That changes what your website is for. It isn't only something a prospective client browses after hearing your name. It's increasingly the document an AI assistant reads when it decides whether to mention you at all. Arriving on the site is no longer the finish line.

Google is still the foundation

None of this makes Google less important. If anything, it matters more, because the AI assistants rely on it:

  • Gemini is Google's own assistant and draws on Google's search results.
  • ChatGPT leans heavily on pages that already rank well in Google search. In a study by AirOps, pages ranking first on Google were cited by ChatGPT 43.2% of the time, about three and a half times as often as pages ranking below Google's top 20.
  • Claude's map search drew on the same kind of business listings you see in Google Maps.

Practical takeaways

Getting recommended by AI is a two-step process:

  1. Rank well in Google and its map listings, so AI assistants find you when they look.
  2. Have a website that says clearly what you do, so they choose you once they arrive.

The firms that came up in our test had done both.

What can an architecture firm do to get recommended by AI?

An architecture firm can do quite a lot to get recommended by AI, and most of it is work you should be thinking about for human visitors anyway. It comes down to being clear, specific and consistent everywhere your firm appears. Here's where we'd start.

1. Say what you do in one clear sentence, and use it everywhere.

Studio Zerbey's homepage says, in a single sentence, what the studio designs and the feeling it's after. The AI assistants repeated that sentence back almost word for word. Write your version, then use the same language on your website, your Google Business Profile, your directory listings and your social profiles.

2. Say what kinds of projects you take on, and at what scale.

All three assistants ended their answers the same way, by asking what kind of project the person had in mind: a new home, a remodel, an addition or an ADU. ChatGPT also asked about budget. Firms that state their project types and typical scale plainly are easy to match once a client answers. AOME's website gives its typical project range, and ChatGPT passed it along.

3. Describe your projects in words, not just photos.

AI assistants read text. A beautiful photo with a one-line caption gives them almost nothing to work with. A few paragraphs about the site, the client's goals and how you solved the problem give them a lot. (We cover this in [post 2].)

4. Write about the things your clients worry about.

All three assistants brought up the same concerns: Permitting, steep slopes and environmentally critical areas. If your firm handles those well, say so, and explain how. Assistants are clearly looking for firms that can speak to it.

5. Keep your listings complete and consistent.

Your firm's name, address, phone number and services should match everywhere they appear, starting with your Google Business Profile. Some assistants pull these details straight from listings, as ChatGPT and Claude did in our test. (More in [post 3].)

6. Ask happy clients for reviews.

Claude described nearly every firm through its reviews. In SOCi's research, the local businesses ChatGPT recommended averaged 4.3 stars. Reviews in clients' own words are some of the most persuasive material an assistant can quote.

7. Build your offsite profile.

Award lists, design publications, local press, and partner websites like your builders' and consultants' sites all add to what assistants "know" about you before anyone asks. This is slow, compounding work, and it pays off over years.

If you'd like to go deeper on any of these, our earlier post, What ChatGPT Doesn't Know About Your Firm, walks through the content, technical and off-site sides in more detail.

The short version

AI assistants like ChatGPT, Gemini and Claude recommend the architecture firms they understand best. They build that understanding from your website, your listings, your reviews and 3rd party mentions around the web, and they put together a fresh and customized recommendation every time someone asks.
That's why the answers keep changing, and why one firm in our test came up every time anyway: its story was clear and told the same way everywhere.

The firms that get recommended make it easy for an AI assistant to say who they are, what they design and who they design it for.

If you'd like to talk about your firm's website...

We're Here!

Mark Figlozzi

Creative Director Mark Figlozzi founded Bizango in 2006. He studied Creative Writing at Oberlin, and believes in the power of storytelling and emotion when it comes to marketing architects. He believes a firm's website should be beautiful and immersive to create desire. We humans make our most important decisions at an emotional level and good taste sells.

FAQs

What if...

Back to blog