Most people already know AI can draft, summarize, and brainstorm. What it's less known for — and what actually moves a goal forward — is holding context on one thing across weeks and nudging you back to it. I've started calling that pairing the new AI superpowers: focus and followthrough, and it's less about a specific feature than a way of setting up a project so the AI carries the thread instead of you re-explaining it every session.
Short answer: Open one dedicated ChatGPT Project or Claude Project per goal, write custom instructions naming the goal, the deadline, and what "done" looks like, then log a short daily check-in in that same thread. The AI keeps the plan current; you still need an outside reminder, since neither tool proactively pings you.
I tested this over three weeks on two different goals — finishing a freelance pitch deck and getting through a backlog of client invoices — using one ChatGPT Project for each instead of my usual habit of opening a fresh chat every time I had a question. The difference wasn't the quality of any single answer. It was that by day four, the AI already knew what I'd done, what I'd skipped, and why, without me copy-pasting a status update into a blank window first.
What you'll need
A ChatGPT account with Projects (available on Free, Plus, and up — Free caps you at 5 files per project, Plus and Go raise that to 25, per OpenAI’s help center) or a Claude account, where Projects require Pro at $20/month since the free tier doesn't include them, per Claude’s own pricing page. Either works; pick whichever you already pay for. You also need one real goal with a deadline — not "be more productive," but something with a finish line, like "submit the grant application by the 15th." Have a rough sense of the steps involved before you start; the AI will refine them, but it can't invent a deadline you haven't set. Last thing: a place outside the AI to remind you to actually open the project each day, since that part is still on you.
Step-by-step: building an AI focus and followthrough system
1. Open one project, one goal
In ChatGPT, click Projects in the sidebar and create a new one named after the goal, not the tool ("Grant Application — Due Aug 15," not "Writing Help"). In Claude, the same workspace is also called a Project. Don't reuse a general-purpose project for this — mixing goals back into one thread is exactly what breaks the followthrough later.
2. Write custom instructions that define "done"
Every project lets you set instructions that apply to every chat inside it. Write the goal, the deadline, and a concrete definition of done: "Done means the deck has 10 slides, a budget page, and one reviewed draft, submitted by August 15." In my testing, skipping the concrete "done" line was the single biggest reason the AI kept regenerating vague, restart-from-scratch plans instead of building on what I'd already done.
3. Break the goal into a week of next actions
Ask for a plan broken into daily or every-other-day steps, small enough that each one is a single sitting. I asked for steps I could finish in under 45 minutes each — anything bigger and I never started it the same day I planned it.
4. Log a short daily check-in in the same thread
This is the step people skip. At the end of each work session, post two or three lines: what you did, what you skipped, and why. Ask the AI to adjust tomorrow's step based on that. In my test, the check-in itself took under two minutes most days, and it's the only reason the plan still matched reality by day ten instead of drifting into a to-do list nobody was updating.
5. Run one weekly review instead of daily replanning
Once a week, ask for a review comparing the plan to what actually happened, then a revised plan for the coming week. Daily re-planning wastes time; weekly is enough cadence to catch drift without turning every session into a strategy meeting.
6. Add an outside nudge
Neither ChatGPT nor Claude will message you first — Projects hold context, they don't push notifications. Set a recurring calendar block or a task-app reminder to open the project each day. I used a plain phone alarm during testing, and it was the least glamorous part of the whole system and also the part that made the other five steps actually happen.
Example prompts you can copy
- "This project is for [goal], due [date]. Done means [specific deliverable]. Break this into daily steps I can each finish in under 45 minutes, starting today."
- "Here's my check-in: I finished [X], skipped [Y] because [reason]. Adjust tomorrow's step based on that, and flag if I'm now behind the deadline."
- "It's the end of the week. Compare this week's plan to what I actually logged, tell me honestly where I'm behind, and give me next week's plan."
- "I'm stuck on [specific step]. Don't give me the whole plan again — just help me get unstuck on this one part."
- "Summarize everything logged in this project so far into a one-paragraph status update I could send to someone else."
The pattern across all five: name the specific thing (the step, the date, the number), and ask for an adjustment to the existing plan rather than a fresh one. Open-ended requests like "help me stay on track" produced generic pep-talk answers in my testing; specific ones produced a plan I could actually act on the same day.
Common mistakes to avoid
The mistake that cost me the most time: starting a new chat instead of staying in the project thread, which meant re-explaining the goal from scratch and losing every prior check-in. Second, skipping the "done" definition — without it, the AI treated the goal as open-ended and kept suggesting more scope instead of narrowing toward a finish line. Third, treating the daily check-in as optional; the two weeks I logged it consistently, the plan stayed accurate, and the one week I didn't, day four's plan was still built on day one's assumptions. Fourth, asking for a full replan every single day instead of a small adjustment, which burned time and reset momentum instead of building on it. Fifth, expecting the AI to remember to remind me — it won't, and I lost two days early on assuming an open project tab counted as a reminder.
ChatGPT Projects vs. Claude Projects vs. a dedicated task app
| ChatGPT Projects | Claude Projects | Dedicated task app (e.g. Routine AI) | |
|---|---|---|---|
| Cost to start | Free (5 files/project); Plus is $20/month for 25 files | No free-tier access; Pro is $20/month for unlimited projects | Free plan available |
| Holds a goal's full context | Yes, across chats in the same project | Yes, across chats in the same project | Yes, as tasks and notes, not conversation |
| Breaks a goal into daily steps | Yes, on request | Yes, on request | Manual, or via its own AI assistant |
| Proactively reminds you | No | No | Some plans, via notifications |
| Best for | Reasoning through a fuzzy goal that changes as you go | Same, if you already pay for Claude | A goal that's really just a checklist with a due date |
If your goal is really just a list of tasks with due dates and no reasoning involved, a dedicated app does that job with less setup — Routine AI is the one I've tested for that, and its free plan covers solo task tracking without needing a $20/month AI subscription at all.
Tools that make this easier
If you don't already have an AI account set up, my guide to ChatGPT’s free plan covers getting started before you build a project on top of it, and my guide to using Claude does the same on that side. Once the daily-followthrough part is working, ChatGPT’s agent mode can take over steps that involve actually visiting websites or filling in forms, instead of you doing the legwork the AI only planned. If you're weighing whether any of this is worth a paid AI plan in the first place, my best AI tool for small business guide compares the cost against what you'd actually use day to day, and my starter kit for AI is a broader look at which accounts to open first if you're setting this up from zero.
My verdict
This system isn't a product, so there's nothing to buy — it's a way of using a project or workspace feature that ChatGPT and Claude already ship. The honest tradeoff is that it only works if you keep logging check-ins; the moment you stop, it degrades into the same forgotten to-do list as every other productivity system. What it got right in my testing was cutting the time I spent re-explaining context to near zero and catching drift within a week instead of a month. If your goal is genuinely just a checklist, skip this and use a task app instead — the AI project only earns its keep when the goal is messy enough to need actual reasoning between steps.
Frequently Asked Questions
Is this free to set up?
Mostly. ChatGPT's free plan includes Projects with a 5-file cap, which is enough for a single goal's instructions and a couple of reference documents. Claude's free tier doesn't include Projects at all — that requires Claude Pro at $20/month, per claude.com’s pricing page.
How long does it take to set up?
About 15 minutes for the first project — naming it, writing the "done" definition, and getting the first week's steps. After that, daily check-ins took me under two minutes each in testing, and the weekly review took about 10.
What's the easiest way to start?
Pick one goal with a real deadline, open a single project for it, and write the "done" definition before you ask for a plan. Skipping straight to "make me a plan" without that context produced vaguer, harder-to-follow steps in my testing.
Does this work with Claude instead of ChatGPT?
Yes — Claude's Projects feature works the same way: custom instructions, persistent context across chats, and a place to log check-ins. The difference is cost, since Claude requires the $20/month Pro plan while ChatGPT's free tier includes a limited version.
Will the AI remind me to check in on my own?
No. Neither ChatGPT nor Claude sends proactive notifications inside a project. You need an outside reminder — a calendar block, a task app, or a plain alarm — or the whole system quietly stops.