Technology and AI

ChatGPT at an Engineering Firm: Uses Worth Trying, and the Rules Around Them

Sam&Sain Co. · Published · Updated · 7 minute read

ChatGPT and similar assistants are language tools: they draft, summarize, reword and explain. At an engineering firm they save time on letters, minutes, summaries and first drafts, and create risk when they stand in for engineering judgment.

On this page

Terms Used

  • General assistant: a chat tool such as ChatGPT, Claude, Gemini or Copilot that answers in text from a prompt.
  • Prompt: what you type or paste in: the request, plus any material it should work from.
  • Data terms: what the provider may do with what you enter, including whether it trains its models on it.
  • Hallucination: in the words of the practice advisory below, an output that “appears to be plausible but turns out to be factually incorrect, inaccurate to the inputs, or otherwise nonsensical”.

What General Assistants Are Good At

ChatGPT, Claude, Gemini and Copilot are language tools. They are good at producing, shortening and reshaping text, and at explaining things in plain words. They predict likely words from what they were trained on and what you give them; they do not look anything up unless the tool is set to, and they do not know your projects unless you supply them. At an engineering firm, that still covers a lot of the day that is not engineering: letters, minutes, summaries, first drafts and the plain-language version of a technical point.

Uses Worth Trying

UseWhat to give itWhat to check
First drafts of letters, transmittals and responses to commentsYour notes and a past letter in the firm’s styleEvery fact, date and commitment
Summaries of a specification, a standard, minutes or an RFPThe document, and the questions you need answeredEach point against the section it cites
Minutes and action listsRough notes and the firm’s minutes formatNames, owners and dates
Plain-language explanations for a client, a board or a permit applicantThe technical text and who will read itThat the meaning hasn’t changed
Checklists of every requirement in an RFP or a contractThe documentNothing missed, section numbers right
Editing a report for length, consistent terms and repeated textThe draftThat no technical statement was altered

In every row, the person who sends the result checks it as if they had written it.

What Stays with the Engineer

General assistants can produce numbers, code references and technical statements that read well and are wrong. So the line is simple:

  • Calculations belong in the firm’s checked tools.
  • Code and standard requirements are read from the code or standard itself.
  • Design decisions, and anything that goes under a seal, are the engineer’s, made and checked the way they always have been.

An assistant can help write the report. It cannot be the reason a number in it is right.

What Ontario’s Regulator Has Adopted

Professional Engineers Ontario has adopted Engineers and Geoscientists BC’s practice advisory, Use of Artificial Intelligence (AI) in Professional Practice, “for use as a guideline in Ontario”, with Ontario’s own statutes and PEO’s Code of Ethics in place of BC’s (PEO, adopted advisory, version of 22 November 2024). Its main points for a firm:

  • Responsibility. Engineers “remain responsible for their work that incorporates or is based on AI output”, and “can face professional consequences for work product containing AI hallucinations” (advisory).
  • Safety. They “should not use or rely on AI outputs for projects involving safety or environmental risks unless they understand the underlying processes and reasoning behind the AI system’s output.”
  • Confidentiality. Data uploaded to an AI tool “may be used to further train the AI model”, which could give other users access to the original user’s confidential information.
  • Records. Help with grammar or paragraph structure, reviewed by the engineer, may not need documenting; AI used “to generate results for a report or to make design decisions” is documented as if the engineer had done the work by hand. Checks may include “noting the make and version of the AI-based system or tool used.”
  • Firms. Firms should know the contract or licence agreement behind each AI product, consider proficiency training, and consider internal policies and procedures.
  • Telling the client. The advisory notes that BC’s regulator has no explicit disclosure requirement, but risks from AI use are assessed and documented, sharing them with the client is considered, and contracts may require more.

Rules to Set Before Anyone Starts: A One-Page Policy

A small firm’s AI policy can fit on one page. The last two points follow what the advisory above says firms should know and consider:

  1. Approved tools and accounts. Business accounts whose terms say your entries are not used to train the model, rather than personal free accounts, and one person who holds the licences.
  2. What may never be entered. For example: client documents marked confidential, personal information, security-sensitive drawings, anything a contract restricts.
  3. Who checks. Every output that leaves the firm is reviewed by the person responsible for it, as if they had written it.
  4. Recording use. Where AI produced results or informed a design decision, note it on the project with the tool and version; where a client’s contract asks for more, follow the contract.
  5. Reading the terms. Who reads each AI product’s licence and data terms, and when they are read again after a change.
  6. Training. What each person learns before using a tool on client work.

Deltek’s 2026 study found 93% of architecture and engineering firms had experienced a cyber attack attempt in the past three years, and one in five a successful attack that caused financial or reputational damage (Deltek, 12 May 2026). Rules about what goes into an AI tool belong with the firm’s other data rules.

Personal information has its own guidance. The Office of the Privacy Commissioner of Canada’s principles for generative AI ask organizations using these tools to “use anonymized or de-identified information within prompts” where possible, and to know “that accountability for decisions rests with the organization, and not with any kind of automated system” (Office of the Privacy Commissioner of Canada, 7 December 2023).

Writing Prompts That Work

  • Say who the text is for and what it must do: “a two-paragraph letter to a property manager explaining why the inspection is needed”.
  • Give it the material: your notes, the relevant section, a past example in the firm’s style.
  • Ask for the format: headings, length, bullet points, tone.
  • Ask it to list what it assumed, so you can check those first.
  • Ask it to point to where each statement came from in what you gave it, so a summary can be checked line by line.

A prompt built that way reads like a short brief to a junior colleague: “Using the attached site notes and last year’s letter as the style, draft a two-paragraph letter to the client’s property manager explaining why a roof inspection is needed before the spring. Plain words, no technical terms. List anything you assumed.” The draft that comes back still needs the sender’s check, but it starts from the firm’s facts.

Checking What Comes Back

Checking an assistant’s answer is quicker when the check is a habit rather than a judgment call each time:

  1. Facts. Every name, date, address and figure traced to your own material.
  2. References. Every section, clause or code reference opened and read; an assistant can cite a clause that says something else.
  3. Numbers. Anything calculated is recomputed in the firm’s own tools.
  4. Assumptions. The list it gave you, read first.
  5. Commitments. Nothing promised in a letter that the scope and the fee don’t cover.

Beyond the Chat Window

A chat window helps one person at a time, with whatever they paste in. The larger gains come when an assistant works from the firm’s own documents, templates and standards, so every engineer starts from the same material, and when its drafts land in the systems the firm already runs. We cover the wider picture in AI at an engineering firm, and proposals in AI for proposal writing.

How We Approach This

We build AI tools that work from your firm’s own documents, templates and standards as part of our software and AI work. The engineering, the checking and the seal stay with your engineers.

Questions

Can Engineers Use ChatGPT for Work?

Yes, for the language work around the engineering: drafts, summaries, minutes and plain-language explanations. Under the practice advisory PEO has adopted, the engineer remains responsible for any work that incorporates AI output, so every result is checked.

Can ChatGPT Do Engineering Calculations?

It can produce numbers that look right and are wrong. Calculations belong in the firm’s checked tools and with the engineer who signs for them.

Is It Safe to Paste Client Documents into ChatGPT?

Only on a business account whose terms keep entries out of model training, and only where the client’s contract allows it. For personal information, Canada’s Privacy Commissioner asks organizations to use anonymized or de-identified information in prompts where possible.

Do Engineers Have to Tell Clients They Used AI?

The advisory PEO adopted notes no explicit disclosure requirement from BC’s regulator, but risks from AI use are assessed and documented and sharing them with the client is considered. A contract or RFP may require disclosure; read it.

Should AI Use Be Recorded on a Project?

Where AI produced results for a report or informed a design decision, yes, documented as if the work had been done by hand, with the tool and its version. Help with grammar or structure, reviewed by the engineer, may not need recording.

Does Using AI Change a Firm’s Professional Liability?

Responsibility stays where it was: under the advisory PEO adopted, the engineer remains responsible for work that incorporates AI output. Ask the firm’s insurer whether its policy says anything about work done with AI.

Sources

Updates

  • : Rewritten at full length: a table of uses with what to give and what to check, what the AI practice advisory PEO adopted asks, a one-page policy, the Privacy Commissioner’s principles on prompts, a sample prompt and checking an answer; Deltek re-read.

Discuss Your Project

Topic

We read every application. If we can help, the first 30-minute video call is free. Or contact us.