ChatEIC: An AI Grant Writer for the EIC Accelerator

ChatEIC is a grant writer for the EIC Accelerator. It has been available for a while, but there has not been a new video about it in a long time, so this is an overview of how the current version works, from the empty project through to the finished draft.

Why a Professional Grant Writer Still Uses It

The most honest argument for the tool is that it is used in real client work. Even as a grant writer with access to a lot of different tools, ChatEIC still gets used. If a client asks for a quick draft for the EIC Accelerator, one gets created.

What it produces is built on deep research rather than on a blank-page generation. You get deep research on policies, deep research on the market and deep research on competitors, and the proposal is then crafted around that material.

How Good Is the Draft

It is a very good draft. Companies have literally submitted the ChatEIC version and passed step one with it.

That is not a recommendation to do so. The recommended approach is to create the draft and then actually edit it: check that everything in it is accurate, and add the things the AI missed. It is a simple tool at heart, and it works best as a starting point rather than as a submission.

Where it earns its place is the gap between doing nothing and hiring someone. It is a good fit if you want a quick draft but do not want to pay a consultant a few thousand to write one.

The Part That Does the Work: Input Validation

The reason the output holds up is the input side. The tool uses very detailed input validation to make sure you actually supply all the data it needs, rather than assuming you already know what the EIC Accelerator is looking for.

So the architecture has two halves. There is a very important input validation part that checks and makes sure you have the input you need. And then, on the outside, there is the draft creation itself, which uses very detailed instructions on how these proposals are supposed to look.

This matters because the official template does not tell you enough. If you read it, you are not actually aware of exactly what you need for the proposal, because it does not give you the granularity of the numbers, the statistics and the technical details you will actually be asked for. A great deal is left up to the imagination, and the form exists to close that gap.

Starting a Project: From Scratch or From a Prefill

Access is organised around credits, where one credit produces one proposal. If you only want a single proposal, one credit is enough, while multiple versions require multiple credits. Alongside that sits a training system that explains how to write particular sections, and the option to buy a direct review.

Creating a project offers two routes. You can start from scratch, which means filling in a lot of form fields by hand, or you can let AI prefill the form from a document you already have. The second route is the point of the tool: you upload one file, the text is extracted and passed to the language model, and the model fills in the form for you. The format does not matter, so a Word document, a PDF or a plain text file all work equally well.

Gathering the Data Without Handing Over Your Company

If you are not sure whether you already hold the data, the form provides a set of data gathering instructions written for language models, which you can copy into whichever agent you already use.

The instructions are worth reading on their own. They begin with the obvious material, such as founding date, company name, website and country, and then become considerably more specialised: customer groups with their pains and gains, named customer examples including website URLs, competitor groups, partners, investors, commercial partners, whether letters of intent exist, who the team members are and which of them are founders, revenues and financial data.

The reason this is offered as a copy-paste block rather than an upload is confidentiality. Pointing your own agent at your own knowledge base and asking it to collect exactly the items on that list is a better arrangement than uploading everything you own about your company to an external tool. You keep control of the source material, and you know precisely what was extracted.

A Worked Example: Lovable

The walkthrough uses Lovable, the vibe coding company, as its example, chosen because it had just received a large investment from the Scaleup Europe Fund. Lovable is far too advanced for the EIC Accelerator in reality, sitting at TRL 9, so it is not a good fit as an applicant. It is a good fit as a demonstration. It is also worth noting that the investment officially makes it deep tech, which is a useful reminder for software companies wondering whether they qualify: if Lovable is deep tech, that category is wider than most people assume.

The uploaded source file ran to roughly forty thousand characters, comfortably below the limit. Once the prefill starts it works in the background and sends an email when the project is ready, so there is no need to sit and wait for it.

The prefilled result is worth examining closely, because it shows where the automation is strong and where it is not. Customer groups came back populated with several named examples each, complete with website URLs. Competitors were largely correct, including the obvious ones. Partners were picked up accurately, including a recent AI infrastructure partnership and the payment and database providers. Investors were maxed out at ten and correctly included the asset manager behind the Scaleup Europe Fund, which does not have a website of its own. Team members were thinner, returning three people where the company employs closer to two hundred, although it did identify the founders correctly. Advisors were not filled at all.

That pattern is the useful lesson. The prefill gets you most of the way on the research-heavy fields and leaves the ones only you can answer. Financial figures, the target TRL, the size of the ask and the project year structure all remain yours to set. For the EIC Accelerator those years matter: 2027 is the first year of the project, 2028 the second, and 2029 the year of commercialisation.

Validation errors are shown but not enforced. A missing URL on one customer group, no commercial partners and too few advisors were all left unresolved in the demonstration, and the proposal still generated. The form is deliberately built so that you do not need to complete every field if you do not want to. It is a helper, not a gate.

What Comes Back

Generation consumes the credit and again runs in the background. In this case it took about fourteen minutes from starting generation to having the first draft in hand. Counting the input validation stage as well, the whole path from upload to draft is roughly forty minutes, and very little of that time requires your attention.

The draft arrives as a full proposal that can be read in the browser or downloaded as a Word file. It picks an acronym, and in this case chose Code Sovereignty, which is a genuinely clever fit given that the Scaleup Europe Fund investment implies a European sovereignty technology. Company names, customer names, competitor names, trademarks and certifications are all hyperlinked to the correct URLs automatically, which removes a tedious manual step.

The structure is complete rather than skeletal. Customer relationships, partners, customer benefits and properly explained customer groups are all present and already formatted, including a risk section. The draft is in that respect more complete than the official EIC template, which is very thin by comparison.

Two extras are worth calling out. The tool also writes a script for the three minute pitch video, structured into problem, solution and commercial sections, with parts allocated for three team members to read on camera. It is not meant to be voiced by AI, but it removes the blank page. It also produces the abstract, which is not part of the proposal itself but is required as part A of the submission, alongside technical details such as the company address. The proposal is part B.

Underneath all of it sits the reason the generation takes as long as it does: three separate passes of deep research covering policy and regulation, competitors, and the market. Each pass is retained in full, with citations hyperlinked in brackets throughout the important sections, and each can be copied out if you want to refine the material in your own system or substitute different sources.

The single most valuable thing it does is less visible than any of that. It works out a narrative. Narrative is what separates a funded proposal from a rejected one, and it is the first thing worth interrogating when a company arrives with a rejection in hand. The draft outlines a problem and then shapes the technology to fit it, so that the problem and the solution belong to each other rather than sitting side by side.

The Honest Limit

Realistically the tool saves around ninety percent of the time, and almost none of the remaining time is spent waiting on it. What it does not do is finish the job.

The important work starts once the draft exists. You have to edit it, verify that every claim is accurate, and bring it within the page limit. You still need your own images rather than AI generated ones, with the financial chart being the one exception that arrives ready to use. Creating a draft quickly is not a reason to skip editing it properly, and a fast first draft is only an advantage if the hours it frees are spent on accuracy.