Automated decisions often arrive with a human-shaped absence. No interviewer explains why your application disappeared. No loan officer says which factor mattered. No housing platform tells you whether a score, filter, or fraud tool blocked the message. The screen simply says no, or says nothing at all.
You may not be able to force a full technical explanation immediately. You can still ask better questions, preserve the right records, and look for routes to human review. This FAQ is general information in a Germany/EU context, not legal advice for your case.
What counts as an automated decision?
An automated decision is a decision made by software with little or no meaningful human involvement. It may rank, reject, approve, flag, suspend, price, prioritize, or route you. The system might be called an algorithm, model, score, matching tool, fraud detector, recommender, risk engine, or screening system.
Common areas include job recruitment, credit, insurance, housing platforms, welfare and benefits administration, education technology, gig-work platforms, social media enforcement, marketplace accounts, and identity verification. Some systems make the final decision. Others shape the decision so strongly that the human step becomes a rubber stamp.
Not every digital form is an automated decision. A human may still read the file. But if a score, filter, or automated rule effectively decides who gets seen, who gets rejected, or who must prove themselves again, it deserves scrutiny.
Was kann ich tun, wenn eine automatisierte Entscheidung mich benachteiligt?
Start by making the decision visible. Save the rejection message, portal screen, email, timestamp, application reference, account name, and any criteria shown before you applied. If the decision came after uploading documents, completing identity checks, answering screening questions, or taking an assessment, note that sequence.
Then ask the organization for:
- the main reason for the decision
- whether automated processing or profiling was used
- what data about you was used
- whether you can request human review
- how to correct wrong data
- how to appeal or submit additional context
Keep the message short. You do not need to accuse the organization of discrimination in the first email. A precise request often gets further than a broad demand for “the algorithm.”
Can I request human review?
Often, yes, especially where the decision has a serious effect on you. Under the GDPR’s rule on solely automated decisions with legal or similarly significant effects, people have protections in many situations, including the ability to seek human involvement, express their view, and challenge the decision. The exact route depends on the sector and facts.
Use plain wording:
I believe this decision may have been made or strongly shaped by automated processing. It significantly affects me. Please provide human review, explain the main reasons for the decision, and tell me what data was used.
Human review should mean more than a person clicking “confirm” after the system has already decided. A meaningful reviewer should have authority to change the outcome, see relevant information, consider your explanation, and check whether the system used wrong or inappropriate data.
What can I ask to see about my data?
You can usually ask an organization for access to personal data it holds about you. In GDPR language, this is the access right. In practical language, you are asking: what information did you use to judge me?
Ask for categories of data, source of data, recipients or partners, retention period, and meaningful information about the logic involved where automated decision-making is used. “Meaningful” does not always mean source code. It can mean the main factors, the kind of data used, and how those factors affected the result.
If you suspect wrong data, ask for correction. Examples include an old address, confused identity, incorrect employment status, outdated debt record, mistaken fraud flag, duplicate account, misspelled name, or document scan failure. A boring data error can produce a life-changing denial.
How can algorithmic discrimination be identified?
Algorithmic discrimination is often identified through patterns, comparisons, and suspicious explanations, not through a single line of code. Warning signs include:
- people with protected characteristics are rejected more often despite similar qualifications
- names, accents, addresses, gaps in employment, disability-related patterns, or migration histories seem to act as proxies
- the system penalizes people who do not fit standard digital traces, such as stable address history or conventional employment records
- identity verification fails more often for certain skin tones, documents, ages, names, or assistive technologies
- appeal outcomes change when a human finally reviews the same information
- the organization cannot explain the main reason for a serious decision
One person’s case may not prove discrimination. It can still be the starting point. If several applicants, tenants, workers, or users compare experiences, patterns become easier to see. Be careful with privacy when comparing. Do not pressure people to share documents publicly.
What if this happened in a job application?
Recruitment tools may screen CVs, rank candidates, score video interviews, parse employment gaps, test personality, or filter by availability. If you are rejected quickly or never reach a human stage, ask whether automated screening was used and which criteria mattered.
You can request access to data used about you and ask for human review where the decision had a significant effect. If discrimination is suspected, anti-discrimination advice or legal advice may be relevant. Digital Dignity Lab focuses on the automated-system questions; broader employment discrimination deadlines and claims need specialist advice because time limits can be short.
For organizations using recruitment tools, do not hide behind the vendor. If you use a tool to screen people, you remain responsible for whether the process is explainable, accessible, and discriminatory in effect.
What if this happened with credit, housing, or insurance?
Credit and insurance decisions may rely on scores, risk categories, address data, payment history, fraud signals, or third-party data. Housing platforms may rank messages, hide profiles, flag accounts, or allow landlords to use filters that reproduce discrimination.
Ask which data sources were used and how to correct errors. If a score or third-party database was involved, ask who provided it. If the organization refuses to explain anything, consider consumer advice, data protection advice, or legal advice. For housing discrimination beyond the algorithmic layer, a tenant association or anti-discrimination advice route may be more useful than a purely technical complaint.
Do not assume the system is neutral because it uses numbers. Scores can carry old inequality in a clean interface.
What if this happened with benefits or a public service?
Public bodies may use automated tools for fraud detection, risk scoring, document routing, appointment systems, or eligibility support. A formal benefits decision in Germany often arrives as a Bescheid, an official written decision. The appeal mechanics for a Bescheid are their own topic, and Economic Security & Social Protection Centre and Justice Access Centre cover those administrative routes in more detail.
For the automated part, ask whether software or profiling affected the decision, what data was used, and how to obtain human review. If you receive an official decision with a deadline, do not wait for the data-access answer before protecting the appeal deadline. Administrative clocks can run faster than transparency processes.
What if a platform suspended, downranked, or demonetized me?
Platforms use automated systems for spam detection, copyright enforcement, nudity detection, hate-speech classification, fraud prevention, recommender ranking, advertising eligibility, and account integrity. A platform decision can affect income, organizing, reputation, or access to community.
Use the platform appeal, but preserve the notice first. Ask which rule was applied, whether automation was used, and what content or behavior triggered the decision. If the decision relates to harassment or abuse reports, Digital Dignity Lab’s article on platform reporting systems explains why moderation notices are often too thin and how to frame an appeal.
If your account is essential for work or public participation, say that in the appeal. Platforms often treat enforcement as content housekeeping when it can be an economic or civic decision.
What if the organization says the algorithm is a trade secret?
Trade secrets do not erase all accountability. An organization may not have to hand over source code, but it can still explain the main factors, data categories, purpose, and review route. It can still correct wrong data. It can still test for discriminatory effects. It can still provide a human appeal.
If the answer is “we cannot tell you anything,” ask again in narrower terms: “Please explain the main reasons for the decision and the categories of personal data used.” If the organization still refuses, a data protection authority, consumer advice body, union, works council, anti-discrimination advice service, or lawyer may be relevant depending on the context.
How should I write the first request?
Use calm, specific language and include identifiers:
I am requesting information about the decision dated [date] regarding [application/account/reference]. Please tell me whether automated processing or profiling was used, the main reasons for the decision, the categories of personal data used, and how I can obtain human review or appeal. If any data is inaccurate, please tell me how to correct it.
Attach only what is necessary to identify the case. Keep a copy. If the matter is urgent, use the official appeal route at the same time rather than waiting for a privacy response.
What can organizations do before harm happens?
Organizations should keep an inventory of automated tools, know what data each tool uses, test outcomes across groups, provide accessible non-digital alternatives where needed, and write decision notices a person can understand. They should require vendors to support explanation, audit, deletion, correction, and human review. They should not buy a scoring system and then act surprised that they are responsible for its consequences.
For affected people, the next step may be one document saved, one request sent, or one adviser contacted. You do not have to prove the whole system is discriminatory before asking why it denied you. A system that affects your opportunities should be able to explain itself in human terms.