Is ChatGPT Plus Worth It for an Architect? A Practice Cost Check

Yes, once it saves more than about 20 minutes a month after checking: at $60 an hour, that is what the $20 fee buys. Move to ChatGPT Business when a second person joins or client files go in: it needs two seats ($40 a month billed annually, $50 monthly) and keeps business data out of model training by default.

That sum hides the real cost, which is checking. A first draft takes seconds; verifying it against drawings, contracts and standards does not, and a draft that cites a clause that does not exist costs far more than $20. Count net minutes saved, not minutes drafted, and test it on your own work for two weeks before deciding.

Follow me on Instagram@sagnikteaches

The break-even sum: what $20 a month buys in minutes

Start with the price, because it is the only number you know for certain. ChatGPT Plus is $20 a month, so it pays for itself when the time it saves is worth more than $20. Divide $1,200 by your hourly value to get the break-even in minutes a month: at $60 an hour it is 20 minutes, at $120 it is 10.

Connect on LinkedInSagnik Bhattacharya

Which hourly value should you use? There are three honest answers, and the lowest is the safest.

Subscribe on YouTube@sagnikteaches

• Cost rate. Annual pay and overheads divided by productive hours. Say $52,000 of pay plus $16,000 of overheads over 1,300 productive hours: $52 an hour.
• Margin on fixed fees. If you win work at fixed fees, a saved hour is margin you keep rather than an hour you can invoice. It is worth at least the cost rate, and more if the hour goes to something that earns.
• The evening you get back. If the drafting used to happen at nine at night, a saved hour is overtime you stop working. Only you can price that, but it is rarely zero.

Use the cost rate for the test below. It understates what a saved hour is worth in most practices, which makes a yes harder to earn and therefore more believable.

Net minutes: the checking tax on every draft

A tool that drafts in seconds can still lose you time, because everything it writes has to be read against something. Minutes of a site meeting: the draft is instant, but each action and owner has to be checked against your notes. A summary of a consultant's report: instant, but every figure has to be traced back to the page. That checking is the tax, and it is why any claim about hours saved means little until it is subtracted.

Here is an illustrative month for a sole architect valuing time at $60 an hour. The task times are invented but typical; replace them with your own.

That is 253 minutes, about 4.2 hours, worth $253 at $60 an hour against a $20 plan: a return of roughly 12 to 1. Now take a haircut. Some of those tasks would have been skipped without the tool, and some outputs will be thrown away. Count half, $126, and it is still 6 to 1 on Plus. On Business, at $40 or $50 for the two-seat minimum, the same half-value return is 3.2 to 1 or 2.5 to 1. Either way, the plan decision is not close for a practice with this much recurring text work. For Business it would be close in a practice with a third of it: about 84 minutes, worth $42 after the haircut, against $40 to $50.

The checking column is where optimistic estimates fall apart, so here is one example. The prompt asked ChatGPT to turn rough meeting notes into minutes. Both the notes and the output are invented for illustration.

Turn these notes into minutes for a design team meeting. Layout: attendees, decisions, actions (owner, due date). Add nothing that is not in the notes. Write "unclear" where the notes are unclear.

Notes: Tues site meeting, kitchen extension: client, contractor, me. Steel delivery now 14th not 12th. Client keeps island if services can move. Contractor to price option B for bi-folds. Condition 4: ask drainage consultant. Contractor's site manager away next week?

Decisions
1. Steel delivery moved to the 14th.
2. Island retained, subject to services being relocated.

Actions
- Contractor: price option B (bi-fold doors). Due: next Tuesday.
- Architect: ask drainage consultant about condition 4. Due: Friday.
- Contractor: confirm site manager cover for next week. Due: unclear.

Three things needed fixing. The output invented two due dates, "next Tuesday" and "Friday", although the notes held neither and the prompt said to add nothing. It turned a question about the site manager into an action nobody had agreed. And it did not flag that "the 14th" had no month. Each fix is small, but the check took seven minutes, which is the 10 in the first row of the table. Skip that check and the minutes you send contain two deadlines nobody agreed.

Plus or Business: what the second seat really costs

A Business Standard seat costs $20 a month billed annually or $25 billed monthly, and a workspace needs at least two paid seats, in any mix of Standard and Premium. OpenAI's pricing documentation, which lists every ChatGPT plan, sets Business at "2+ users" and describes a dedicated workspace with admin controls, SAML single sign-on and multi-factor authentication, with no training on your business data by default. Here is what that does to the monthly bill as a practice grows.

Two things stand out. Billed annually, a Business Standard seat is the same $20 as Plus, so for two or more people the choice has little to do with price; it is about the annual commitment and the controls. And a sole architect pays a $20 to $30 premium for Business, which is 20 to 30 minutes of net savings at $60 an hour. That premium buys a default (business data is not used for training), admin controls you may not need on your own, and a second seat you can hand to a partner, an associate or your office manager. Before choosing annual billing, check in the workspace's billing settings whether removing a seat takes effect at once or only at renewal, and whether any refund applies.

Plus is built for one person: it is a personal plan, and Business starts at two users. A login shared between two architects also breaks OpenAI's terms, which do not allow sharing credentials, and it puts two people's client chats in one history. Premium seats ($100 a month annually, $125 monthly) give five times Standard's usage. You would only need one if a single person hit the usage limit most days, which is unusual for drafting and summarising. For the general team-plan comparison, ChatGPT Plus vs ChatGPT Business covers the features side by side.

What the $20 includes that shortens the checking

Three Plus features cut checking time more than any model upgrade, because they make the draft start from your wording instead of the internet's. Projects keep instructions and reference files together: Plus allows 25 files per project, so each job or each document type can have its own. Custom instructions, up to 5,000 characters on Plus against 1,500 on the free plan, hold standing rules such as "plain English, no marketing language, list every assumption". And file uploads on paid plans run to 80 files every three hours, against three a day on the free plan, which is the difference between summarising one report and working through a whole consultant package in an afternoon.

Take a practice that sets up a project called Fee proposals, loads three past proposals with client names removed plus its standard terms, and writes instructions to reuse the practice's wording, list every assumption, and leave a bracket wherever a fee or date is needed. In the table above the fee-proposal check took 65 minutes. With the practice's own wording as the starting point there is less new text to verify, so say the check falls to 40 minutes. That is 25 minutes saved on a task done once a month, worth $25 at $60 an hour, which already exceeds the whole Plus fee. The figure is illustrative; the lesson is where to look for savings. Set up the reference files once and the checking column shrinks on every later use.

Here is what those project instructions might look like once filled in for a small practice that works on domestic extensions and refurbishments. They run to about 750 characters, so they also fit the smaller allowance on the cheaper plans.

Project: Fee proposals (domestic work)
You draft fee proposals for a small architectural practice.
- Reuse the structure and wording of the three example proposals in this project.
- Plain English. No marketing language, no superlatives.
- Stages follow the practice's own list: brief, concept, planning, technical design, tender, site.
- Never invent a fee, percentage, date or programme. Write [FEE], [DATE] or [WEEKS] instead.
- Never name a planning outcome or approval as likely.
- End with a list headed "Assumptions to confirm", one line each.
- If the brief mentions a heritage-protected building, a shared wall with a neighbour
  or a boundary dispute, add the line: "Specialist advice may be needed: [WHO]".
- Metric units throughout.

Two lines in that block do most of the work. The bracket rule stops the draft from guessing a fee, which is the error that costs most if it slips through, and the assumptions list turns the check from a re-read into a tick-list. When a draft comes back with a figure where a bracket should be, that is the first thing to fix and a sign the instruction needs to sit higher in the list.

On Business, projects can be shared with colleagues in the workspace. That matters for the second seat: a colleague can use the same set-up instead of rebuilding it. It is also why a practice that adopts ChatGPT seriously should keep the instructions and reference files in one agreed place, not in one person's head.

Go at $8 or Pro from $100: the plans either side of Plus

Two other personal plans bracket Plus, and the same break-even sum settles both. Go costs $8 a month, so at $60 an hour it pays once it saves 8 minutes. It keeps the 25-file projects that Plus has, but custom instructions stop at 1,500 characters instead of 5,000, it is pitched at lighter use than Plus, its listing says it may include ads, and it leaves out ChatGPT Work, the agent that carries out multi-step tasks, which starts at Plus. Pro starts at $100 a month, with five times Plus's usage on that tier and dearer tiers above it.

For an architect the 1,500-character cap is the practical test of Go. The fee-proposal block above fits inside it, but a practice that wants one standing set of rules covering tone, units, stages, exclusions and an assumptions list across all its work will run out of room and end up repeating the overflow in every project. Write the rules you actually want, count the characters, and if they fit and your log shows you seldom hit a limit, try Go for a month. The $12 difference from Plus is worth 12 minutes at $60 an hour, so the question is only whether the extra instruction space and allowance save that much.

Pro is a different calculation. Against Plus, its extra $80 needs 80 more minutes of net saving a month at $60 an hour, and that only comes from time currently lost waiting for a usage limit to reset. If one person works through large consultant packages every afternoon and hits the Plus limit most days, it can pay. Otherwise a second Plus seat, or Business, buys more for the practice than one Pro seat. Moving between Go, Plus and Pro also changes nothing about confidentiality: all three are personal plans with the same model-training switch, covered next.

Training settings and client files decide the plan

For an architect the plan question is often a confidentiality question wearing a price tag. On Plus, a switch in privacy settings controls whether your chats are used to improve the models. The label changes from time to time, so look for the model-training switch rather than a name, and check it is off before the first client detail goes in. Temporary Chat is not used for training, though OpenAI may keep a copy for up to 30 days for safety, and deleting a chat does not erase it at once: OpenAI says deleted chats are removed within 30 days. On Business, the default runs the other way: OpenAI lists no training on your business data by default.

My own sorting rule for an architectural practice, which is a judgement and not OpenAI's advice, has three bands:

• Safe on any plan: your own writing with no client detail. Website copy, CPD notes, generic email templates, fee-proposal wording with names removed.
• Plus with training off, or Business: project descriptions with the client's name and address stripped, anonymised meeting notes, summaries of public planning documents.
• Business only, and only if the client agreement allows it: drawings, models, client financial details, consultants' reports marked confidential, anything under an NDA.

The plan does not override your contracts. Read the client agreement and your professional indemnity insurer's terms for any clause on third-party AI before uploading anything with a client's name on it. The copyright and data side of this is covered in copyright and client data risks of AI in architecture practices, and whether ChatGPT trains on client data on business versus free plans goes deeper on the settings.

A two-week hours-saved test with a filled-in log

This is the part most cost comparisons skip. Run it for two weeks on real work.

1. Pick five recurring text tasks: for example meeting minutes, consultant chasers, a fee proposal, a design statement, a report summary.
2. For each, record the usual time. Use your last timesheet, or time the first instance the old way with a stopwatch.
3. Do the following instances with ChatGPT and time everything: writing the prompt, reading the draft, checking it against your sources, and editing.
4. Mark each output as used, partly used or not used.
5. Count used outputs in full, partly used ones at half, and unused ones at zero.
6. Multiply the two-week total by about 2.2 to get a month. Compare it with three times the break-even from the first table.

The counted total is 102 minutes in two weeks, so about 224 minutes a month. At $60 an hour that is $224, eleven times the Plus fee, 5.6 times Business at $40 and 4.5 times Business at $50. Against a threshold of three times break-even (60 minutes for Plus, 120 for Business annual), the log passes both. If a log came out under one and a half times break-even, I would not pay; between the two, I would extend the test a fortnight before deciding.

Do the arithmetic in a spreadsheet, not in the chatbot. Language models are unreliable at sums, which is a reason to keep the log yourself; why AI is bad at maths explains what to check.

Then check it again at the three-month mark, because a plan that passes in week two can be idle by month three once the novelty fades and tasks drift back to the old way. Keep the same log for one week, double it, and compare with the same thresholds. Say the week shows 30 counted minutes: doubled, that is 60 minutes a fortnight, about 130 a month, still more than six times the Plus break-even at $60 an hour, so keep paying. If the doubled week falls under one and a half times break-even, cancel or drop a plan; if it has grown, that is your evidence for a second seat.

Four ways the sum goes wrong

Checking left out. A report summary arrives with a flow rate of 4.2 where the report says 2.4. It is a transposed pair of digits, easy to miss, and tracing it costs ten minutes. Every figure in a summary needs a page reference, and you should ask for them: "give the page number for each figure".

Reading drawings. OpenAI's developer guide to image inputs, which describes how its models handle pictures, says images are scaled down to fit pixel limits before the model reads them, that small text should be enlarged to be read reliably, that rotated or upside-down text may be misread, and that the models struggle with precise spatial positions. A drawing sheet breaks all four rules at once: tiny dimension strings, text running up the page, and meaning that depends on where a line sits. Give it text it can read, such as schedules, specifications and notes, crop to one detail at a time if you must show a drawing, and treat any number it reads off a sheet as unchecked.

Invented references. A draft specification or design statement may cite a standard or a clause number that does not exist, or one that exists with a different meaning. Never paste a reference you have not opened at source.

Paying ahead of the evidence. A practice moves to Business, or buys Premium seats for everyone, before any log exists. The cheapest mistake in this whole area is the plan that was bought on a Friday afternoon because the demo looked good.

Sole practitioner, two partners, five staff: who buys what

A sole architect doing domestic extensions. The log above is this practice. Plus at $20 passes comfortably if client detail stays out of the prompts. If a client's agreement bans AI training on project data, or you would rather evidence a default than rely on a personal setting, Business costs $20 to $30 more a month and still passes at 5.6 times break-even.

A two-partner practice. Business billed annually is $40 for two seats, the same as two Plus accounts, and gives the practice a workspace instead of two personal histories. Each partner has their own login. The decision is the annual commitment, not the price.

A five-person practice where two people write most documents. Business does not require a seat for every employee, only two seats in total. Buying two seats ($40 annually, $50 monthly) for the people who write fee proposals and reports, and nothing for the other three, is a sensible start. Revisit in three months with real usage in front of you. Anyone not on the plan should not be pasting client material into a personal account, which is a policy point more than a purchasing one; see company AI accounts instead of personal logins.

When an architect should skip the subscription

• You cannot name three text-heavy tasks that recur every month.
• Your client agreements ban third-party AI processing and you cannot change them.
• Your two-week log comes out below one and a half times break-even.
• Most of your week is drawing and modelling in CAD or BIM, so the text tasks it helps with are a small share of your hours.

In any of those cases, use the free plan for occasional tasks and revisit when something changes. The assistant matters too: whether ChatGPT, Claude or Gemini fits architecture work best is a separate question, answered in Claude vs ChatGPT vs Gemini for architects. If you would rather run this test with someone who has seen how practices set it up, that is what my AI implementation consultation covers.

Plan questions architects ask

Can two architects share one ChatGPT Plus login?

No. OpenAI's terms do not allow sharing account credentials, and Plus is a personal plan. Sharing also puts two people's client chats in one history. If two of you will use it, ChatGPT Business with two seats costs the same per seat as Plus when billed annually.

Is ChatGPT Business cheaper than two Plus accounts?

Not cheaper, but level on price if you pay annually. A Business Standard seat is $20 a month billed annually or $25 monthly, against $20 for Plus. Two Business seats cost $40 or $50 a month, and you gain admin controls and a no-training-by-default position for business data.

Does ChatGPT Plus use my client documents for training?

Consumer plans have a switch in privacy settings that controls whether your chats are used to improve the models, so check it is off before you paste anything about a client. Business plans do not train on business data by default. Neither setting replaces reading your client contract and your insurer's terms.

Should I pay for Claude or Gemini instead?

The three assistants overlap on drafting and summarising, so the sensible way to choose is the same two-week log on each. Time five recurring tasks, count only the output you used, and compare net minutes. Pay for whichever saves the most after checking, not the one with the best demo.

Further reads

• How Much Does AI Cost a Small Architecture Practice per Month? — A full monthly budget for a small practice, beyond one subscription.
• AI Policy for Architects: A Ready-to-Use Checklist and Clauses — Write the client-data rules your staff should follow.
• How to Measure Time Saved After Rolling Out AI in a Small Firm — A wider method for proving time saved across a whole practice.
• How Architects Use AI for Fee Proposals, Briefs and Paperwork — Fee proposals and briefs, the tasks most likely to pay back.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: OpenAI's pricing documentation (Go $8, Plus $20, Pro from $100, Business $20 a user billed annually or $25 monthly for 2+ users, admin controls, SAML SSO, MFA, no training on business data by default), read on 29 October 2026; OpenAI's developer guide to image inputs; OpenAI help pages on Projects, custom instructions, file uploads, Temporary Chat and deleted chats, as checked in October 2026.