AI can be useful during ERC proposal preparation when it is used to challenge, organise and clarify your thinking.
It is much less useful when it is asked to invent the science, generate unsupported claims or produce generic grant language around a project it does not understand.
The ERC Scientific Council recognises that researchers may use AI technologies or other external help for activities such as brainstorming, literature search, revision, translation and summarisation. The applicant remains fully responsible for authorship, accuracy, plagiarism and good scientific conduct.
Official source: ERC Scientific Council position on AI
Before you use any prompt
Three rules matter.
1. Keep intellectual ownership
AI should help you examine your scientific argument. It should not decide what your research programme is.
2. Verify every factual claim and citation
Treat generated references, numbers and statements as unverified until you check the original source.
3. Protect confidential or unpublished material
Do not paste sensitive scientific ideas, unpublished results, personal data or confidential consortium information into a system unless you understand and accept how that system handles the content.
Prompt 1 - separate knowledge, assumptions and unknowns
Use this before you define the central question.
I am developing an ERC Starting Grant idea in [field]. Below is my description of the scientific problem. Separate it into: (1) what is established knowledge, (2) what appears to be an assumption or dominant interpretation, and (3) what is genuinely unresolved. Do not propose a project yet. Identify ambiguities and tell me which statements would require evidence or citation.
Scientific problem: [paste concise description]
Why use it: it helps expose whether the proposal is built around a real unknown or around an optimisation task.
Prompt 2 - stress-test the central question
My proposed central scientific question is: [question].
Act as a sceptical expert researcher. Test whether the question is scientifically important, precise enough to organise a five-year programme, and genuinely uncertain. Identify hidden assumptions, competing explanations and any wording that makes the question look predetermined. Do not rewrite the project unless I ask.
Why use it: the central question needs to survive scientific challenge before it becomes polished proposal text.
Prompt 3 - test ERC-level ambition
Here is my project concept: [concept].
Analyse the ambition at four levels: (1) optimisation, (2) extension into a new regime, (3) explanation of an unresolved mechanism, and (4) potential field-level change. Tell me which level the current concept most closely resembles and what scientific evidence would justify a stronger claim. Do not use promotional language.
Why use it: ambition is easier to discuss when it is separated from adjectives and specifications.
Prompt 4 - challenge the hypothesis
My central proposition or hypothesis is: [hypothesis].
List the strongest scientific reasons it could be wrong. Identify at least two plausible alternative explanations. For each explanation, state what observation, experiment or theoretical result would discriminate between them. Do not assume my preferred hypothesis is correct.
Why use it: a useful ERC hypothesis should be vulnerable to evidence.
Prompt 5 - build an outcome tree
My central hypothesis is: [hypothesis].
Construct three scientifically plausible outcome classes: strongly supported, partially supported / boundary conditions identified, and rejected. For each outcome, explain what new knowledge is produced, what interpretation changes and what future research direction becomes possible. Avoid inventing facts that are not in my input.
Why use it: this helps develop scientific optionality rather than a brittle single-path project.
Prompt 6 - test Part I / B1 clarity
Read the following ERC Part I draft as an expert scientist outside my exact subfield. Without rewriting it, identify:
- the central scientific question;
- the unresolved limitation;
- the main proposition or hypothesis;
- the objectives;
- the strongest field-level consequence;
- anything you could not infer confidently.
Then tell me where the argument becomes difficult to follow and which paragraph carries too many ideas.
Draft: [paste text]
Why use it: if an informed reader cannot extract the scientific architecture, the proposal may need restructuring rather than copy-editing.
Prompt 7 - check whether B1 reads like a methods paper
Review this ERC Part I section. Estimate how much of the text is devoted to scientific motivation, state-of-the-art limitation, central question, objectives, overall strategy and detailed methodology. Identify places where implementation detail arrives before the scientific reason to care is clear. Do not shorten the text automatically; explain the imbalance first.
Why use it: Part I needs a scientific argument, not a compressed B2.
Prompt 8 - design a conceptual figure
Convert the scientific logic below into a conceptual figure plan suitable for a five-page ERC Part I. The figure must clarify the science rather than decorate the page. Use no more than 5-7 visual elements. Show the relationship between current understanding, unresolved limitation, central question or proposition, decisive tests and plausible outcomes. Suggest a declarative figure title.
Scientific logic: [paste concise description]
Why use it: figures are powerful when they reduce cognitive load.
Prompt 9 - separate scientific risk from execution risk
Review the project description below and separate risks into:
- scientific uncertainty: uncertainty in the answer or mechanism;
- execution risk: risk that I cannot generate the required evidence.
For scientific uncertainty, explain what could still be learned under different outcomes. For execution risk, propose realistic mitigation options. Do not remove scientific risk merely to make the project look safer.
Why use it: high scientific risk and poor execution planning are not the same thing.
Prompt 10 - test objectives as scientific decisions
Here are my project objectives: [objectives].
For each objective, answer: what scientific uncertainty is resolved when this objective is completed? If an objective mainly describes an activity, propose a question that would make its scientific purpose explicit. Do not create new work packages.
Why use it: objectives are stronger when they resolve uncertainty rather than only list tasks.
Prompt 11 - challenge the independence case
Based only on the CV summary and proposal concept below, identify what evidence supports the claim that this is an independent scientific programme rather than a continuation of my previous supervisor or group. Identify where the independence case is weak or ambiguous. Do not infer achievements I have not provided.
CV summary: [text] Proposal concept: [text]
Why use it: independence needs to be visible in the scientific trajectory, not merely stated.
Prompt 12 - simulate a Step 1 reader
Act as an ERC panel member reading only Part I and the CV/Track Record. You are an expert scientist but not necessarily a specialist in my exact niche. Give me:
- the strongest reason to continue to Step 2;
- the three biggest scientific uncertainties that remain;
- the single sentence you think describes the proposal;
- any claim that feels important but insufficiently supported.
Do not evaluate detailed feasibility beyond what is needed to understand the scientific strategy.
Why use it: it tests what the first-stage argument actually communicates.
Prompt 13 - simulate a specialist Step 2 reviewer
Read the project as a specialist reviewer with access to Part I and Part II. Focus on whether the methodology can discriminate between the proposed explanations, whether dependencies are explicit, whether risks have meaningful responses, and whether the resources follow the scientific architecture. Identify the five questions you would ask before recommending funding.
Why use it: Step 2 requires deeper confidence in implementation.
Prompt 14 - prepare adversarial interview questions
Generate 20 concise scientific questions an ERC panel could ask about this proposal. Prioritise assumptions, alternative explanations, feasibility bottlenecks, why the PI is uniquely positioned to lead the programme, team design, budget logic and what happens if the central hypothesis fails. Avoid generic questions such as “why is this important?” unless you make them specific to my text.
Why use it: good interview preparation tests scientific ownership rather than memorised answers.
Prompt 15 - detect generic or AI-like proposal language
Mark any sentence in this draft that could plausibly appear in hundreds of unrelated grant proposals because it is generic, promotional or vague. For each marked sentence, explain what concrete scientific information is missing. Do not rewrite it into more marketing language.
Why use it: proposal language should carry scientific information, not filler.
Prompt 16 - test the field-level consequence
If the strongest version of this project succeeds, describe the plausible scientific consequences over three time horizons: immediate (during the grant), medium-term (5 years after), and field-shaping (10-15 years). Separate consequences that follow logically from the project from speculative claims that would require additional assumptions.
Why use it: this tests whether the larger vision is credible and explicit.
What not to ask AI to do
Avoid prompts such as:
- "write me a winning ERC proposal";
- "invent a breakthrough hypothesis";
- "generate citations that support my idea";
- "make this sound more groundbreaking" without scientific evidence;
- "predict my probability of winning".
These requests tend to create confident prose without improving the science.
A better workflow
Use AI after you have defined the scientific content yourself:
- write the problem in your own words;
- identify the unknowns;
- formulate the central question;
- map possible outcomes;
- use prompts to challenge ambiguity and assumptions;
- revise the science yourself;
- use AI again for a second critical pass;
- verify every external fact against the original source.
The ERC Blank-Page Builder provides a structured starting point before you use any of these prompts.
AI is not a replacement for peer review
An AI system does not know whether your result is truly surprising to your field, whether a proposed experiment is realistic in your laboratory, or how a specific panel will judge scientific novelty.
Use it as one source of challenge. Keep scientific peers, Host Institution support, NCP guidance and independent review in the process.
Who can help with an ERC proposal? ->
Official source
The ERC Scientific Council's current position states that external help, including AI, does not remove the applicant's responsibility for authorship, plagiarism and good scientific conduct: Current position of the ERC Scientific Council on AI.