Humbyte Blog
Revising a thesis with AI again and again: how to stop re-collecting your requirements and context every time
When you use AI to help write a thesis, the hardest part is usually not the first draft but the many rounds of revision that follow.
A thesis may go through many rounds of revision. Today you adjust the line of argument, tomorrow you unify the terminology, and later you add explanations, rework the wording, or reorganize content based on feedback from your supervisor or reviewers.
The thing is, no revision starts from zero. The AI needs to know not only what to change this time, but also which requirements must always be followed, which content is already settled, and which parts must not be changed casually.
1. A thesis revision has more than one set of requirements
When revising a thesis with AI, you usually deal with several kinds of information:
- Long-term requirements:the topic, core arguments, terminology, writing style, citation format and so on, which must stay consistent across many rounds of revision.
- Current requirements:what this round needs to fix, such as tightening the wording, strengthening the argument, adjusting the logic, or responding to review comments.
- Reference material:text you are already happy with, wording you settled on earlier, and anything else related to the current revision.
- Revision results:which of the versions the AI produced can be kept, and which still need work.
Not all of this needs to be sent to the AI every time. Different tasks need different combinations, but the long-term requirements and settled content must not get lost.
If you put it all together by hand every time, it soon turns into a lot of repeated work.
2. Why does the AI stop following your requirements as the revisions go on?
A single chat can run for dozens of rounds or more.
At first, the AI knows which terms must be consistent, which phrasing is not allowed, and which conclusions are settled. But as the chat grows, new requirements, old versions and discarded revision ideas all get mixed together in the history.
Then things like this can happen:
- Terms you explicitly asked to keep consistent get replaced with other wording.
- Content that was already settled gets changed by accident in a later revision.
- You asked to adjust one part, but the AI changed parts that did not need touching.
- To fix the same problem, you have to explain earlier requirements all over again.
This does not mean an AI will always forget requirements as a chat gets longer, but complex, lengthy context does raise the risk of missing or mixing up constraints.
A practical approach is to start a new chat regularly and give it only the requirements and reference material the current revision really needs.
3. A new chat solves one problem but creates another
A new chat cuts down interference from the old history, but it also means you have to provide the necessary information again.
For example, before continuing to revise a passage, you may need to:
- Find the long-term writing requirements for the thesis in earlier chats.
- Find the specific requirements for this revision.
- Find the content you were already happy with, as reference.
- Put these materials into the new chat in a sensible order.
- Add the text you want to revise now and start the next round.
If the material is scattered across several AI chats, your thesis document and your notes, you keep switching, searching, copying and pasting.
A thesis is revised many times, so this routine may repeat many times too.
The real hassle is not copying text once. It is re-organizing information you already organized long ago, again and again.
4. Save and reuse this content with MindDesk
MindDesk is a browser extension I built. It does not write your thesis for you. It helps you manage the prompts and answers you use again and again in AI chats.
You can save the information from your thesis revisions separately:
Save long-term and revision requirements
Save your usual writing requirements and revision rules as prompts, organized by thesis or task. Next time, just pick them instead of writing them again.
Keep the content you are happy with
When an AI answer is good, save it as a Context right below the answer. Settled wording, revision results and anything you need to refer back to can all be kept.
Combine context for each revision task
When you are about to start a new chat, check the long-term requirements, specific revision instructions and related reference material you need this time in the side panel, then adjust their order.
There is no need to copy your entire chat history; choose only the information this task needs.
Insert into a new AI chat in one click
Once combined, insert the content into the AI site’s input box in one click and continue revising.
All of this happens in the browser side panel, so you do not have to dig through old chats or keep switching apps.
5. Not just for theses
The same approach works for other work that is revised over a long period, such as research proposals, technical reports, project documents and business copy.
They have one thing in common: the work does not end in one AI chat, and earlier requirements and intermediate results keep being used in later tasks.
That is the problem MindDesk sets out to solve:
making the requirements you have written, the content you have organized and the answers you are happy with usable again in your next AI chat, instead of starting over every time.
MindDesk supports ChatGPT, Claude, Gemini, Grok, DeepSeek, Qwen, Doubao and other AI sites, and saved content can be reused across different AIs.