AI Use Statement

This statement explains how IdeaScore AI, the artificial-intelligence features of IdeaScore, works with your data, what it can and cannot do, and the choices you have. It is published at /legal/ai and forms part of the Terms of Service and the Privacy Policy. Capitalised terms have the meanings given in the Terms of Service.

1. Our Commitments

1. Your data never trains a model. Applications, evaluations, results, chats, tuning, memory and settings are never used to train or improve any artificial-intelligence model, ours or our engine provider's or anyone else's.
2. People decide. IdeaScore AI suggests; the programme's people decide. Every decision about an Applicant (shortlisting, selection, rejection, ranking, funding) stays with the Customer's team.
3. You are in control. The Customer decides whether to use IdeaScore AI at all, for which programmes and applications, and with what instructions.
4. Only what is needed. IdeaScore AI receives only what a request needs. Before anything is sent to it, we swap out the names and contact details of the people in a Customer's programme: names are replaced with labels such as "Founder 1" or "Evaluator A", and email addresses, phone numbers, profile links and ID numbers are removed, before anything is sent, within the limits set out in section 3.
5. You can see the reasons. Each IdeaScore AI score comes with a rationale and the evidence it relied on, for a person to check.
6. Applicants are told. Where IdeaScore AI helps assess, score or rank applications, the Service tells the Applicants concerned that it was used (section 7).

2. What IdeaScore AI Does

IdeaScore AI can, when the Customer uses it:

• evaluate applications against the Customer's rubric, giving a suggested score, a rationale and evidence for each criterion (in Swift, Pro and Deep tiers, which differ in depth and cost);
• summarise applications and read pitch decks;
• answer questions in chat about a programme's applications, citing the applications it used, and, where the organisation has not switched Deep Research off (Deep Research is a mode in which IdeaScore AI also searches the public web) and a User chooses Deep Research for a question, search the web for the question and keep the pages the answer used with the chat;
• help set up programmes, forms and rubrics, and draft text such as emails and letters; and
• power the help assistant, which sees only the page you are on and simple counts, never application data.

3. What It Receives, and Where

For a run, IdeaScore AI sends the following to our AI engine provider: the answers to the application form, uploaded pitch decks, the rubric, and the Customer's tuning, memory and programme documents (for chat, also the question). Before anything is sent, the names of the people in the Customer's programme (founders, team members, Applicants, evaluators, mentors and staff) are replaced with labels such as "Founder 1" or "Evaluator A", and email addresses, phone numbers, profile links and identity numbers (such as Aadhaar and PAN) are removed. A pitch deck is sent with these details blacked out in its text, keeping its layout and pictures; when that cannot be done safely, its text is sent instead, with the same details removed. When the answer comes back, the Service puts the names back only for people allowed to see them; in a blind review they stay hidden. Team and company names are kept. These limits apply: only names that appear in the Customer's programme are recognised, and a change to them takes effect within a minute; words that are part of an image (for example a photo of a business card) or text drawn as outlines cannot be read; a pitch deck whose fonts cannot be read as letters is sent as text only, with the same details removed, while a font that maps letters to the wrong characters cannot be detected, so names drawn in it may remain visible in the deck sent; for the people of the application being read (or of every team in a question about several teams) a capitalised name is always replaced (except the few words described below), and a lower-case one of four letters or more unless it is also an everyday word; for anyone else, a name that is also an everyday word is replaced only where it reads as a name (not at the start of a sentence or a line), except that written in Devanagari such a name of anyone in the programme is also replaced on its own and with a common case ending, and names in lower case or short names in capitals only in requests not tied to particular teams; names written in unusual ways (one-letter names, two-letter names not written like "Li", misspellings, plurals, or another script such as Devanagari when the programme lists the name only in Latin letters) may not be recognised; a few words IdeaScore AI uses itself (the tier names "Deep", "Swift" and "Pro", and role words such as "Founder" or "Team") are never replaced on their own, while other everyday words that are also names (such as "Grant" or "Park") follow the rules above; in very large programmes, up to 20,000 people of each kind and 80,000 people in all are recognised; and with Deep Research, IdeaScore AI reads public web pages, which may name an Applicant's founders.

Our engine providers are listed on the Sub-processors page, with their processing regions: a main provider, and a second one that answers a request only when the main provider cannot. We use each under terms that do not allow it to use our customers' data to train or improve its models, and that limit its use of the data to providing the service to us. Those terms may allow a provider to keep inputs and outputs for a limited period, for example to detect and prevent misuse of its service; we do not promise that a provider keeps nothing. Our engine providers process data outside India, and no India-only processing is offered. The Service shows the Customer where IdeaScore AI processes applications, and an owner or administrator must accept processing outside India once, in IdeaScore AI's settings, before the first run.

Sensitive answers. Answers to questions marked as sensitive in the Service (for example a scholarship's family income, category or disability questions; see the Privacy Policy, section 4) are left out of everything IdeaScore AI receives, with one exception: when a person allowed to check eligibility asks IdeaScore AI to explain which of the programme's published eligibility rules an application meets, IdeaScore AI receives only the answers those rules check, without the Applicant's name or contact details. References written by referees reach IdeaScore AI only as the version with the referee's and the Applicant's names removed, when it summarises references for the reviewers.

An organisation may instead connect its own AI engine key. Its own agreement with that provider then governs what the provider does with the data.

4. What It Keeps, and For How Long

• Evaluation results: scores, rationales, evidence, the model version and the Credits charged. Kept with the programme, and removed when the application, evaluation or programme is deleted.
• Text extracted from decks and short digests of each application, kept so a deck is read only once and the right applications can be found for a chat question. Rebuilt when an application changes, removed with it.
• Chats: questions, answers and the applications and web pages they cite. Private to the User who started them. Deleted automatically 12 months after last use, and a User can delete their own chats at any time.
• Tuning, memory and programme documents, with who changed them and when, until an administrator changes or removes them.
• Feedback on answers, and a credit ledger entry for every run (an append-only billing record).

5. Who Sees It

IdeaScore AI results, tuning, memory and connectors are visible to the organisation's owners and the programme's administrators. Each person's chats are private to them.

Evaluators. Evaluators never see IdeaScore AI's scores or its evaluation of an application. Whether they see IdeaScore AI's other output about an application they review (its summaries of the application and the pitch deck, and its similarity and writing-pattern signals) depends on the evaluator visibility setting the organisation's owner chooses, which a programme's administrators may make stricter but not looser: hidden (the default), shown after the evaluator submits their own score, or always shown. In a blind round evaluators see none of it, whatever the setting.

Applicants do not see IdeaScore AI's scores, rationales or other output about their application. They are told when IdeaScore AI has been or may be used (section 7).

6. Limits You Should Know

IdeaScore AI uses large language models. Like all such models, it:

• can be wrong, misread a document, or state something false with confidence;
• can reflect biases in the material it was trained on, or in the instructions and rubric it is given;
• can give different answers to similar inputs, or when run again;
• checks claims only against public sources, and only with Deep Research. On the Deep tier (and on the Pro tier when Deep Research is chosen for a run), IdeaScore AI checks claims such as market size, competitors, patents and named partners against public web sources and says what it could not confirm. It cannot see private records, so it cannot confirm matters such as revenue, traction, funding or team credentials that public sources do not show; a claim it reports as supported may still be untrue, and one it could not confirm may still be true. Without Deep Research it checks nothing outside the application, and it never sees information that is not in the application or in public sources;
• may be less reliable on very short, very long, poorly scanned, non-English or highly technical applications; and
• may cite web pages that are out of date or inaccurate when Deep Research is used.

So: use IdeaScore AI as a second reader, not a judge. Have a person review every score and rationale before relying on it, look out for patterns that could disadvantage a group of Applicants, and do not use IdeaScore AI output as the sole basis for any decision that significantly affects a person. Section 7 of the Terms of Service makes this a contractual requirement.

7. Telling Applicants

7.1 What the Service tells Applicants. Where IdeaScore AI is used to help assess, score or rank applications, the Applicants concerned are told so in plain words, by the name IdeaScore AI:

• on the application form, before an Applicant submits, while the Customer has IdeaScore AI switched on: that the Customer may use IdeaScore AI to help its reviewers read and score applications, and that people make every decision;
• in the applicant portal: whether IdeaScore AI has helped assess that Applicant's application or, where it has not, that it may; and
• in decision letters sent through the Service, where IdeaScore AI scored that Applicant's application in the round the letter concerns.

Where a Customer has switched IdeaScore AI off, or its plan does not include it, and IdeaScore AI has not assessed an Applicant's application, the Service does not say that IdeaScore AI was used.

7.2 The Customer's own notice. These notices add to, and do not replace, the Customer's own privacy notice to Applicants, for which the Customer is responsible as Data Fiduciary (controller) under section 3.1 of the Privacy Policy. Customers should say in that notice whether IdeaScore AI is used to help review applications, and that people make the decisions. We recommend a sentence such as: "We may use IdeaScore AI to help our reviewers read and compare applications. It does not make decisions; our panel does. Your application is not used to train any AI model."

7.3 Decisions communicated outside the Service. A Customer that tells Applicants of decisions other than through the Service's decision letters is responsible for telling them, in that communication or its own notice, that IdeaScore AI was used.

8. Choices and Opt-outs

• Organisation level. An owner can switch IdeaScore AI off for the whole organisation in Settings, in which case no IdeaScore AI request is sent to the engine provider until an owner switches it on again. The Customer can also simply not run IdeaScore AI.
• Programme and application level. IdeaScore AI runs only on the programmes and applications a Customer's administrator chooses.
• Processing region. IdeaScore AI processes data outside India; no India-only processing is offered. IdeaScore AI does not run for an organisation until an owner or administrator has accepted this once.
• Deep Research. On by default for every organisation (since October 2026); an owner or administrator can switch it off for the whole organisation in IdeaScore AI's settings, and, once they have switched Deep Research off, no IdeaScore AI request searches the web, on any tier, until they switch Deep Research on again. While it is on, it is used in chat only when a User chooses Deep Research for a question; for evaluations it is always used on the Deep tier, offered as a choice on the Pro tier, and not offered on the Swift tier.
• Chats. Users can delete their own chat history at any time.
• API. Each API key has a daily IdeaScore AI credit cap, which can be set to 0 to stop IdeaScore AI use through that key.
• Applicants. An Applicant who does not want IdeaScore AI used on their application should tell the programme's organisers, who decide. An Applicant can also ask the organisation for a copy of their data, or for its deletion, from the applicant portal.
• Deletion. An organisation owner can ask us to export or delete everything IdeaScore AI holds for the organisation.

9. Law and Standards

We designed IdeaScore AI with the Digital Personal Data Protection Act, 2023, the GDPR (including Article 22 on automated decision-making), and India's AI governance guidance in mind, including the principles of human oversight, transparency and accountability. We will update this statement as laws and guidance on artificial intelligence develop.

10. Contact

Questions about IdeaScore AI and your data: privacy@ideascore.online. Concerns about an IdeaScore AI output: support@ideascore.online, or our Grievance Officer at grievance@ideascore.online.