Architectural Prompting: before adopting another AI tool, 5 questions to ask yourself in the studio

The risk of starting from the instrument

The most common temptation is to choose an application first and then look for a way to use it. This approach easily leads to the accumulation of accounts, subscriptions and personal procedures that remain isolated. One person uses AI to prepare minutes, another to work on images, yet another to search for information in documents. Single tests can work, but they hardly become a shared method.

AI produces a more stable benefit when it starts from a specific activity, not from the functions offered by a platform. Before adding a new tool, it may therefore be useful to answer five questions.

1. Which activity really repeats itself?
   It is best to start with what is often done: preparing the minutes, researching the specifications, reorganizing observations or producing recurring documents. A repetitive activity is easier to observe, describe and improve. To identify it, just write down, for a few days, the operations carried out several times or those that require you to continuously retrieve information already searched for.
2. Can the expected outcome be described precisely?
   “Use AI for the project” does not define a task. “Extracting all the requirements relating to insulation from a specification, indicating the page of origin” is instead a precise task. Before even writing a prompt, it may be useful to complete this sentence: “At the end of the operation I want to obtain…“. The answer can be a list, a table, a summary, a comparison of versions or a first draft. The important thing is that the result is recognizable and verifiable.
3. What documents are needed?
   AI doesn’t fix a messy archive on its own. If the documents are not up to date, the versions are not recognizable or the necessary sources are missing, the result will also be unreliable. To prepare a report, a transcript, agenda and list of participants may be needed; for a document check, specifications, updated documents and regulatory references. This question often brings up a problem that predates AI: the firm has a lot of information, but doesn’t always keep it in an easily searchable form.
4. Where is the necessary knowledge found today?
   In many studies the most important information is not in documents, but in people’s memories. A senior professional knows where to look for a prescription, remembers how a similar problem was solved, or knows the reasons for a decision made years earlier. The problem is therefore not only to find information more quickly, but to prevent the knowledge produced from remaining tied to a single person. Neat archives, procedures, checklists, report templates and previous cases can become a shared and reusable basis. AI can help consult it, but first you need to decide what is worth keeping.
5. How will the result be checked?
   Control should be defined before using AI. You need to determine who verifies the output, which sources need to be checked, and which errors would be particularly serious. An internal summary requires different precautions than a regulatory verification or a text intended for a contractual document. The AI ​​can carry out an initial reading, sort the materials and flag elements for further investigation. However, the responsibility for the evaluation remains with the professional, especially when standards, costs, safety or design decisions are involved.

A test to do in a week

To understand where to start, you don’t need to immediately prepare a complex strategy. For a week you can write down:

• repeated activities;
• information searched more than once;
• documents that are difficult to find;
• the steps that always depend on the same person;
• operations that require manual transcriptions or comparisons.

In the end, just one activity is chosen and described through five elements:

1. Activity: read the coordination minutes.
2. Source materials: minutes and transcripts.
3. Expected result: list of decisions, managers and deadlines.
4. Accepted sources: only the documents provided.
5. Control: Project manager review.

At this point it will be easier to choose the right tool, formulate the request and evaluate the result based on a concrete need.

From isolated testing to a working method

Artificial intelligence can be useful even far from images and three-dimensional models. It can intervene in documentation, archives, information searches and repetitive operations that occupy a significant part of the day. The decisive step, however, is not to continually add new tools. It consists of recognizing an activity, describing it, preparing the sources and defining the necessary control.

The book addresses a question close to the one proposed in this article: which operations does it make sense to entrust to AI, at which stage of the project and with what control? Images, spatial alternatives, technical documents, regulatory analysis, agents and archives are brought back to a common method.

Alongside practical exercises, the volume collects cases of design studios that have already introduced AI into their work and includes a conversational digital extension to delve deeper into content and update tools and applications. In a rapidly changing industry, learning every new software is impossible. Building a criterion for choosing what to use is a more lasting investment.

Sources and insights

• Kiana Buchberger, How Architects Can Simplify Project Workflows Beyond CAD and BIM, ArchDailyJuly 10, 2026.
• Agustina Iniguez, Rethinking the Architecture Firm for the AI ​​Era, ArchDailyMay 11, 2026.
• Luciana Mastrolia, Federica Joe Gardella, Generative artificial intelligence for designers. Methods and tools for augmented design practiceMaggioli Editore, July 2026.