ChatGPT Project Instructions Character Limit: Official Answer (2026)

People searching for the official ChatGPT project instructions character limit want one number from OpenAI. OpenAI does not publish one. The help center article on Projects explains what the instructions field does and lists file limits per plan, and then says nothing about how long the instructions can be.

That absence is the real answer, and it changes how you should think about the field. Below is what OpenAI does document, where the widely repeated 8,000-character figure comes from, a two-minute test to find your own ceiling, and the part most guides skip: what to put in the characters you have so that a project in ChatGPT writes like you instead of like everyone.

What Does OpenAI Officially Say About the ChatGPT Project Instructions Character Limit?

Nothing numerical. OpenAI's Projects in ChatGPT help article, updated within the past month at the time of writing, describes how to add instructions through Project settings and gives an example ("Act like my marketing mentor. Be concise. Use bullet points. Ask clarifying questions."). It states that project instructions apply only within that project and override your global custom instructions. It then moves on to files, memory and sharing without ever naming a character cap.

That is unusual for OpenAI, because the neighbouring feature is documented precisely. The custom instructions FAQ answers "Is there a character limit?" with two numbers by plan. The Projects article has no equivalent question. If the limit mattered to OpenAI's support team as a frequent complaint, it would be there. The reasonable reading is that the field is large enough that few users hit it, and OpenAI reserves the right to change it without a release note.

So the honest answer to the navigational query is: the official limit is unpublished. Anyone quoting a precise figure, including this site in earlier guides, is reporting observed behaviour, not documentation.

The Numbers OpenAI Does Publish

OpenAI documents limits for the layers around project instructions, and those limits shape what you should keep in the instructions field versus elsewhere.

Two details from the official pages matter more than the missing number. First, the 5,000-character paid limit on global instructions arrived in July 2026; before that every plan had 1,500. OpenAI changes these ceilings quietly, which is why a figure from a 2025 forum thread should not be treated as current. Second, project instructions override global custom instructions inside the project. They do not stack. If your global field says "formal" and your project field says "casual", the project wins, and the project also has to carry anything from the global field you still want applied.

For the global fields specifically, our guide to the ChatGPT custom instructions character limit covers what fits in each one.

Where the 8,000-Character Figure Comes From

The 8,000 number has two sources, neither of them the Projects documentation. The GPT builder enforces 8,000 characters on Custom GPT instructions, and that limit is visible in the builder interface and widely discussed on OpenAI's developer forum. Projects launched a year after Custom GPTs with a similar instructions field, and users who pasted long blocks reported the same truncation point.

That is decent evidence. It is also the kind of evidence that goes stale without anyone noticing, because ChatGPT does not show a character counter in the Projects settings field and does not warn you when it truncates a paste. A user on a plan that quietly received a different ceiling would have no signal. Our earlier complete guide to ChatGPT project instructions repeated the 8,000 figure for that reason, and the ChatGPT Projects setup guide describes the field as holding "several thousand characters", which is the more defensible phrasing.

Use 8,000 as a planning assumption. Do not use it as a fact you would repeat to a colleague without checking.

How to Measure Your Own Project Instructions Ceiling in Two Minutes

The test is simple and gives you a number that is true for your plan and today's interface, which is better than any published figure.

1. In a text editor with a character counter, build a block of known length. Numbered lines work well: "0001 lorem ipsum", "0002 lorem ipsum", and so on, each line a fixed width. Make it 10,000 characters.
2. Open a throwaway project, open Project settings, paste the block into instructions, and save.
3. Close and reopen Project settings. Scroll to the end of what survived and read the last line number.
4. Multiply by your line width. That is your ceiling. If the full 10,000 survived, repeat with 20,000.

Write the result down with the date. If you rely on a long instructions block, repeat the test after any visible ChatGPT interface change, because that is when ceilings tend to move. The Projects article was updated twelve days before this post was written, and the custom instructions limit moved three months earlier. Neither change was announced inside the product.

Why the Character Count Is the Wrong Thing to Optimize

Most people searching for this limit are not about to paste 8,000 characters of marketing-mentor instructions. They are trying to make a project write in their voice, they have filled the field with description, and the drafts still read like ChatGPT. They assume the field is too small for what they need. Usually the field is fine and the content is the problem.

Here is a typical project instructions block for a writing project. It is 310 characters:

You are my writing assistant. Write in my voice: professional yet friendly, clear and confident, warm but not casual. Keep things concise. Avoid jargon. Match the tone of a senior consultant who is approachable and direct. Sound human, not like an AI.

Every phrase in that block describes thousands of people. "Professional yet friendly" is a region of a map, and the model will land in the centre of that region every time, which is the same centre it lands in for every other user who wrote those words. Doubling the block to 620 characters of synonyms moves nothing. Hitting the ceiling with 8,000 characters of adjectives moves nothing either, because the problem is the unit, not the quantity.

A Tone Label Spends Characters Without Buying a Voice

AI does not know your tone. It knows words that describe tone, and it has seen those words attached to millions of writers who are not you. A voice lives somewhere else: in sentence length and its variance, in where you place qualifiers ("probably" at the start of a clause or buried in the middle), in which transitions you use and which you never use, and in punctuation habits such as how often a colon appears or whether you ever write a one-word sentence.

Those properties are countable from a few pages of your own writing, and when they are stated as rules the model can check its own draft against them. "Warm" cannot be checked. "Average sentence 16 words, range 6 to 28, at least one sentence under 8 words per paragraph" can. The second rule costs about 90 characters and does more than the first does at any length. The difference between custom instructions and a Style Profile is exactly this difference: one describes how you hope to sound, the other records how you measurably write.

If you want to see what a measured description looks like before writing one, a free Writing DNA Snapshot analyses a sample of your writing and returns the properties above as numbers you can paste straight into a project.

What a Character Actually Buys You: Measurable Rules

A rule earns its characters when the model could violate it and you could point to the violation. Use that test on every line in your instructions block.

Each rule on the right is between 60 and 120 characters. Fifteen of them run to roughly 1,500 characters, which fits inside even the Free-plan global custom instructions limit, let alone a project field whose observed ceiling is several times that. The limit stops being a constraint the moment the content becomes measurable.

Does a Better Model Make the Limit Matter Less?

It makes the content of the field matter more, and the length matter about the same. Model capability and the conditioning you supply about yourself are separate layers. OpenAI improves the first with every release. It cannot supply the second, because the second is data about one person, and only that person has it.

What a more capable model does change is how faithfully it executes what you give it. Older models would drift from a precise rule after a few exchanges; current ones hold it. So the same 1,500 characters of measurable rules produce a closer match on every model upgrade, while the same 1,500 characters of adjectives produce a more polished version of the generic centre. The upgrade multiplies whatever is in the field. A measured profile appreciates. A tone label does not.

This is also why a profile is worth writing once and carrying between tools. The rules above are not ChatGPT-specific. A Writing Twin for ChatGPT, Claude and Gemini is the same measured profile pasted into each product's instructions field, and its value rises with every model release on every platform.

A Project Instructions Block That Fits Any Plan

Below is a complete project instructions block for a work-writing project. It is about 1,250 characters, so it would fit the Free-plan global field with room to spare, and it leaves most of a project field empty for project-specific context.

ROLE: Draft emails, memos and Slack messages in my voice for my team and clients.

STRUCTURE: Open with the ask or the conclusion. Context in sentence two, never sentence one. Paragraphs of 2 to 4 sentences. One idea per paragraph.

SENTENCES: Average 15 words. Range 5 to 30. At least one sentence under 8 words in every paragraph. Fragments allowed for emphasis.

HEDGES: None in the first sentence of a paragraph. At most one per paragraph, placed mid-sentence ("the vendor will probably ship Friday", never "Probably the vendor").

PUNCTUATION: Colons to introduce a point. No semicolons. No exclamation marks. No em-dashes.

TRANSITIONS I USE: "So", "Which means", "One more thing", "In practice". TRANSITIONS I NEVER USE: "Moreover", "Furthermore", "That said", "Additionally", "I hope this finds you well".

WORDS I NEVER USE: circle back, going forward, reach out, touch base, bandwidth, move the needle, low-hanging fruit.

REGISTER SHIFTS: To my manager, the ask is the last sentence and the body is evidence. To a peer, the ask is the first sentence. To a client, add one sentence of context before the ask and sign off with my first name only.

CLOSING: End on the next step with a date. No "let me know if you have any questions".

Every line is checkable. If a draft opens with "I hope this finds you well", the draft violated a rule and you can say which one. If you wrote "warm and professional" instead, you would have nothing to point at. Full setup steps for the project itself, including memory settings and file placement, are in the ChatGPT Projects setup guide.

Where Knowledge Files Fit When Instructions Run Out

If you do approach the ceiling, the overflow does not belong in instructions anyway. Project files are the right home for writing samples, long banned-phrase lists and format templates, and they are governed by the file limits OpenAI does publish (5, 25 or 40 files by plan). Files are consulted when the model judges them relevant to the current message, while instructions apply to every message. That makes instructions the place for rules and files the place for evidence.

A sound split for a writing project:

• Instructions: the measurable rules, under 2,000 characters, applied every turn.
• Files: three to five labelled samples of your real writing ("peer email", "client proposal", "board update"), each a page or less, so the model has reference material when a draft resembles one of them.
• Memory: left to its job of storing facts about your work, which it does well, and which are not your voice.

Our guide to ChatGPT project knowledge files and persona covers the file layer in detail. For what a complete measured profile contains, start with what a writing style profile is.

Frequently Asked Questions

What is the official ChatGPT project instructions character limit?

There is no officially published number. OpenAI's help center article on Projects describes the instructions field, states that project instructions override your global custom instructions, and lists file limits per plan, but it does not state a character cap for the instructions field. The only character limits OpenAI publishes are for global custom instructions: 1,500 characters on Free and Go, 5,000 on Plus, Pro, Business, Enterprise and Education plans.

Is the ChatGPT project instructions limit 8,000 characters?

8,000 characters is the figure most guides repeat, including earlier guides on this site. It matches the limit the GPT builder enforces for Custom GPT instructions, and user testing has reported the same ceiling in Projects. OpenAI has never documented it for Projects, and limits in the ChatGPT interface have changed without announcement before. Treat 8,000 as a working assumption you verify yourself, not as an official number.

How do I check the character limit of my own ChatGPT project instructions?

Open any project's settings, paste a block of text whose length you already know (a counter in any text editor gives you the number), save, reopen the settings and compare what survived. If the field truncated silently, the surviving length is your ceiling on your plan and interface version. The test takes two minutes and gives you a number that is true today, which is more than any blog post can promise.

Does the ChatGPT project instructions character limit matter for writing in my own voice?

Less than people assume. A tone label such as "professional yet friendly" uses a handful of characters and describes thousands of writers. The rules that actually reproduce a voice are measurable: sentence-length range, where hedges sit, which transitions never appear, how often you use a colon. A block of those rules fits in roughly 1,200 to 2,000 characters, well under any limit you are likely to hit.

Do project instructions add to my global custom instructions or replace them?

They replace them inside the project. OpenAI's Projects article states that project instructions apply only within that project and override your global custom instructions. Anything from your global setup that you still want in force, such as your banned-word list, has to be repeated in the project field. In shared projects the override is stricter still: collaborators' personal custom instructions and memories are not used at all.

---

The limit you were searching for does not exist in OpenAI's documentation, and the limit you measure yourself will almost certainly be larger than what a good voice setup needs. The characters that matter are the ones that state something checkable about how you write. If you would rather measure than guess, a free Writing DNA Snapshot turns a sample of your writing into those numbers, ready to paste into a ChatGPT project, a Claude project, or anywhere else you write with AI.