ChatGPT, Claude & the Assistants

Three general-purpose assistants dominate day-to-day design use, and they are more alike than the marketing suggests. ChatGPT (OpenAI) is the most widely used and the most feature-broad - strong all-rounder, excellent at image understanding and generation in one place, huge ecosystem of custom GPTs, and reliably capable across almost any task. Claude (Anthropic) is prized for long-form writing, careful reasoning, working over long documents, and a thoughtful, less-generic prose voice - many designers reach for it for anything involving serious reading or writing. Gemini (Google) has deep integration with Google Workspace (Docs, Gmail, Drive) and very large context, which matters if your practice lives in that ecosystem.

Beyond the big three you will meet others worth knowing: Microsoft Copilot (ChatGPT-based, woven into Windows and Microsoft 365), Perplexity (built around cited web search - better when you need sourced, current answers), and open-weight models like Llama and Mistral that some studios run privately for confidentiality. There are also assistants inside tools you already use - Copilot in Microsoft 365, AI in Notion, and AI features arriving in design software.

The honest truth about differences: on any given month one model tops a benchmark, and by the next another has leapfrogged it. For everyday design work - reasoning, drafting, summarising, analysing pasted material - a current paid model from any of the big three is more than good enough. The gap between assistants is far smaller than the gap between a good prompt and a bad one. Do not choose based on last week's leaderboard.

Every major assistant has a capable free tier and a paid tier (typically around USD 20 a month, with pricier 'pro' tiers above that). For casual use, free is genuinely useful. But if AI becomes part of how you work, paid is one of the easiest tools to justify - it is not mainly about a smarter model, it is about removing the frictions that break your flow.

Paid tiers generally unlock: the strongest model (free often routes you to a smaller, faster one, or the best model only for a few messages a day); higher usage limits so you are not cut off mid-task; larger file and image uploads; generous or exclusive access to features like image generation, deep-research modes, voice, and code execution; and better reliability at busy times. The exact boundaries shift constantly as providers reshuffle tiers, so check current terms rather than trusting a blog post.

A sensible path: start free, use it seriously for a couple of weeks on real tasks, and notice where you hit walls - the model feels weaker than you have seen, you run out of messages, an upload is refused, a feature is locked. If those walls interrupt real work, pay for the one assistant you use most. One paid subscription you actually exploit beats three free tabs you skim. And for a studio, note that consumer subscriptions and business/API plans differ on data-use and privacy terms - which matters once client information is involved (Module 9).

Modern assistants are multimodal - they take more than typed text, and this is where they earn their place in a design workflow. File upload: drop in a PDF brief, a code extract, a spreadsheet, a Word doc, and ask the model to summarise, extract, compare or answer questions grounded in it - exactly the 'put the real document on the desk' move from Lesson 1.1. Image input: upload a photo, a plan, a moodboard, a screenshot, or a sketch, and the model can describe it, critique it, extract information, or use it as context - "what style is this room?", "list the issues in this drawing," "caption these ten site photos." Image generation (in ChatGPT and Gemini) is built in for concept visuals, though dedicated tools go further (Module 3). Web search grounds answers in current, citable sources, patching the knowledge-cutoff problem. Code execution / data analysis lets some assistants run calculations and chart a spreadsheet you upload.

Two setup features quietly change everything. Custom instructions (or 'personalisation') let you tell the assistant, once, who you are and how you want it to respond - "I am an architect in India; answer concisely, use metric, flag uncertainty, skip the disclaimers" - and it applies to every chat. Set these on day one; they save you re-explaining yourself forever.

Custom instructions (set once, applied to every chat):
  I am an interior designer working in India. Use metric units
  and Indian context. Be concise and direct - skip filler and
  disclaimers. Challenge weak reasoning rather than just agreeing.
  Flag anything you are not confident about, and never invent
  figures, products or citations.

Projects (ChatGPT Projects, Claude Projects, Gemini Gems) are persistent workspaces that hold their own instructions and reference files, so an assistant can 'know' a specific project's brief, standards and context across many conversations. For a designer juggling several jobs, a project per job is the tidiest way to keep the model grounded in the right material - and it is the on-ramp to the custom assistants of Module 8.

The standalone chat window is not the only place AI shows up, and it helps to know where the general assistant fits against the alternatives. Increasingly, AI is embedded inside software you already use: Copilot across Microsoft 365, AI in Notion and Google Docs, and a growing wave of AI features arriving inside design tools - Revit, Rhino, SketchUp, Photoshop, Forma. These in-app assistants have less general muscle than a top chat model, but they win on context: they can see your document, your model or your spreadsheet directly, so you skip the copy-paste. For work that lives inside one application, the embedded assistant is often the smoother choice; for open-ended thinking, drafting and cross-cutting tasks, the standalone chat assistant is more capable.

Beyond the general assistants sit specialised AI tools the later modules cover - Midjourney and Stable Diffusion for images, dedicated text-to-3D and render-enhancement tools, code assistants like GitHub Copilot. A general chat assistant is a generalist; these are purpose-built and go deeper in their lane. A realistic 2026 studio setup is not one tool but a small kit: a daily-driver chat assistant for thinking and writing, whatever AI is baked into your core design software, and one or two specialised tools for images or rendering. The chat assistant is the hub you reason and draft in; the others are called in for their specialty.

One more distinction worth holding: consumer apps versus the API. Everything above is the consumer, point-and-click layer. Underneath, the same models are available programmatically through an API, which is how custom assistants, automations and tool-chains get built (Module 8) - and which typically carries different, stronger data-use and privacy terms, mattering the moment client information is involved. You do not need the API to get enormous value; the chat apps alone carry most designers a very long way. But knowing the layers exist tells you where to go when the chat window stops being enough - and stops you assuming the friendly chat box is all there is.

Here is a decision that will not waste your life. If you live in Google Workspace, Gemini's native integration is a strong default. If your work is writing- and reading-heavy - long documents, specs, careful prose - Claude is a natural fit. If you want the broadest features and best image generation in one place, ChatGPT is the safe generalist. If you mostly need sourced, current facts, add Perplexity for search. Any of these will serve you well; there is no wrong answer among the leaders.

Then commit. The single biggest mistake is tool-hopping - skimming five assistants shallowly, chasing every release, never learning any of them well. Depth beats breadth by a wide margin: knowing one assistant's quirks, features, custom instructions and project setup makes you far more productive than a superficial tour of all of them. Learn one deeply as your daily driver, and keep a second in your back pocket for the occasional job it does better (many designers pair ChatGPT for images with Claude for writing).

Stay current without drowning: skim releases from the big providers a few times a year, re-test your daily driver when a major new model lands, and switch only if something is meaningfully, repeatably better for your tasks - not because a benchmark moved. The assistants will keep leapfrogging each other forever; your workflow does not have to. Module 10 returns to keeping up sanely. For now: pick one, set your custom instructions, make a project, and get to work.