Most architectural writing is not design writing. It is the document around the design — the sheet you fill in, the response to a question you already know the answer to, the note confirming what was said at 8am on site. Those are templates with firm-specific vocabulary, and that is exactly the shape of work a language model is good at producing quickly.
Where AI clearly helps
- RFP and qualification responses. The same firm profile, the same four reference projects, the same mandatory formatting. Writing it once and re-prompting it per opportunity saves real hours around a deadline.
- Project descriptions for competition entries. A 300-word statement of concept, material choice and spatial sequence. The model is reliable here as long as you feed it the actual drawings and numbers.
- Meeting notes and action lists. Long, dull, and read by everyone afterward. A short summary with owners and dates keeps a project from drifting.
- Material and finish schedules. Provide the list and it will format it consistently — the part that used to be copy-paste between three documents.
- Post-occupancy follow-up letters. The check-in six months after handover, asking what is not working. Polite, brief, no design argument.
- Client welcome documents. What to expect in weeks one through six, who to call, where the site entrance is. Every practice writes the same one.
Where it nearly got us in trouble
- Invented dimensions and areas. A statement that says the living room is 34 square metres when it is 22 is the kind of error that ends up in a client conversation. Never let a model produce a number you have not measured.
- Compliance language softened. "Meets current standards" is meaningless if the sentence around it is wrong. Accessibility and fire requirements are not prose to be paraphrased by a model that has never read the local code.
- Material claims. Fire ratings, slip resistance, emission classes, warranty periods for a specific product. These come from the manufacturer, not from a language model.
- Competition text that describes something you did not draw. A generated narrative about "a porous south face" when the facade is glass. Jurors read carefully, and clients read it back to you later.
- Over-confident scheduling. Phase dates, council timeframes, consultation windows. A model has no idea what your local authority actually took last spring.
Editing time, from our 30-day benchmark
Our 30-day test measured roughly 22 minutes of editing per 1,000 words on the low-cost dedicated writer and about 8 minutes on the general assistant, which works out near $7 per piece of your time at $30 per hour. For an architect billing design rates, the honest argument is not the hourly math. It is that the qualifications response gets finished on the same evening instead of the following weekend.
A workflow that holds up
Keep one folder per project with the drawings, the measured schedule and the client's brief in it. Generate the competition statement from those files, then walk the text against the drawings line by line — dimensions first, material second, and the word "adjacent" third, because that is the word models use when they have not looked. For RFP answers, use the general assistant to assemble the section order and the compliance table, then fill every number by hand. Read any document the way a client will: skimming for the one figure they care about. If that figure is wrong, the rest of the text does not save you.
Start with the qualifications response
The low-cost plan is enough for the repetitive half of a practice's writing.
Try Rytr →What architects ask
Can AI write a competition entry statement?
As a first draft from your own drawings and numbers, yes. It cannot know what the building looks like, so the narrative has to be anchored to something you actually drew.
Is it safe for RFP answers?
Yes for structure and tone. Every project reference, fee and date in the finished document should be one you supplied — not one the model reproduced from a previous submission.
Does a paid plan make sense for a small practice?
The cheap plan covers correspondence and short statements. The premium general assistant is worth it if you also assemble long bid documents regularly.
Frequently Asked Questions
Can AI write a competition entry statement?
As a first draft from your own drawings and numbers, yes. It cannot see the building, so anchor every claim to something you actually drew.
Is it safe to use for RFP responses?
Yes for structure and tone. Every reference, fee and date should be supplied by you rather than reproduced from an earlier submission.
Which plan fits a one-architect practice?
The low-cost plan covers client notes and short statements. A premium general assistant pays off if long bid documents are routine.