Most organisations preparing an Erasmus+ application now open ChatGPT, Claude or Gemini at some point. That makes sense, as AI tools are good at turning rough notes into clean paragraphs, summarising long documents and getting you past a blank page.

The problem is not that AI is used. The problem is what happens when the proposal is written by AI rather than thought through by the partnership. AI produces fluent text, but Erasmus+ evaluators do not score fluency. They score relevance, design, partnership, and impact against fixed award criteria, and they read dozens of proposals per round and notice when a text sounds right but says very little.

After 10 years of writing and reviewing Erasmus+ proposals, these are the patterns we see most often in applications drafted mainly with chatbots, and why they cost points.

What the Programme Guide 2026 says about AI

The Erasmus+ Programme Guide 2026 addresses AI directly. In Part C, under “Original content and authorship”, it says that applicants using generative AI tools “should be conscious of the potential of plagiarism” and must ensure the application’s “appropriateness and accuracy, its compliance with intellectual property regulations, as well as its original content.”

The same section explains that all proposals are automatically screened by the National Agencies and the Executive Agency for double funding and plagiarism risks. If an application is similar to an already funded or to-be-funded project, a similarity report is generated, and without a justification from the applicant, the proposal may be rejected. Plagiarism makes a proposal ineligible. So the Guide does not ban AI. It puts the responsibility for accuracy and originality fully on the applicant.

1. Generic text that fits any project

Ask a chatbot to describe “a youth project on digital inclusion” and you will get a polished answer. Ask it again in another browser and you will get almost the same answer. Thousands of applicants are doing exactly this.

Evaluators recognise the pattern: broad statements about “the rapidly changing digital landscape”, “fostering European values” and “empowering young people”, with no names, numbers or local detail. Text like this could describe any project, which means it describes none.

It also matters for the similarity screening described above. Text generated from similar prompts tends to look similar. That is a risk you do not want in a programme that actively checks for it.

What scores instead: concrete facts about your organisation, your target group and your local context that no other applicant could write.

2. A needs analysis that is invented, not found

For Cooperation Partnerships (KA220), the Relevance criterion explicitly asks whether “the proposal is based on a genuine and adequate needs analysis”. For Small-scale Partnerships (KA210), Quality of project design asks whether the objectives “address the needs and goals of the participating organisations and the needs of their target groups”.

An AI tool does not know your learners, your community or the gaps your staff see every day. If you do not give it that information, it fills the space with plausible general needs, sometimes supported by statistics that are outdated, from the wrong country, or simply made up.

Evaluators look for a line that runs from a real, specific need to the objectives, the activities and the results. A generic needs section breaks that line at the very first step.

What scores instead: needs that come from your own evidence: surveys, staff experience, participant feedback, local data, previous projects. AI can help you structure this evidence. It cannot produce it.

3. Sections that do not match each other

An Erasmus+ application is not a set of separate essays. Objectives, activities, work plan, partner roles, results, budget and impact all have to tell the same story.

When sections are generated one by one in a chat window, they drift. The objectives mention four target groups, the activities only reach two. A result promised in the impact section has no activity that produces it. A partner described as “lead on dissemination” has no task in the work plan.

The KA210 and KA220 award criteria check exactly this: a work plan that is “clear, complete and effective”, a project that is “cost-effective and allocates appropriate resources to each activity”, and an “allocation of tasks” that “demonstrates the commitment and active contribution of all participating organisations”. Inconsistencies are among the easiest things for an evaluator to spot and to justify a lower score with.

4. Rules and figures the AI got wrong

Chatbots answer with confidence even when they are wrong. They mix up Programme Guide editions, Key Actions and National Agency rules. We regularly see AI-drafted texts with the wrong number of required partners, outdated priorities, mixed-up deadlines, or budget logic from the old cost-based model.

A few examples of details that must be exactly right in the 2026 Guide:

  • KA210 projects choose between two lump sums, EUR 30 000 and EUR 60 000. To be funded, a proposal must score at least 60 points, and at least half the points in each award criterion.
  • KA220 projects choose between three lump sums, EUR 120 000, EUR 250 000 and EUR 400 000. The threshold is 70 points, again with at least half the points in each criterion.

That last rule matters more than most applicants realise. A proposal can score well overall and still be rejected because one section, often the one written fastest, falls below half of its points.

What to do: check every rule, amount and deadline against the current Programme Guide and your National Agency’s call, never against a chatbot answer. Our free Erasmus+ Action Explorer 2026 is a quick starting point for deadlines, grants and eligibility by action.

5. A budget and work plan that do not add up

Under the lump sum model, the activities you describe must justify the amount you request. AI tools often produce ambitious activity lists that would cost far more than the lump sum, or the opposite: a large lump sum with too little work behind it. Both are visible to an evaluator assessing cost-effectiveness.

The same applies to mobility budgets in KA1, where travel bands, subsistence and organisational support follow fixed unit costs. A chatbot will happily estimate these from memory. Our free Erasmus+ Travel Calculator and Gantt Chart Maker help you check the numbers and the timeline against each other before you write.

6. Nobody in the partnership owns the text

This is the risk that shows up after approval. If the proposal was generated rather than discussed, partners often do not recognise their own commitments. The coordinator has to implement activities nobody planned properly, and the final report has to explain why results differ from what was promised.

Evaluators can often feel this at the application stage too. Partner roles are vague, coordination mechanisms are generic, and there is no sign that the organisations actually talked to each other.

How to use AI well in your Erasmus+ proposal

None of this means you should stop using AI. Used carefully, it saves real time. A sensible division of work looks like this:

Good uses of AI:

  • Turning your own notes, meeting minutes and survey results into a first structure
  • Shortening and tightening text you wrote to fit character limits
  • Checking grammar and readability, especially when writing in a second language
  • Brainstorming dissemination channels or indicator ideas, which you then adapt and verify

Leave to people:

  • The needs analysis and the evidence behind it
  • Objectives, activities and results, and how they connect
  • Partner roles, budget choices and the work plan
  • Every rule, amount and deadline, checked against the Programme Guide
  • The final read-through, section by section, against the award criteria

And always keep the Guide’s warning in mind: whatever tool you use, the content must be accurate, original and yours.

What about getting outside help?

The same Part C section of the Programme Guide 2026 also advises applicants against relying on external entities or individuals to draft their application. The reasoning is that thinking through the project is part of building the organisation’s capacity, and that outside writers who work for several applicants increase the risk of similar texts. The Guide also states that the grant cannot pay for drafting the application, so any outside support is a cost your organisation covers itself.

We think this advice is right about the risks, and it shapes how we work:

  • The project stays yours. The idea, the needs, the target groups and the decisions come from your organisation and your partners. Our role is to ask the questions evaluators will ask, structure your answers and turn them into a proposal that holds together.
  • Every text is written for one project only. We do not reuse proposals, sections or templates between clients. Each application is built from your own context, which is also the best protection against similarity screening.
  • You stay involved section by section. You review and shape every part, so the partnership understands and owns what it commits to.
  • We check against the award criteria. Every section is read the way an evaluator reads it, against the criteria and minimum thresholds of your Key Action.

That is the difference between a proposal that reads well and one that scores well.

Need a second pair of eyes on your proposal?

If you are preparing a KA1, KA210, KA220, Capacity Building or European Youth Together application and want it developed with an evaluator’s perspective, see our Proposal Development service. You can also start with our free tools, such as the Erasmus+ Eligibility Checker or the Concept Note Maker, to put your idea on paper first.

When you are ready, tell us about your project: your organisation, the Key Action and your deadline.

Scroll to Top